CN104751455A - Crop image dense matching method and system - Google Patents

Abstract

The invention discloses a crop image dense matching method, comprising the following steps: calculating a matching point in the right image for a pixel in the center area of the target to be matched in the left image; on the basis of obtaining the center matching point of the object, The center point of the target and its matching point are the sub-images of the two images intercepted as the center; in the two sub-images, the image outside the boundary of the analysis target is removed by the boundary segmentation algorithm; for each pixel of the analysis target, search in the right sub-image The pixel point closest to its STFT vector is used as the matching point, so as to obtain each pixel matching point, and obtain the dense matching of the target crop; on the basis of the dense matching result of the sub-image, the source image is obtained through the inverse transformation of the sub-image Dense matching results. The matching method and system of the present invention can achieve more accurate dense matching results for captured crop images, can meet the purpose of crop state analysis, and provide a basis for further real-time analysis and processing.

Description

A crop image dense matching method and system

technical field

The invention relates to the field of agricultural informatics, in particular to a crop image dense matching method and system.

Background technique

Image matching is an important part of the field of pattern recognition and computer vision. Accurate image matching is the basis for subsequent processing such as image target positioning, motion detection, 3D reconstruction, and information fusion. The quality of the processing results will directly affect the quality of subsequent analysis work.

In terms of crop image analysis, in order to be able to analyze crop status information, such as the shape of the affected area of sick leaves, leaf size, fruit size, plant height, etc., it is necessary to perform dense matching calculation on the crop image, that is, to obtain each pixel of the target image in Location correspondences in multiple images. However, although in the industrial field, dense matching research has achieved certain research results, but in terms of agricultural crop images, due to the similarity of crop images and background color, brightness and texture, dense matching of crop images is a complex task. Work. Less progress has been made on dense matching for crop images.

At present, the research on dense matching mainly includes methods based on optical flow field and matching methods based on SIFT flow.

The so-called optical flow field is to assign a velocity vector to each pixel in the image when there is relative motion between the camera and the scene object, and obtain a motion field about the image called the optical flow field. In terms of optical flow field research, so far, there are about dozens of various methods and improved methods, which are roughly divided into four categories conceptually: gradient-based methods, region-based methods, energy-based methods, and phase-based methods. . Although the dense matching algorithm based on the optical flow field has achieved great results in industrial applications, the dense matching of crop images collected in the natural environment is often more complicated than the industrial dense matching, mainly due to the existence of different leaves, different fruits and backgrounds. Due to the similar color and brightness of crop images, occlusion differences at different shooting angles, and differences in the reflected light of crops at different angles, the brightness consistency based on the optical flow field is difficult to apply to crop images, which seriously affects the accuracy of optical flow calculations.

The SIFT matching algorithm is a scale-space-based matching algorithm proposed by David in 2004. SIFT has stable matching capabilities under image scaling, rotation, and brightness changes. It is recognized as one of the two most successful matching algorithms in the past 10 years. The other matching algorithm is the surf matching algorithm proposed by Bay in 2008. However, both the SIFT matching algorithm and the surf matching algorithm only find the most matching point pairs in the two images, and are not dense matching algorithms. With the help of the idea of optical flow field, CeLiu et al. proposed a SIFT flow dense matching method based on the consistency of SIFT eigenvectors. The basis of SIFT flow calculation is the gray gradient consistency of structural features. In some industrial applications with simple structures, SIFT streams can achieve high-precision matching effects. However, in the field of agriculture, the structure of crop images in the natural environment is very complex, and the parts of the crops and the background have high similarity, which affects the SIFT matching effect of crop images. Due to the rotation invariance of SIFT calculation, the edge region of the crop image may also have a large similarity with the non-matching point structure. The inventor used several typical optical flow field algorithms and SIFT flow algorithms to calculate the dense matching of sweet potato leaf, strawberry and eggplant leaf images. The experimental results show that there is a large gap, which is difficult to meet the needs of crop analysis.

At present, the research on dense matching of complex crop images in natural environment is almost blank, which is mainly due to the complex structure and messy distribution of crop images in natural environment. At present, some important achievements are mostly in the field of industrial application, which is difficult to apply in the field of agriculture, and cannot achieve the purpose of crop state analysis.

Contents of the invention

The purpose of the present invention is to overcome the shortcomings and deficiencies of the prior art, and provide a crop image dense matching method.

