CN117292278A - Digital orchard fruit tree positioning method, device, equipment and medium - Google Patents

Abstract

This application relates to a digital orchard fruit tree positioning method, device, equipment and medium. The method includes: updating the backbone network in the Mask R-CNN model to a Swin-Transformer network, and converting the Hybrid Task Cascade architecture into a The detector is embedded into the Mask R-CNN model to build a fruit tree image instance segmentation model; input the lychee fruit tree remote sensing image of the lychee fruit tree into the trained fruit tree image instance segmentation model for instance segmentation, and determine the fruit tree segmentation result, based on the fruit tree The recognition model identifies the lychee fruit tree remote sensing image of the fruit tree segmentation result, determines the number and pixel coordinates of the lychee fruit tree in the lychee fruit tree remote sensing image; determines the global positioning system coordinates of the fruit tree based on the digital terrain map; uses the matching algorithm to classify the lychee fruit tree in the litchi fruit tree remote sensing image. The quantity and pixel coordinates are paired with the global positioning system coordinates of the lychee trees in the orchard digital terrain map to determine the location of the lychee trees. This application can save costs and eliminates the need to install a positioning device.

Description

Digital orchard fruit tree positioning methods, devices, equipment and media

Technical field

The present application relates to the field of agricultural production, and in particular to a digital orchard fruit tree positioning method, corresponding devices, electronic equipment and computer-readable storage media.

Background technique

The litchi orchard is a complex ecosystem that is affected by many factors that affect tree growth, such as ecological environment, weather, terrain, soil, pests and diseases, and cultivation techniques. Traditional orchard water, fertilizer, and pesticide management relies entirely on experience, with human factors dominating the field, resulting in extensive management and low efficiency. Traditional orchards rarely use information tools for precise management, and do not collect production data to understand the necessary growth data of trees. In order to increase yields, traditional orchards often use improved varieties, increase soil fertility, and strengthen traditional pest and disease control, which consumes manpower and agricultural resources. However, there are still uncertainties in yield and lychee quality. The traditional orchard model can no longer meet the development requirements of modern orchards. Upgrading traditional orchards to smart orchards is the only way for orchard development. The fruit industry occupies an important position in my country's rural economic development, but the overall level of orchard production management still lags behind the international advanced level, especially in terms of digitalization, informatization, and intelligent management technology. Therefore, it is necessary to systematically summarize the progress made in digital orchard research, further clarify the future development direction, and provide technical support for the development of informatized and intelligent orchards.

In recent years, with the development of deep learning technology, drone remote sensing and computer vision technology have been widely used in agricultural production. Researchers use remote sensing images to segment fruit trees. For example, traditional machine learning methods, such as watershed algorithm and seed region growing algorithm, are used to segment fruit trees. For example, semantic segmentation methods such as UNet are used to separate fruit trees from the background. However, there is still relatively little research on digital orchard maps, and progress towards this goal is still slow, and there are also some problems that need to be further solved. For example, due to problems such as feature matching errors and improper processing of splicing edges, the trees in the spliced image are generally ambiguous, which has a great impact on the subsequent extraction of relevant information on the tree crown. Canopy information is obtained from the original image, but the original image does not contain the location information of the fruit trees, and the solution of installing positioning devices for each tree is very costly.

In summary, due to problems such as feature matching errors and improper processing of splicing edges, the trees in the spliced image are generally blurred, which has a great impact on the subsequent extraction of relevant information on the tree crown; obtained from the original image Canopy information, but the original image does not contain the location information of the fruit trees, and the solution of installing a positioning device for each tree has problems such as huge costs. The applicant has made corresponding explorations in order to solve this problem.

Contents of the invention

The purpose of this application is to solve the above problems and provide a digital orchard fruit tree positioning method, corresponding devices, electronic equipment and computer-readable storage media.

In order to meet the various purposes of this application, this application adopts the following technical solutions:

A digital orchard fruit tree positioning method proposed to meet one of the purposes of this application includes the following steps:

Respond to the orchard fruit tree positioning instructions and obtain remote sensing images of lychee fruit trees corresponding to each lychee fruit tree in the orchard;

Update the backbone network in the Mask R-CNN model to the Swin-Transformer network, and embed the Hybrid Task Cascade detector of the hybrid task cascade architecture into the Mask R-CNN model to build a fruit tree image instance segmentation model;

The lychee fruit tree remote sensing image corresponding to the lychee fruit tree is input into the trained fruit tree image instance segmentation model for instance segmentation, the fruit tree segmentation result is determined, and the lychee corresponding to the fruit tree segmentation result is identified based on the preset fruit tree recognition model. Remote sensing images of fruit trees, determining the number and pixel coordinates of lychee fruit trees in the remote sensing images of lychee fruit trees;

Construct a digital topographic map of the orchard, and determine the global positioning system coordinates corresponding to each fruit tree based on the digital topographic map;

Based on the preset matching algorithm, the number and pixel coordinates of the lychee fruit trees in the remote sensing image of the lychee fruit trees are matched with the global positioning system coordinates of each lychee fruit tree in the orchard digital topographic map, and the corresponding position of the lychee fruit tree is determined. Complete the positioning of lychee fruit trees in the digital orchard.

Optionally, the step of training the fruit tree image instance segmentation model includes the following steps:

Mark the lychee fruit tree part in the lychee fruit tree remote sensing image corresponding to the lychee fruit tree and its corresponding picture serial number to determine the training set and verification set of the fruit tree image instance segmentation model;

The fruit tree image instance segmentation model is subjected to back propagation iterative training based on the training set and the verification set, so that in each back propagation iterative training, the lychee fruit tree remote sensing image corresponding to the annotated lychee fruit tree is The error between the prediction box result and the actual annotation result obtained by inputting into the fruit tree image instance segmentation model is used to calculate the target loss function;

The model parameters are updated according to the target loss function until the change value of the target loss function is less than the preset threshold or the number of training times is greater than the preset number, then the model parameters are saved and the training of the fruit tree image instance segmentation model is completed.

Optionally, the step of inputting the lychee fruit tree remote sensing image corresponding to the lychee fruit tree to the trained fruit tree image instance segmentation model for instance segmentation and determining the fruit tree segmentation result includes the following steps:

The Swin-Transformer network in the fruit tree image instance segmentation model divides the input lychee fruit tree remote sensing image corresponding to the lychee fruit tree into several small blocks of equal size, and embeds each small block to determine each small block. Characteristic representation within;

Multi-layer translation windows are used to perform self-attention calculations on each small patch to exchange information between the features of adjacent small patches;

Feature extraction is performed based on multi-layer translation window self-attention and MLP layers, and at the same time, multi-scale feature fusion is performed on the output of each block;

Perform feature pyramid operation on features at multiple scales, and use global pooling and linear layers for classification output.

