CN116465840A - A device and method for quickly identifying the age of tangerine peel - Google Patents

Abstract

The application provides a device and method for quickly identifying the age of tangerine peel, which relates to the technical field of medicinal material detection, including a multi-spectral camera, a light source, and a processor electrically connected to the multi-spectral camera and the light source. The light source includes a plurality of LED lamp beads with different wavelengths and adjustable intensities; the light source is adjusted to meet the preset distance and angle relationship with the sample to be tested. The year results of the sample, the distance between the multispectral camera and the sample to be tested meet the preset requirements. The imaging effect is evaluated by adjusting the parameters of the device, so that the device is in the best state to identify the age of the sample to be tested, and improve the accuracy of the results; realize fast, non-destructive, non-contact, low-cost, and low-use threshold identification of the age of tangerine peel.

Description

A device and method for quickly identifying the age of tangerine peel

technical field

The application relates to the technical field of medicinal material detection, and in particular to a device and method for rapidly identifying the age of tangerine peel.

Background technique

Chenpi is a food and medicinal product made from mature citrus peels dried and stored for a long time. The longer the storage time, the higher the medicinal value; For this reason, it is necessary to identify the age of tangerine peel.

Currently, there are mainly two methods for identifying the age of tangerine peel: a kind of traditional sensory identification method, which is to judge the age of tangerine peel by its appearance and smell characteristics. The sensory identification method is still a common method for identifying the age of tangerine peel because it is more convenient and quick, but this method relies on the experience of the appraiser, has strong subjectivity and randomness, and is difficult to identify the age of tangerine peel very accurately. Another kind is physical and chemical analysis identification method, and this method adopts some physical and chemical means, adopts specific reagent and instrument to process and analyze the dried orange peel to be tested. Physical and chemical analysis and identification methods can obtain relatively accurate analysis results, but they require professional technicians and experimental equipment, and the analysis process is complex, time-consuming, and costly, making it difficult to widely promote and use in real life.

Contents of the invention

The purpose of the embodiment of the present application is to provide a device and method for quickly identifying the age of orange peel, which can quickly, conveniently and accurately identify the age of orange peel.

An aspect of the embodiment of the present application provides a device for quickly identifying the age of tangerine peel, including a multispectral camera, a light source, and a processor electrically connected to the multispectral camera and the light source, the light source includes a plurality of LED beads with different wavelengths and adjustable intensities; the light source is adjusted to meet the preset distance and angle relationship with the sample to be tested, and then emits a beam toward the sample to be tested in the target area. The processor analyzes and calculates and outputs the year result of the sample to be tested, wherein the distance between the multispectral camera and the sample to be tested meets a preset requirement.

Optionally, a dark box is also included, and the multispectral camera and the light source are both arranged in the dark box; an object stage is also arranged in the dark box, and the target area is preset on the object table, and the sample to be tested is placed in the target area.

Another aspect of the embodiment of the present application provides a method for quickly identifying the age of tangerine peel, which is applied to the processor in the above-mentioned device for quickly identifying the age of tangerine peel. The method includes: adjusting the angle and distance between the light source and the sample to be tested according to a preset algorithm, so that the light source and the sample to be tested meet the preset distance and angle relationship;

After the light source and the sample to be tested satisfy the preset distance and angle relationship, multiple LED lamp beads in the light source respectively emit light beams towards the sample to be tested located in the target area;

receiving light beams reflected by the light beams of a plurality of LED lamp beads after irradiating the sample to be tested through a multi-spectral camera, so as to obtain original spectral image data of the sample to be tested under illumination of different wavelengths, wherein the distance between the multi-spectral camera and the sample to be tested meets preset requirements;

performing a region-of-interest extraction operation according to the original spectral image data to obtain region-of-interest data;

According to the data of the region of interest, input the age identification model of orange peel to obtain the age of the sample to be tested.

Optionally, the method also includes:

Adjusting the distance between the multispectral camera and the target area, so that the images collected by the multispectral camera can cover the target area to meet the preset requirements.

Optionally, the adjusting the angle and distance between the light source and the sample to be tested according to a preset algorithm, so that the light source and the sample to be tested meet a preset distance and angle relationship, comprising:

During the process of adjusting the angle and distance between the light source and the sample to be tested, collecting a standard spectral image of a standard reflective whiteboard in the target area through the multispectral camera;

According to the spectral image of the standard reflective whiteboard and the formula Calculating to determine whether the angle and distance between the light source and the sample to be tested satisfy the preset distance and angle relationship; wherein, Indicates the variance of the gray value of the image at the i-th wavelength, is the average gray value of the image at the i-th wavelength, W and H are the number of pixels in the width and height of the target area respectively, f _i (x, y) represents the gray-scale value of the image at the coordinates (x, y) at the i-th wavelength.