Another object of the present invention is to provide a crop image dense matching system.

The purpose of the present invention is achieved through the following technical solutions:

A crop image dense matching method, including steps in the following order:

S1. Analyze the center matching of the target: For a pixel in the center area of the target to be matched in the left picture, calculate its matching point in the right picture through the SIFT flow field;

S2. Analyzing target subimage segmentation: on the basis of obtaining the target center matching point, intercepting the subimages of two images centered on the target center point and its matching point;

S3. Subgraph boundary segmentation and subgraph dense matching: In the two subgraphs, the image outside the boundary of the analysis target is removed through the boundary segmentation algorithm; for each pixel of the analysis target, search for the shortest distance to its SIFT vector in the right subgraph The pixel points are used as matching points, so as to obtain each pixel matching point and obtain the dense matching of the target crop;

S4. Dense matching in the original image: on the basis of the dense matching result of the sub-image, through the inverse transformation of the sub-image intercepted in step S2, the dense matching result of the source image is obtained.

In step S2, the size of the target sub-image segmentation is set according to the size of the monitoring target. In order to ensure that the target to be analyzed can completely appear in the two sub-pictures.

In step S3, the boundary segmentation algorithm is a canny boundary segmentation algorithm.

Another object of the present invention is achieved through the following technical solutions:

A crop image dense matching system, including a thin client, a camera connected to the thin client, a first AP, a power supply module for the thin client and the first AP, a server, and a second AP connected to the server .

The power supply module includes a controller, a storage battery, a solar panel, and an inverter respectively connected to the controller, and the inverter is respectively connected to the thin client and the first AP. The thin client, the camera, and the first AP of the crop image dense matching system of the present invention are located next to the crops to be monitored, usually in the wild, so it is more convenient to use solar energy for the power supply module, and it is not necessary to lay special power supply cables, which is also more energy-saving and environmentally friendly.

The cameras mentioned are two Logitch Pro9000 cameras, the distance between the two cameras is 6-12 cm from the ground, and the directions of the two cameras are the same. Many experiments have proved that the Logitch Pro9000 camera can capture higher quality crop images under different lighting conditions. Adjust the left and right camera position and shooting direction. The two cameras are at about the same height, with a distance of 6-12 cm; properly adjust the direction of the cameras so that both cameras can obtain a relatively complete image of the target, and then fix the positions and shooting directions of the two cameras.

Compared with the prior art, the present invention has the following advantages and beneficial effects:

1. The present invention mainly studies the dense matching method of crop images in the natural environment. Most of the current research results are mainly in the field of industrial applications, and these results cannot be directly applied for crop analysis purposes. The invention of the present invention can achieve more accurate dense matching results for captured crop images in a natural environment, can meet the purpose of crop state analysis, and provide a basis for further real-time analysis and processing.

2. Crop images play an extremely important role in the analysis of crop status. The growth status of crops and information about pests and diseases can be obtained through crop image analysis. With the development of embedded technology and wireless transmission technology, people can collect images of crops and transmit them anytime and anywhere, so that a large amount of image information of crops can be obtained. However, information is abundant and knowledge is scarce. People hope to obtain rich crop-scale state information such as the shape and size of pest and disease areas, leaf and fruit size and plant height, and crop dynamic change process from a large number of collected images. However, since the existing image dense matching algorithms cannot meet the accuracy requirements of crop state analysis, people cannot obtain rich crop-scale information from images.

The inventor found through a large number of crop image experiments that the main factor affecting the dense matching of crop images is caused by the boundaries of leaves and fruits. The invention starts from this feature and achieves dense crop matching by analyzing the sub-image dense matching method of the target. the goal of.

Description of drawings

Fig. 1 is the flowchart of a kind of crop image dense matching method described in the present invention;

Fig. 2 is a schematic structural diagram of a crop image dense matching system according to the present invention.

Detailed ways

The present invention will be further described in detail below in conjunction with the embodiments and the accompanying drawings, but the embodiments of the present invention are not limited thereto.