Optionally, the step of inputting the lychee fruit tree remote sensing image corresponding to the lychee fruit tree to the trained fruit tree image instance segmentation model for instance segmentation and determining the fruit tree segmentation result includes the following steps:

The Hybrid Task Cascade detector in the fruit tree image instance segmentation model uses the Swin-Transformer network as the backbone network to perform feature extraction on the input remote sensing image of the lychee fruit tree corresponding to the lychee fruit tree;

An anchor point frame is used to extract the candidate target area in the litchi fruit tree remote sensing image, and the candidate target area is passed through a multi-level detection and segmentation head to detect and segment the target area;

Perform bounding box regression and instance segmentation mask prediction for each target region;

Deduplication is performed on overlapping candidate target areas to determine the fruit tree segmentation results.

Optionally, the step of constructing a digital topographic map of the orchard and determining the global positioning system coordinates corresponding to each fruit tree based on the digital topographic map includes the following steps:

Construct a digital topographic map of the orchard, and determine the GPS information and elevation information corresponding to each fruit tree in the digital topographic map of the orchard;

The digital topographic map of the orchard is processed into blocks, and the fruit tree recognition model is used to identify the sub-blocks of the digital topographic map to determine the recognition results of the sub-blocks of the digital topographic map;

Perform data splicing on the digital terrain map sub-block recognition results to determine the corresponding pixel coordinates of each fruit tree in the orchard;

Convert the pixel coordinates of the fruit trees into the global positioning system coordinates corresponding to each fruit tree.

Optionally, based on a preset matching algorithm, the number and pixel coordinates of lychee fruit trees in the remote sensing image of lychee fruit trees are matched with the global positioning system coordinates of each lychee fruit tree in the orchard digital topographic map to determine the corresponding lychee fruit tree. The steps for the location include the following steps:

Calculate the pixel distance from the pixel coordinates of the litchi fruit tree in the litchi fruit tree remote sensing image to the midpoint of the litchi fruit tree remote sensing image, and sort them from near to far;

Convert the pixel distance into a global positioning system coordinate distance, and determine the global positioning system coordinates with the same number of fruit trees in the litchi fruit tree remote sensing image;

Based on the semi-vector function, vector angles are used to compare and filter the global positioning system coordinates to match the global positioning system coordinates of the lychee fruit tree, and determine the corresponding position of the lychee fruit tree in the digital orchard.

Optionally, based on a preset matching algorithm, the number and pixel coordinates of lychee fruit trees in the remote sensing image of lychee fruit trees are matched with the global positioning system coordinates of each lychee fruit tree in the orchard digital topographic map to determine the corresponding lychee fruit tree. The steps for the location include the following steps:

In response to the pesticide spraying instruction, the number and pixel coordinates of the lychee fruit trees in the remote sensing image of the lychee fruit trees are matched with the global positioning system coordinates of each lychee fruit tree in the orchard digital terrain map based on the semisine function to determine the corresponding lychee fruit tree. s position;

Pesticide spraying is performed on each lychee fruit tree in the lychee orchard according to the corresponding position of the lychee fruit tree to complete pest prevention of the lychee fruit trees.

A digital orchard fruit tree positioning device adapted to another purpose of this application includes:

The image acquisition module is configured to respond to the orchard fruit tree positioning instructions and acquire remote sensing images of lychee fruit trees corresponding to each lychee fruit tree in the orchard;

Segmentation model building module, set to update the backbone network in the Mask R-CNN model to a Swin-Transformer network, and embed the Hybrid Task Cascade detector of the hybrid task cascade architecture into the MaskR-CNN model to build a fruit tree Image instance segmentation model;

The pixel coordinate determination module is configured to input the litchi fruit tree remote sensing image corresponding to the litchi fruit tree into the trained fruit tree image instance segmentation model for instance segmentation, determine the fruit tree segmentation result, and identify the fruit tree based on the preset fruit tree recognition model. A remote sensing image of lychee fruit trees corresponding to the fruit tree segmentation result, and determining the number and pixel coordinates of lychee fruit trees in the remote sensing image of lychee fruit trees;

The global positioning coordinate determination module is configured to construct a digital terrain map of the orchard, and determine the global positioning system coordinates corresponding to each fruit tree based on the digital terrain map;

The fruit tree positioning module is configured to match the number and pixel coordinates of the lychee fruit trees in the remote sensing image of the lychee fruit trees with the global positioning system coordinates of each lychee fruit tree in the orchard digital terrain map based on a preset matching algorithm to determine the lychee fruit tree. Corresponding positions to complete the positioning of lychee fruit trees in the digital orchard.

An electronic device adapted to another object of the present application is provided, including a central processor and a memory. The central processor is used to call and run a computer program stored in the memory to execute the digital orchard fruit tree positioning method described in the present application. A step of.

A computer-readable storage medium is provided to meet another object of the present application, which stores a computer program implemented according to the digital orchard fruit tree positioning method in the form of computer-readable instructions. The computer program is called and run by the computer. , execute the steps included in the corresponding method.

Compared with the existing technology, this application aims at problems such as feature matching errors and improper processing of splicing edges. The trees in the spliced image are generally ambiguous, which has a great impact on the subsequent extraction of relevant information on the tree crown; from the original The canopy information is obtained from the image, but the original image does not contain the location information of the fruit trees, and the solution of installing a positioning device for each tree has problems such as huge costs. This application includes but is not limited to the following beneficial effects:

First, the matching algorithm proposed in this application performs matching by obtaining the GPS information of fruit trees in maps and drone images, eliminating the need to install a positioning device and saving costs;

Secondly, the Swin-CMN model proposed in this application has superiority in lychee canopy segmentation, and has better performance indicators and segmentation effects than other networks;

Third, the digital orchard fruit tree positioning method proposed in this application can map the high-resolution canopy in the drone image to the map, thereby providing high-resolution real canopy information of the lychee tree. It provides great help for subsequent processing. This method has high accuracy and repeatability, solves the problems of extensive and low efficiency of traditional orchard management, and provides effective technical support for the effective development of smart orchards.