Optionally, after adjusting the angle and distance between the light source and the sample to be tested according to a preset algorithm, so that the light source and the sample to be tested meet the preset distance and angle relationship, the method further includes:

Adjusting the exposure parameters of the multi-spectral camera, so that the gray value of no more than 5% of the pixels in the target area of the imaging picture reaches the maximum value or the minimum value.

Optionally, the multi-spectral camera respectively receives the light beams reflected by the light beams of multiple LED lamp beads after irradiating the sample to be tested, so as to obtain the original spectral image data of the sample to be tested under different wavelengths of illumination, including:

receiving the light beams of the sample to be tested in multiple different bands fed back by the multispectral camera, so as to acquire the original spectral image data of the sample to be tested in multiple different bands.

Optionally, after receiving the reflected light beams of the plurality of LED light beads by the multi-spectral camera after irradiating the sample to be tested, so as to obtain the original spectral image data of the sample to be tested under illumination of different wavelengths, the method further includes:

The key point features of the original spectral image data of a plurality of different bands are respectively extracted by the ORB feature detection module;

Using a brute force matching module to match the key point features, so that the key point features of the original spectral image data of a plurality of different bands overlap;

The original spectral image data is converted according to the matched key point features, so as to realize pixel-level alignment of the original spectral image data in multiple different bands, and obtain multiple registered and transformed spectral images.

Optionally, placing a standard reflective whiteboard with known reflectivity in the target area to replace the sample to be tested; converting the original spectral image data according to the matched key point features to achieve pixel-level alignment of the original spectral image data in multiple different bands, and after obtaining multiple converted images, the method further includes:

Under the same shooting conditions as shooting the sample to be tested, cover the lens of the multispectral camera with a lens cover, and obtain dark current data of the multispectral camera;

After removing the lens cover, obtain the standard spectral image of the standard reflective whiteboard fed back by the multispectral camera;

According to the standard spectral image of the standard reflective whiteboard, the reflectance-corrected spectral image is calculated by the formula _I1 =( _I0 -B)/((WB)/r); wherein, _I1 is the corrected spectral image data, _I0 is the spectral image data after registration conversion, B is the dark current data, W is the standard reflective whiteboard data, and r is the reflectance of the standard reflective whiteboard.

Optionally, performing the region-of-interest extraction operation according to the original spectral image data to obtain the region-of-interest data includes:

randomly copying the reflectance-corrected spectral image under any band for making a mask, and using a median filter method to denoise the mask image;

binarizing the mask image by an adaptive threshold segmentation method to divide the background area and the sample image area to be tested;

performing morphological processing of opening and closing operations on the binarized mask image in turn, eliminating white spots in the background area and black spots in the sample image area to be tested;

Extracting from the mask image to obtain the outer frame of the sample image area to be tested;

Selecting 40% to 50% of the center of the outer frame as the region of interest of the sample image region to be tested;

Based on the coordinates of the mask image, the reflectance-corrected multiple spectral image data are cropped to extract the data of the region of interest.

Optionally, the input of the orange peel age identification model according to the data of the region of interest to obtain the year of the sample to be tested includes:

Use 1*1 convolutional layer to reduce the dimension of the input spectral image channel number;

Extract image feature by image feature extraction network module, described image feature is image one-dimensional vector;

Use an average pooling layer with a kernel size of 9*9 to reduce the number of pixels in the input spectral image;

Adjusting the number of image channels by 1*1 convolution, performing feature extraction and feature combination between image channels represented by different wavelengths, to obtain a three-dimensional feature image;

After taking out the outputted three-dimensional feature images sequentially by pixels, rearranging them into two-dimensional feature vector groups;

The two-dimensional feature vector group is input to the spectral feature extraction network module to extract spectral features, and the spectral features are spectral one-dimensional vectors;

splicing the image one-dimensional vector and the spectrum one-dimensional vector;

The year probability distribution of the sample to be tested is obtained through the fully connected layer and the Softmax activation function.

The device and method for quickly identifying the age of tangerine peel provided by the embodiments of the present application irradiates the sample to be tested with a light source, and the multispectral camera receives the beam reflected by the sample to be tested and feeds back the detection information to the processor. After the distance meets the preset requirements, use this device to identify the age of the sample to be tested. By adjusting the parameters of the device to evaluate the imaging effect, the device is in the best state to identify the age of the sample to be tested, so as to improve the accuracy of the year of the sample to be tested. In addition, when the device is used to identify the age of the sample to be tested, there is no need to touch the sample to be tested, and non-destructive, non-contact detection is realized. The detection speed is fast, the cost is low, and the threshold for use is low.

Description of drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present application, the accompanying drawings required in the embodiments of the present application will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present application, and therefore should not be considered as limiting the scope. For those of ordinary skill in the art, other related drawings can also be obtained according to these drawings without creative work.

Fig. 1 is the device structure schematic diagram of the rapid identification of the age of orange peel provided by the present embodiment;

Fig. 2 is one of the method flowcharts of the rapid identification of the year of orange peel provided by the present embodiment;

Fig. 3 is the second flow chart of the method for quickly identifying the age of orange peel provided in this embodiment.