As shown in Figure 1, a crop image dense matching method includes the steps in the following order:

S1. Analyze the center matching of the target: For a pixel in the center area of the target to be matched in the left picture, calculate its matching point in the right picture through the SIFT flow field;

S2. Analyzing target subimage segmentation: on the basis of obtaining the target center matching point, intercepting the subimages of two images centered on the target center point and its matching point;

S3. Subgraph boundary segmentation and subgraph dense matching: In the two subgraphs, the image outside the boundary of the analysis target is removed through the boundary segmentation algorithm; for each pixel of the analysis target, search for the shortest distance to its SIFT vector in the right subgraph The pixel points are used as matching points, so as to obtain each pixel matching point and obtain the dense matching of the target crop;

S4. Dense matching in the original image: on the basis of the dense matching result of the sub-image, through the inverse transformation of the sub-image intercepted in step S2, the dense matching result of the source image is obtained.

In step S2, the size of the target sub-image segmentation is set according to the size of the monitoring target.

In step S3, the boundary segmentation algorithm is a canny boundary segmentation algorithm.

As shown in Figure 2, a crop image dense matching system includes a thin client, a camera connected to the thin client, a first AP, a power supply module for the thin client and the first AP, and a server, which is connected to the server The second AP.

The power supply module includes a controller, a storage battery, a solar panel, and an inverter respectively connected to the controller, and the inverter is respectively connected to the thin client and the first AP.

The cameras mentioned are two Logitch Pro9000 cameras, the distance between the two cameras is 6-12 cm from the ground, and the directions of the two cameras are the same.

details as follows:

The solar power supply system is used to power the wireless transmission and image acquisition system. The solar power supply system is composed of 18V100W monocrystalline silicon solar panel, 12V to 220V500W pure sine wave inverter, 12V100AH lead-acid battery and solar controller. The wireless transmission is realized by using the long-distance point-to-point Internet connection between the first Ap and the second Ap.

The equipment for collecting crop images in the embodiment is two Logitch Pro9000 cameras. Many experiments have proved that the Logitch Pro9000 cameras can collect higher-quality crop images under different lighting conditions. Adjust the left and right camera position and shooting direction. The two cameras are approximately at the same height, with a distance of 6-12 cm. The directions of the two cameras are roughly the same. Properly adjust the direction of the camera so that both cameras can obtain a relatively complete image of the target, and then fix the position and shooting direction of the two cameras.

Connect two Logitch Pro9000 cameras with Tianyuan Tengchuang brand thin client at the same time with USB interface, and install Windows operating system on the thin client. The acquisition of crop images requires the thin client to control two cameras to simultaneously capture crop images. At the same time, the capture software uses VC to call OpenCV to complete the encoding.

The above-mentioned embodiment is a preferred embodiment of the present invention, but the embodiment of the present invention is not limited by the above-mentioned embodiment, and any other changes, modifications, substitutions, combinations, Simplifications should be equivalent replacement methods, and all are included in the protection scope of the present invention.

Claims (6)

1. crop map is as a dense matching method, it is characterized in that, comprises the step of following order:

S1. the center coupling of evaluating objects: to a pixel of the central area of left figure target to be matched, go out its match point in right figure by SIFT Flow Field Calculation;

S2. evaluating objects dividing sub-picture: on the basis obtaining target's center's match point, intercept the subgraph of two width images centered by target's center's point and its match point;

S3. subgraph boundary segmentation and subgraph dense matching: in two subgraphs, removes the image outside evaluating objects border by boundary segmentation algorithm; To the pixel of each evaluating objects, in right subgraph, search and the pixel of its SIFT vector minimum distance are as match point, thus obtain each pixel matching point, obtain the dense matching of target crop;

S4. the dense matching in former figure: on the basis of subgraph dense matching result, the inverse transformation of the subgraph intercepted by step S2, obtains source images dense matching result.

2. crop map according to claim 1 is as dense matching method, it is characterized in that, in step S2, the size of described target dividing sub-picture will set according to the size of monitoring objective.

3. crop map according to claim 1 is as dense matching method, it is characterized in that, in step S3, described boundary segmentation algorithm is canny boundary segmentation algorithm.

4. crop map is as a dense matching system, it is characterized in that: comprise thin client, the camera be connected with thin client, an AP, also comprises the supply module for thin client, an AP power, and server, the 2nd AP that is connected with server.

5. crop map according to claim 4 is as dense matching system, it is characterized in that: described supply module comprises controller, and the accumulator, solar panels, the inverter that are connected with controller respectively, described inverter is connected with thin client, an AP respectively.

6. crop map according to claim 4 is as dense matching system, it is characterized in that: described camera is two Logitch Pro9000 cameras, and two camera distance 6 ~ 12 centimetres, ground, two camera directions are consistent.