Description of drawings

The above and/or additional aspects and advantages of the present application will become apparent and readily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

Figure 1 is a flow chart of the implementation of the digital orchard fruit tree positioning method in the embodiment of the present application;

Figure 2 is a schematic flow chart of the digital orchard fruit tree positioning method in the embodiment of the present application;

Figure 3 is a network structure diagram of the Swin-CMN model in the embodiment of this application;

Figure 4 is the recognition result of instance segmentation of litchi remote sensing images in the embodiment of the present application;

Figure 5 is the recognition result of the orchard map recognition in the embodiment of the present application;

Figure 6 is an implementation rendering of determining the location of a lychee fruit tree based on a preset matching algorithm in an embodiment of the present application;

Figure 7 is a visual rendering of the digital map system in the embodiment of the present application;

Figure 8 is a functional block diagram of the digital orchard fruit tree positioning device in the embodiment of the present application;

Figure 9 is a schematic structural diagram of a computer device in an embodiment of the present application.

Detailed ways

The embodiments of the present application are described in detail below. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements with the same or similar functions. The embodiments described below with reference to the drawings are exemplary and are only used to explain the present application and cannot be construed as limiting the present application.

Those skilled in the art will understand that, unless expressly stated otherwise, the singular forms "a", "an", "the" and "the" used herein may also include the plural form. It should be further understood that the word "comprising" used in the description of this application refers to the presence of stated features, integers, steps, operations, elements and/or components, but does not exclude the presence or addition of one or more other features, Integers, steps, operations, elements, components and/or groups thereof. It will be understood that when we refer to an element being "connected" or "coupled" to another element, it can be directly connected or coupled to the other element or intervening elements may also be present. Additionally, "connected" or "coupled" as used herein may include wireless connections or wireless couplings. As used herein, the term "and/or" includes all or any unit and all combinations of one or more of the associated listed items.

It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical terms and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. It should also be understood that terms, such as those defined in general dictionaries, are to be understood to have meanings consistent with their meaning in the context of the prior art, and are not to be used in an idealistic or overly descriptive manner unless specifically defined as here. to explain the formal meaning.

Those skilled in the art can understand that the "client", "terminal" and "terminal device" used here include both devices with wireless signal receivers, devices that only have wireless signal receivers without transmission capabilities, and A device with receiving and transmitting hardware capable of conducting two-way communications on a two-way communications link. Such devices may include: cellular or other communication devices such as personal computers and tablet computers, which have a single line display or a multi-line display or a cellular or other communication device without a multi-line display; PCS (Personal Communications Service, personal communication system) ), which can combine voice, data processing, fax and/or data communication capabilities; PDA (Personal Digital Assistant, personal digital assistant), which can include a radio frequency receiver, pager, Internet/Intranet access, web browser, notepad , calendar and/or GPS (Global Positioning System, Global Positioning System) receiver; conventional laptop and/or handheld computer or other device having and/or including a conventional laptop and/or handheld radio frequency receiver computer or other device. As used herein, "client", "terminal" and "terminal device" may be portable, transportable, installed in a vehicle (air, sea and/or land), or adapted and/or configured to be used locally Operate, and/or operate in distributed form, at any other location on Earth and/or space. The "client", "terminal" and "terminal device" used here can also be communication terminals, Internet access terminals, music/video playback terminals, for example, they can be PDA, MID (Mobile Internet Device, mobile Internet device) and/or A mobile phone with music/video playback function can also be a smart TV, set-top box and other devices.

The hardware referred to by names such as "server", "client", and "service node" in this application are essentially electronic devices with equivalent capabilities to personal computers, and are equipped with a central processing unit (including arithmetic units and controllers). ), memory, input devices, output devices and other necessary components disclosed by the von Neumann principle. The computer program is stored in its memory. The central processor transfers the program stored in the external memory into the memory to run, and executes the program. The instructions in it interact with input and output devices to complete specific functions.

It should be pointed out that the concept of "server" in this application can also be extended to the situation of server cluster. According to the network deployment principles understood by those skilled in the art, the servers should be logically divided. In physical space, these servers can be independent of each other but can be called through interfaces, or they can be integrated into a physical server. of a computer or a cluster of computers. Those skilled in the art should understand this modification, but it should not limit the implementation of the network deployment method of the present application.

Unless expressly specified, one or several technical features of this application can either be deployed on the server for implementation and the client remotely calls to obtain the online service interface provided by the server to implement access, or can be directly deployed and run on the client to implement access.

The neural network models cited or possibly cited in this application, unless expressly specified, can be deployed on a remote server and remotely called on the client, or can be deployed on a client with competent device capabilities for direct calling. In some embodiments, , when it runs on the client, its corresponding intelligence can be obtained through transfer learning, so as to reduce the requirements for client hardware running resources and avoid excessive occupation of client hardware running resources.

Various data involved in this application, unless expressly specified, can be stored remotely in a server or in a local terminal device, as long as they are suitable for being called by the technical solution of this application.

Those skilled in the art should know that although the various methods of the present application are described based on the same concept and are common to each other, these methods can all be executed independently unless otherwise specified. Similarly, the various embodiments disclosed in this application are all proposed based on the same inventive concept. Therefore, the same expressed concepts, as well as the concepts that are appropriately transformed for convenience only, although the conceptual expressions are different, should be regarded as equivalent. understand.

Unless the mutually exclusive relationship between the various embodiments to be disclosed in this application is explicitly stated, the relevant technical features involved in the various embodiments can be cross-combined to flexibly construct new embodiments, as long as such combination does not deviate from the present disclosure. The application must be creative and can meet the needs of the existing technology or solve some deficiencies in the existing technology. Those skilled in the art will be aware of modifications to this.

Based on the above exemplary scenarios, please refer to Figure 1 and Figure 2. In one embodiment, the digital orchard fruit tree positioning method of the present application includes the following steps:

Step S10: Respond to the orchard fruit tree positioning instruction and obtain remote sensing images of lychee fruit trees corresponding to each lychee fruit tree in the orchard;

The terminal device can respond to the orchard fruit tree positioning instructions and obtain remote sensing images of lychee fruit trees corresponding to each lychee fruit tree in the orchard; to obtain the remote sensing images of lychee fruit trees, a DJI Phantom 4RTK drone can be used to map lychees at heights of 20m, 30m, and 50m respectively. For shooting in the orchard, set the heading overlap to 80% and the side overlap to 60%. The lychee fruit tree varieties photographed include Nuomi Ci, Guiwei, Xianjinfeng, etc.

In some embodiments, the original pictures are screened to filter out invalid pictures such as blurry and overexposed pictures, the pictures taken by the drone are imported into the three-dimensional modeling software to generate orthophotos, and the orthophotos are combined according to 1024*1024 and Randomly crop the 512*512 size to obtain 800 pieces of data. Perform data enhancements such as sharpening and blurring on the original data to obtain 4,000 data, increase the amount of training data, improve the generalization ability of the model, add noise data, and improve the robustness of the model. You can use Labe lme for data annotation. , converted into a data set in COCO format.

Step S20: Update the backbone network in the Mask R-CNN model to the Swin-Transformer network, and embed the Hybrid Task Cascade detector of the hybrid task cascade architecture into the Mask R-CNN model to construct fruit tree image instance segmentation Model;

Refer to Figure 3, update the backbone network in the Mask R-CNN model to the Swin-Transformer network, and embed the Hybrid Task Cascade detector of the hybrid task cascade architecture into the Mask R-CNN model to construct the fruit tree image Instance segmentation model, perform instance segmentation on the litchi fruit tree remote sensing image, embed the Swin Transformer network as the backbone network, and divide the input image H×W×3 into non-overlapping patch sets through patch partition, where each patch size is 4× 4, then the feature dimension of each patch is 4×4×3=48, and the number of patch blocks is H/4×W/4. Stage1 first uses a linear embedding to change the patch feature dimension after division into C, and sends it to the Swin Transformer Block for W-MSA and SW-MSA processing respectively. The operation of stage2-stage4 is the same. First, through a patch merging, the input is merged according to 2x2 adjacent patches. In this way, the number of sub-patch blocks becomes H/8×W/8, and the feature dimension becomes 4C. Use linear Embedding compresses 4C into 2C, adjusts the channel dimension to twice the original through a fully connected layer, and then sends it to the Swin Transformer Block for W-MSA and SW-MSA processing.

The main idea of the Hybrid Task Cascade detector is to introduce a cascade structure based on Faster R-CNN and simultaneously utilize multi-task learning to improve detection performance. The output of RPN is spliced with the first layer feature map, and then the feature maps from the second to fourth layers are spliced with the first layer feature map in turn to obtain a cascaded feature pyramid. On the basis of fused features, bounding box generation and classification are performed for each pyramid layer using multi-task learning of two tasks, namely the bounding box generation task and the category classification task.

In some embodiments, the FBLoss loss function L _FB is proposed based on the edge features of the litchi tree. The calculation method is as follows

L _FB =λ _FL L _FL +λ _SM L _SM +λ _B L _B

Among them, L _FL is the focus loss function, and the calculation formula is

L _FL =-(1-p _t ) ^γ log(p _t )

(1-p _t ) ^γ is the adjustment factor, and γ≥0 is the adjustable focus parameter.

L _SM is the smoothing loss function, and the calculation formula is:

x is the numerical difference between the predicted box and the real box.

L _B is the boundary loss function, and the calculation formula is

L _B =∫ _Ω ΦG(q)s _θ (q)dq

ΦG is the level set representation of the boundary, s _θ represents the softmax probability output of the segmentation network, and D _G is a distance map relative to the boundary.

In some embodiments, the lychee canopy training image data set is divided into a training set, a test set and a verification set, hyper-parameters such as initial learning rate and training rounds are set, the Swin-CMN model is trained, and after the training is completed, the Inference weight files and segmentation model performance metrics for the Swin-CMN model.

Preferably, the data set is divided into a training set, a test set, and a verification set in a ratio of 8:1:1. Set the initial learning rate to 0.02, the gradient descent method to SGD, the learning rate strategy to warmup, epochs (training rounds) to 200, batch_size (batch size) to 8, train on RTX3090, and obtain the model weight file after training. .

Furthermore, the model performance index can be obtained through the following steps. The average average precision (mAP) is selected as an index to evaluate the accuracy of the training model. mAP (Mean Average Precision) is an evaluation index obtained by arithmetic averaging the average precision (Average Precision) of multiple object categories. It is used to evaluate the overall effect of the target detection algorithm. The commonly used mAP value is between 0-1. The higher the value, the better the effect of the target detection algorithm. In image segmentation, a curve can be drawn for each category based on precision P (Precision) and recall R (Recall), and the AP value is calculated using the area under the curve (AUC). Multiple indicators are calculated as follows:

TP represents the number of samples that are predicted to be positive by the model and are actually positive; FP represents the number of samples that are predicted to be positive by the model but are actually negative; FN represents the number of samples that are actually positive but are predicted to be negative by the model; P is the accuracy Rate refers to the proportion of samples that are actually positive among the samples predicted by the model to be positive; R is the recall rate, which refers to the proportion of samples that are predicted to be positive by the model among the samples that are actually positive. A high accuracy rate indicates that the true positive samples account for a high proportion of the predicted positive samples, and the model has strong recognition ability; a high recall rate indicates that the true positive samples account for a high proportion of samples identified as positive by the model, and the model coverage ability is strong.

Step S30: Input the remote sensing image of the lychee fruit tree corresponding to the lychee fruit tree into the trained fruit tree image instance segmentation model for instance segmentation, determine the fruit tree segmentation result, and identify the phase of the fruit tree segmentation result based on the preset fruit tree recognition model. Corresponding remote sensing image of lychee fruit trees, determine the number and pixel coordinates of lychee fruit trees in the remote sensing image of lychee fruit trees;

Before inputting the lychee fruit tree remote sensing image corresponding to the lychee fruit tree into the trained fruit tree image instance segmentation model for instance segmentation, the fruit tree image instance segmentation model is trained and the lychee fruit tree corresponding to the lychee fruit tree is marked. The part of the lychee fruit tree in the remote sensing image and its corresponding picture serial number are used to determine the training set and verification set of the fruit tree image instance segmentation model; the fruit tree image instance segmentation model is backpropagated according to the training set and the verification set. Iterative training, so that in each backpropagation iterative training, the predicted frame result obtained by inputting the remote sensing image of the lychee fruit tree corresponding to the annotated lychee fruit tree into the fruit tree image instance segmentation model is different from the actual annotation result. error, calculate the target loss function; update the model parameters according to the target loss function, until the change value of the target loss function is less than the preset threshold or the number of training times is greater than the preset number, save the model parameters and complete the fruit tree image instance segmentation Model training.

Please refer to Figure 4. After training the fruit tree image instance segmentation model, the lychee fruit tree remote sensing image corresponding to the lychee fruit tree is input to the trained fruit tree image instance segmentation model to perform instance segmentation and determine the fruit tree segmentation result. The lychee fruit tree remote sensing image corresponding to the fruit tree segmentation result is identified based on the preset fruit tree recognition model, and the number and pixel coordinates of the lychee fruit tree in the lychee fruit tree remote sensing image are determined.

Step S40: Construct a digital terrain map of the orchard, and determine the global positioning system coordinates corresponding to each fruit tree based on the digital terrain map;

Please refer to Figure 5 to construct a digital topographic map of the orchard, and determine the global positioning system coordinates (GPS) information and elevation information corresponding to each fruit tree in the digital topographic map of the orchard; process the digital topographic map of the orchard into blocks and use a fruit tree identification model Identify the digital terrain map sub-blocks to determine the digital terrain map sub-block identification results; perform data splicing on the digital terrain map sub-block identification results to determine the pixel coordinates corresponding to each fruit tree in the orchard; convert the fruit tree pixel coordinates into each The corresponding global positioning system coordinates of the fruit trees are stored, and the (GPS) global positioning system coordinate data of all lychee fruit trees in the orchard is stored.

Step S50: Match the number and pixel coordinates of lychee fruit trees in the remote sensing image of lychee fruit trees with the global positioning system coordinates of each lychee fruit tree in the orchard digital terrain map based on a preset matching algorithm to determine the corresponding location of the lychee fruit tree. location to complete the positioning of lychee fruit trees in the digital orchard.

Referring to Figure 6, calculate the pixel distance from the pixel coordinates of the litchi fruit tree in the litchi fruit tree remote sensing image to the midpoint of the litchi fruit tree remote sensing image, and sort them from near to far; convert the pixel distance into a global positioning system coordinate distance, and determine GPS coordinates with the same number of fruit trees in the remote sensing image of lychee fruit trees; based on the semi-vector function and vector angle, the GPS coordinates are compared and screened to match the global positioning system coordinates of the lychee fruit trees, and the digital position of the lychee fruit trees is determined. Corresponding location in the orchard.

Specifically, the distance values between the midpoint GPS of the litchi fruit tree remote sensing image and the GPS of all the orchard fruit trees can be calculated based on the semisine function (Haversine formula), sorted from small to large, and the pixel coordinates of the fruit trees in the remote sensing image to the midpoint coordinate pixels of the image can be calculated. The distance value is interchanged with the GPS coordinates, that is, the GPS information of the fruit tree in the litchi fruit tree remote sensing image is calculated.

According to the vector relationship between the fruit trees in the litchi fruit tree remote sensing image and the whole orchard fruit tree, the angle between the fruit trees in the two situations was calculated to improve the accuracy of fruit tree pairing. The specific steps of the matching algorithm are as follows:

For fruit tree recognition in litchi fruit tree remote sensing images, n fruit trees are identified. When n is less than or equal to 1, the GPS of the matching fruit trees in the entire orchard is directly used; when n is greater than 1, the k-th fruit tree is compared with the k+1-th fruit tree. The distance between the fruit tree and the midpoint of the image. When the distance is greater than 1 meter, the original pairing scheme (1<k<=n) is maintained.

When the distance is less than 1 meter, we construct a vector triangle using the first fruit tree, the image midpoint, and the k-th fruit tree, compare their pixel vector angles with the GPS vector angle, and calculate the difference p.

Construct a triangle using the first fruit tree, the image midpoint, and the k+1th fruit tree, compare the GPS vector angle with the k-th fruit tree pixel vector angle in step B2, and calculate the difference p1.

Calculated as follows:

Among them, a, b, and c are respectively the opposite sides of the pixel triangle surrounded by the midpoint of the image, the first fruit tree, and the k-th fruit tree. a1, b, and c are respectively the midpoint of the image, the first fruit tree, and the k-th fruit tree. The opposite sides of the pixel triangle surrounded by +1 fruit trees; x, y, and z are the opposite sides of the GPS triangle surrounded by the midpoint of the image, the first fruit tree, and the k-th fruit tree respectively. x1, y1, and z1 are respectively are the opposite sides of the GPS triangle surrounded by the midpoint of the image, the first fruit tree, and the k-th fruit tree. x, y, z, x1, y1, and z1 are calculated by the Haversine formula for The longitude and latitude coordinates of the first fruit tree and the k+1th fruit tree are calculated.

Compare the values of p and p1. If p<p1, then the k-th fruit tree corresponds to the k+1-th fruit tree GPS, and the k-th fruit tree and the k+1-th fruit tree are exchanged.

The calculated GPS is the GPS information of the fruit trees in the effective area of the image captured by the drone.

In some embodiments, please refer to Figure 7. A digital orchard system display interface can be built based on the Vue framework, the fruit tree information stored in the file is read, and the corresponding location information of the lychee fruit trees is visualized in the system.

It can be seen from the above embodiments that compared with the existing technology, this application aims at problems such as feature matching errors and improper processing of splicing edges. Trees in the spliced image are generally blurred, and it is difficult to extract relevant information on the subsequent tree crowns. It has a great impact; the canopy information is obtained from the original image, but the original image does not contain the location information of the fruit trees, and the solution of installing a positioning device for each tree has huge costs and other issues. This application includes but is not limited to the following beneficial effects:

First, the matching algorithm proposed in this application performs matching by obtaining the GPS information of fruit trees in maps and drone images, eliminating the need to install a positioning device and saving costs;

Secondly, the Swin-CMN model proposed in this application has superiority in lychee canopy segmentation, and has better performance indicators and segmentation effects than other networks;

Third, the digital orchard fruit tree positioning method proposed in this application can map the high-resolution canopy in the drone image to the map, thereby providing high-resolution real canopy information of the lychee tree. It provides great help for subsequent processing. This method has high accuracy and repeatability, solves the problems of extensive and low efficiency of traditional orchard management, and provides effective technical support for the effective development of smart orchards.

Based on any embodiment of the present application, the steps of training the fruit tree image instance segmentation model include the following steps:

Mark the lychee fruit tree part in the lychee fruit tree remote sensing image corresponding to the lychee fruit tree and its corresponding picture serial number to determine the training set and verification set of the fruit tree image instance segmentation model;

The fruit tree image instance segmentation model is subjected to back propagation iterative training based on the training set and the verification set, so that in each back propagation iterative training, the lychee fruit tree remote sensing image corresponding to the annotated lychee fruit tree is The error between the prediction box result and the actual annotation result obtained by inputting into the fruit tree image instance segmentation model is used to calculate the target loss function;

The model parameters are updated according to the target loss function until the change value of the target loss function is less than the preset threshold or the number of training times is greater than the preset number, then the model parameters are saved and the training of the fruit tree image instance segmentation model is completed.

On the basis of any embodiment of the present application, the step of inputting the lychee fruit tree remote sensing image corresponding to the lychee fruit tree into the trained fruit tree image instance segmentation model for instance segmentation and determining the fruit tree segmentation result includes the following steps:

The Swin-Transformer network in the fruit tree image instance segmentation model divides the input lychee fruit tree remote sensing image corresponding to the lychee fruit tree into several small blocks of equal size, and embeds each small block to determine the size of each small block. Feature representation within blocks;

Multi-layer translation windows are used to perform self-attention calculations on each small patch to exchange information between the features of adjacent small patches;

Feature extraction is performed based on multi-layer translation window self-attention and MLP layers, and at the same time, multi-scale feature fusion is performed on the output of each block;

Perform feature pyramid operation on features at multiple scales, and use global pooling and linear layers for classification output.

On the basis of any embodiment of the present application, the step of inputting the remote sensing image of the lychee fruit tree corresponding to the lychee fruit tree into the trained instance segmentation model of the fruit tree image for instance segmentation and determining the fruit tree segmentation result includes the following steps:

The Hybrid Task Cascade detector in the fruit tree image instance segmentation model uses the Swin-Transformer network as the backbone network to perform feature extraction on the input remote sensing image of the lychee fruit tree corresponding to the lychee fruit tree;

An anchor point frame is used to extract the candidate target area in the litchi fruit tree remote sensing image, and the candidate target area is passed through a multi-level detection and segmentation head to detect and segment the target area;

Perform bounding box regression and instance segmentation mask prediction for each target region;

Deduplication is performed on overlapping candidate target areas to determine the fruit tree segmentation results.

On the basis of any embodiment of the present application, the orchard digital terrain map is constructed, and the step of determining the global positioning system coordinates corresponding to each fruit tree based on the digital terrain map includes the following steps:

Construct a digital topographic map of the orchard, and determine the GPS information and elevation information corresponding to each fruit tree in the digital topographic map of the orchard;

The digital topographic map of the orchard is processed into blocks, and the fruit tree recognition model is used to identify the sub-blocks of the digital topographic map to determine the recognition results of the sub-blocks of the digital topographic map;

Perform data splicing on the digital terrain map sub-block recognition results to determine the corresponding pixel coordinates of each fruit tree in the orchard;

Convert the pixel coordinates of the fruit trees into the global positioning system coordinates corresponding to each fruit tree.

On the basis of any embodiment of the present application, based on a preset matching algorithm, the number and pixel coordinates of lychee fruit trees in the remote sensing image of lychee fruit trees are matched with the global positioning system coordinates of each lychee fruit tree in the orchard digital terrain map to determine The steps of positioning the lychee fruit tree correspondingly include the following steps:

Calculate the pixel distance from the pixel coordinates of the litchi fruit tree in the litchi fruit tree remote sensing image to the midpoint of the litchi fruit tree remote sensing image, and sort them from near to far;

Convert the pixel distance into a global positioning system coordinate distance, and determine the global positioning system coordinates with the same number of fruit trees in the litchi fruit tree remote sensing image;

Based on the semi-vector function, vector angles are used to compare and filter the global positioning system coordinates to match the global positioning system coordinates of the lychee fruit tree, and determine the corresponding position of the lychee fruit tree in the digital orchard.

Specifically, the distance values between the midpoint GPS of the litchi fruit tree remote sensing image and the GPS of all the orchard fruit trees can be calculated based on the semisine function (Haversine formula), sorted from small to large, and the pixel coordinates of the fruit trees in the remote sensing image to the midpoint coordinate pixels of the image can be calculated. The distance value is interchanged with the GPS coordinates, that is, the GPS information of the fruit tree in the litchi fruit tree remote sensing image is calculated.

According to the vector relationship between the fruit trees in the litchi fruit tree remote sensing image and the whole orchard fruit tree, the angle between the fruit trees in the two situations was calculated to improve the accuracy of fruit tree pairing. The algorithm idea is as follows:

For fruit tree recognition in litchi fruit tree remote sensing images, n fruit trees are identified. When n is less than or equal to 1, the GPS of the matching fruit trees in the entire orchard is directly used; when n is greater than 1, the k-th fruit tree is compared with the k+1-th fruit tree. The distance between the fruit tree and the midpoint of the image. When the distance is greater than 1 meter, the original pairing scheme (1<k<=n) is maintained.

When the distance is less than 1 meter, we construct a vector triangle using the first fruit tree, the image midpoint, and the k-th fruit tree, compare their pixel vector angles with the GPS vector angle, and calculate the difference p.

Construct a triangle using the first fruit tree, the image midpoint, and the k+1th fruit tree, compare the GPS vector angle with the k-th fruit tree pixel vector angle in step B2, and calculate the difference p1.

Calculated as follows:

Among them, a, b, and c are respectively the opposite sides of the pixel triangle surrounded by the midpoint of the image, the first fruit tree, and the k-th fruit tree. a1, b, and c are respectively the midpoint of the image, the first fruit tree, and the k-th fruit tree. The opposite sides of the pixel triangle surrounded by +1 fruit trees; x, y, and z are the opposite sides of the GPS triangle surrounded by the midpoint of the image, the first fruit tree, and the k-th fruit tree respectively. x1, y1, and z1 are respectively are the opposite sides of the GPS triangle surrounded by the midpoint of the image, the first fruit tree, and the k-th fruit tree. x, y, z, x1, y1, and z1 are calculated by the Haversine formula for The longitude and latitude coordinates of the first fruit tree and the k+1th fruit tree are calculated.

Compare the values of p and p1. If p<p1, then the k-th fruit tree corresponds to the k+1-th fruit tree GPS, and the k-th fruit tree and the k+1-th fruit tree are exchanged.

The calculated GPS is the GPS information of the fruit trees in the effective area of the image captured by the drone.

On the basis of any embodiment of the present application, based on a preset matching algorithm, the number and pixel coordinates of lychee fruit trees in the remote sensing image of lychee fruit trees are matched with the global positioning system coordinates of each lychee fruit tree in the orchard digital terrain map to determine The steps of positioning the lychee fruit tree correspondingly include the following steps:

In response to the pesticide spraying instruction, the number and pixel coordinates of the lychee fruit trees in the remote sensing image of the lychee fruit trees are matched with the global positioning system coordinates of each lychee fruit tree in the orchard digital terrain map based on the semisine function to determine the corresponding lychee fruit tree. s position;

Pesticide spraying is performed on each lychee fruit tree in the lychee orchard according to the corresponding position of the lychee fruit tree to complete pest prevention of the lychee fruit trees.

Referring to Figure 8, a digital orchard fruit tree positioning device adapted to one of the purposes of this application is provided, including an image acquisition module 1100, a segmentation model building module 1200, a pixel coordinate determination module 1300, a global positioning coordinate determination module 1400 and a fruit tree positioning module. Module 1500. Among them, the image acquisition module 1100 is configured to respond to the orchard fruit tree positioning instructions and acquire remote sensing images of lychee fruit trees corresponding to each lychee fruit tree in the orchard; the segmentation model building module 1200 is configured to update the backbone network in the Mask R-CNN model to Swin -Transformer network, and embed the Hybrid Task Cascade detector of the hybrid task cascade architecture into the Mask R-CNN model to construct a fruit tree image instance segmentation model; the pixel coordinate determination module 1300 is configured to compare the lychee fruit tree with the The corresponding lychee fruit tree remote sensing image is input to the trained fruit tree image instance segmentation model for instance segmentation, and the fruit tree segmentation result is determined. Based on the preset fruit tree recognition model, the lychee fruit tree remote sensing image corresponding to the fruit tree segmentation result is identified, and the result is determined. The number and pixel coordinates of lychee fruit trees in the remote sensing image of lychee fruit trees; the global positioning coordinate determination module 1400 is configured to construct a digital terrain map of the orchard, and determine the global positioning system coordinates corresponding to each fruit tree based on the digital terrain map; the fruit tree positioning module 1500 , set to match the number and pixel coordinates of lychee fruit trees in the remote sensing image of lychee fruit trees with the global positioning system coordinates of each lychee fruit tree in the orchard digital topographic map based on a preset matching algorithm, and determine the corresponding location of the lychee fruit tree. location to complete the positioning of lychee fruit trees in the digital orchard.

Based on any embodiment of the present application, please refer to Figure 9. Another embodiment of the present application further provides an electronic device. The electronic device can be implemented by a computer device. As shown in Figure 9, a schematic diagram of the internal structure of the computer device . The computer device includes a processor, a computer-readable storage medium, a memory, and a network interface connected by a system bus. Among them, the computer readable storage medium of the computer device stores an operating system, a database and computer readable instructions. The database can store a sequence of control information. When the computer readable instructions are executed by the processor, the processor can implement a Digital orchard fruit tree positioning method. The processor of the computer device is used to provide computing and control capabilities and support the operation of the entire computer device. Computer readable instructions may be stored in the memory of the computer device. When executed by the processor, the computer readable instructions may cause the processor to execute the digital orchard fruit tree positioning method of the present application. The network interface of the computer device is used for communication with the terminal connection. Those skilled in the art can understand that the structure shown in Figure 9 is only a block diagram of a partial structure related to the solution of the present application, and does not constitute a limitation on the computer equipment to which the solution of the present application is applied. Specific computer equipment can May include more or fewer parts than shown, or combine certain parts, or have a different arrangement of parts.

In this embodiment, the processor is used to execute the specific functions of each module and its sub-modules in Figure 8, and the memory stores program codes and various types of data required to execute the above modules or sub-modules. Network interfaces are used for data transmission to user terminals or between servers. The memory in this embodiment stores the program codes and data required to execute all modules/sub-modules in the digital orchard fruit tree positioning device of the present application, and the server can call the server's program codes and data to execute the functions of all sub-modules.

This application also provides a storage medium storing computer-readable instructions. When the computer-readable instructions are executed by one or more processors, the one or more processors execute the digital orchard fruit tree positioning described in any embodiment of this application. Method steps.

This application also provides a computer program product, including a computer program/instruction, which when executed by one or more processors, implements the steps of the digital orchard fruit tree positioning method described in any embodiment of this application.

Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments of the present application can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the program is executed, it may include the processes of the above method embodiments. Among them, the aforementioned storage medium can be a computer-readable storage medium such as a magnetic disk, an optical disk, a read-only memory (Read-Only Memory, ROM), or a random access memory (Random Access Memory, RAM).

The above are only some of the embodiments of the present application. It should be pointed out that those of ordinary skill in the technical field can also make several improvements and modifications without departing from the principles of the present application. These improvements and modifications can also be made. should be regarded as the scope of protection of this application.

In summary, this method can map the high-resolution canopy in the drone image to the map, thereby providing high-resolution real canopy information of the lychee tree and providing great convenience for subsequent processing. help. This method has high accuracy and repeatability, solves the problems of extensive and low efficiency of traditional orchard management, and provides effective technical support for the effective development of smart orchards.

Claims (10)

1. The digital orchard fruit tree positioning method is characterized by comprising the following steps of:

responding to the positioning instruction of the fruit tree of the orchard, and acquiring a litchi fruit tree remote sensing image corresponding to each litchi fruit tree in the orchard;

updating a backbone network in a Mask R-CNN model into a Swin-Transformer network, and embedding a Hybrid Task Cascade detector of a mixed task cascade architecture into the Mask R-CNN model to construct a fruit tree image instance segmentation model;

Inputting litchi fruit tree remote sensing images corresponding to the litchi fruit trees into a trained fruit tree image instance segmentation model for instance segmentation, determining fruit tree segmentation results, identifying litchi fruit tree remote sensing images corresponding to the fruit tree segmentation results based on a preset fruit tree identification model, and determining the number and pixel coordinates of litchi fruit trees in the litchi fruit tree remote sensing images;

constructing a digital topographic map of an orchard, and determining global positioning system coordinates corresponding to each fruit tree based on the digital topographic map;

and matching the number and pixel coordinates of the litchi trees in the litchi tree remote sensing image with the global positioning system coordinates of each litchi tree in the orchard digital topographic map based on a preset matching algorithm, and determining the corresponding positions of the litchi trees so as to finish the positioning of the litchi trees in the digital orchard.

2. The method of claim 1, wherein the step of training the fruit tree image instance segmentation model comprises the steps of:

labeling a litchi fruit tree part and a corresponding picture sequence number in a litchi fruit tree remote sensing image corresponding to the litchi fruit tree to determine a training set and a verification set of the fruit tree image example segmentation model;

Performing back propagation iterative training on the fruit tree image example segmentation model according to the training set and the verification set, so that in each back propagation iterative training, a target loss function is calculated according to errors of a predicted frame result and an actual labeling result, which are obtained by inputting the litchi fruit tree remote sensing image corresponding to the litchi fruit tree after labeling into the fruit tree image example segmentation model;

and updating model parameters according to the target loss function until the change value of the target loss function is smaller than a preset threshold value or the training times are larger than preset times, storing the model parameters and finishing training of the fruit tree image instance segmentation model.

3. The method for locating fruit tree in digital orchard according to claim 1, wherein the step of inputting the litchi fruit tree remote sensing image corresponding to the litchi fruit tree to the trained fruit tree image instance segmentation model for instance segmentation and determining fruit tree segmentation results comprises the following steps:

dividing an input litchi tree remote sensing image corresponding to the litchi tree into a plurality of small blocks with the same size by a Swin-transducer network in the fruit tree image example segmentation model, and embedding each small block to determine characteristic representation in each small block;

Performing self-attention calculation on each small block by adopting a multi-layer translation window so as to perform information interaction on the characteristics of the adjacent small blocks;

feature extraction is performed based on the multilayer translation window self-attention and the MLP layer, and multi-scale feature fusion is performed on the output of each block;

and carrying out feature pyramid operation on the features of a plurality of scales, and carrying out classified output by adopting a global pooling layer and a linear layer.

4. The method for locating a fruit tree in a digital orchard according to claim 3, wherein the step of inputting the litchi fruit tree remote sensing image corresponding to the litchi fruit tree to the trained fruit tree image instance segmentation model to carry out instance segmentation and determining the fruit tree segmentation result comprises the following steps:

the Hybrid Task Cascade detector in the fruit tree image example segmentation model adopts a Swin-Transformer network as a main network, and performs feature extraction on an input litchi fruit tree remote sensing image corresponding to the litchi fruit tree;

extracting candidate target areas in the litchi tree remote sensing image by adopting an anchor point frame, and carrying out multi-stage detection and segmentation on the candidate target areas so as to detect and segment the target areas;

Carrying out regression of a boundary box and prediction of an instance segmentation mask on each target area;

and performing de-duplication on the overlapped candidate target areas to determine the fruit tree segmentation result.

5. The method of claim 4, wherein the step of constructing a digital topography of the orchard, and determining global positioning system coordinates corresponding to each fruit tree based on the digital topography, comprises the steps of:

constructing an orchard digital topography map, and determining GPS information and elevation information corresponding to each fruit tree in the orchard digital topography map;

dividing the digital topographic map of the orchard into blocks, and identifying the digital topographic map sub-blocks by adopting a fruit tree identification model to determine digital topographic map sub-block identification results;

performing data splicing on the digital topographic map sub-block identification result, and determining pixel coordinates corresponding to each fruit tree in an orchard;

and converting the pixel coordinates of the fruit trees into global positioning system coordinates corresponding to each fruit tree.

6. The method for locating a fruit tree in a digital orchard according to claim 1, wherein the step of matching the number and pixel coordinates of litchi fruit trees in the litchi fruit tree remote sensing image with global positioning system coordinates of each litchi fruit tree in the digital topographic map of the orchard based on a preset matching algorithm to determine the corresponding positions of the litchi fruit trees comprises the following steps:

Calculating the pixel distance from the pixel coordinates of the litchi fruit tree in the litchi fruit tree remote sensing image to the midpoint of the litchi fruit tree remote sensing image, and sequencing from near to far;

converting the pixel distance into a global positioning system coordinate distance, and determining global positioning system coordinates which are the same as the number of the fruit trees in the litchi fruit tree remote sensing image;

and comparing and screening the global positioning system coordinates by adopting vector angles based on a semi-orthometric function to match the global positioning system coordinates of the litchi fruit trees, and determining the corresponding positions of the litchi fruit trees in the digital orchard.

7. The method according to any one of claims 1 to 6, wherein the step of determining the corresponding position of the litchi tree by pairing the number and pixel coordinates of the litchi tree in the litchi tree remote sensing image with the global positioning system coordinates of each litchi tree in the orchard digital topography map based on a preset matching algorithm comprises the steps of:

responding to pesticide spraying instructions, and matching the number and pixel coordinates of litchi trees in the litchi tree remote sensing image with the global positioning system coordinates of each litchi tree in an orchard digital topographic map based on a semi-normal function to determine the corresponding positions of the litchi trees;

And spraying pesticides on each litchi fruit tree in the litchi fruit garden according to the corresponding position of the litchi fruit tree so as to finish pest prevention of the litchi fruit tree.

8. A digital orchard fruit tree positioning device, comprising:

the image acquisition module is used for responding to the positioning instruction of the fruit trees in the orchard and acquiring litchi remote sensing images corresponding to the litchi fruit trees in the orchard;

the segmentation model construction module is used for updating a backbone network in the Mask R-CNN model into a Swin-transform network, and embedding a Hybrid Task Cascade detector of the mixed task cascade architecture into the Mask R-CNN model so as to construct a fruit tree image example segmentation model;

the pixel coordinate determining module is used for inputting litchi fruit tree remote sensing images corresponding to the litchi fruit trees into the trained fruit tree image instance segmentation model to conduct instance segmentation, determining fruit tree segmentation results, identifying the litchi fruit tree remote sensing images corresponding to the fruit tree segmentation results based on a preset fruit tree identification model, and determining the number and pixel coordinates of the litchi fruit trees in the litchi fruit tree remote sensing images;

the global positioning coordinate determining module is used for constructing a digital topographic map of the orchard and determining global positioning system coordinates corresponding to each fruit tree based on the digital topographic map;

The fruit tree positioning module is used for matching the number and pixel coordinates of litchi fruit trees in the litchi fruit tree remote sensing image with the global positioning system coordinates of each litchi fruit tree in the orchard digital topographic map based on a preset matching algorithm, and determining the corresponding positions of the litchi fruit trees so as to finish the positioning of the litchi fruit trees in the digital orchard.

9. An electronic device comprising a central processor and a memory, characterized in that the central processor is arranged to invoke a computer program stored in the memory for performing the steps of the method according to any of claims 1 to 7.

10. A computer-readable storage medium, characterized in that it stores in the form of computer-readable instructions a computer program implemented according to the method of any one of claims 1 to 7, which, when invoked by a computer, performs the steps comprised by the corresponding method.