Icons: 10-obscura; 100-multispectral camera; 101-light source; 102-stage; 200-sample to be tested.

Detailed ways

The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

In the description of this application, it should be noted that the orientation or positional relationship indicated by the terms "inner", "outer", etc. is based on the orientation or positional relationship shown in the drawings, or the orientation or positional relationship that is usually placed when the product of the application is used. It is only for the convenience of describing the application and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation of the application. In addition, the terms "first", "second", etc. are only used for distinguishing descriptions, and should not be construed as indicating or implying relative importance.

It should also be noted that, unless otherwise clearly specified and limited, the terms "setting" and "connection" should be understood in a broad sense, for example, it may be a fixed connection, a detachable connection, or an integral connection; it may be a direct connection, or an indirect connection through an intermediary, or an internal connection between two components. Those of ordinary skill in the art can understand the specific meanings of the above terms in this application in specific situations.

Among the existing identification methods for the age of tangerine peel, the disadvantage of the sensory identification method is that it depends on human experience, has strong subjectivity and randomness, and is difficult to identify the age of tangerine peel very accurately; and the physical and chemical analysis identification method has the disadvantage of requiring professional technicians and experimental equipment. The analysis process is complicated, time-consuming, and costly, and it is difficult to be widely used in real life.

In view of this, in order to solve the above problems, the embodiment of the present application provides a device and method for quickly identifying the age of tangerine peel, which can perform rapid, non-destructive, non-contact, low-cost, and low-use threshold identification of the age of tangerine peel.

Specifically, as shown in FIG. 1 , an aspect of the embodiment of the present application provides a device for quickly identifying the age of tangerine peel (hereinafter referred to as the device), including: a multispectral camera 100, a light source 101, and a processor electrically connected to the multispectral camera 100 and the light source 101 respectively. The camera 100 respectively receives the light beams reflected by the light beams of multiple LED lamp beads irradiating the sample 200 to be tested and feeds back the detection information under different wavelengths of light to the processor. The processor analyzes and calculates and outputs the year results of the sample 200 to be tested, wherein the distance between the multispectral camera 100 and the sample 200 to be tested meets the preset requirements.

A multi-spectral imaging system is established through the above-mentioned device, wherein the multi-spectral camera 100 is developed based on a MEMS chip, and can quickly acquire spectral images in multiple different bands. The light source 101 is composed of multiple LED lamp beads with different wavelengths and whose intensity can be adjusted in a cross arrangement. By adjusting the light intensity of the lamp beads of each wavelength, the distribution of the light intensity of the entire light source 101 with wavelength can be adjusted to achieve the optimal lighting effect.

For example, in the present application, the multispectral camera 100 is placed vertically above the sample to be tested 200, and there are two light sources 101, which are symmetrically arranged on both sides above the sample to be tested 200; when the device is used for identification, the light source 101 emits a beam toward the sample to be tested 200, and the sample to be tested 200 reflects the beam after receiving it, and reflects the reflected beam to the multispectral camera 100. The year result of the sample to be tested 200 is obtained and output, so as to obtain the year result of the sample to be tested 200 .

It should be noted that, during the use of the above-mentioned device, before the identification of the sample 200 to be tested, the position between the light source 101 and the sample to be tested 200, the multispectral camera 100 and the sample to be tested 200 needs to be adjusted, so as to adjust the parameters of the multispectral imaging system and evaluate the imaging effect; specifically, the preset distance and angle relationship between the light source 101 and the sample to be tested 200 should be satisfied, so that the lighting effect is in the best state; similarly, the distance between the multispectral camera 100 and the sample to be tested 200 meets the preset requirements, so that The device can identify the sample 200 to be tested under the best condition.

In addition, the device also includes a dark box 10, in which the multispectral camera 100 and the light source 101 are both arranged; in the dark box 10, a stage 102 is also arranged, and a target area is preset on the stage 102, and the sample 200 to be tested is placed in the target area.

The black box 10 is assembled with black opaque boards to reduce the interference of the external environment on the imaging system; the bottom of the dark box 10 is a black metal platform with holes, on which support rods and brackets can be installed for fixing the multispectral camera 100 and the light source 101; the preset target area on the stage 102 is used to place the sample 200 to be tested, and the placement surface is black.

Therefore, the device for quickly identifying the age of orange peel provided in the embodiment of the present application irradiates the sample 200 to be tested through the light source 101, the multispectral camera 100 receives the light beam reflected by the sample 200 to be tested and feeds back the detection information to the processor, and the processor analyzes and calculates and outputs the year result of the sample 200 to be tested; wherein, the light source 101 includes a plurality of LED lamp beads with different wavelengths and whose intensity can be adjusted. 101 needs to be adjusted to meet the preset distance and angle relationship with the sample to be tested 200. After the distance between the multispectral camera 100 and the sample to be tested 200 meets the preset requirements, the device is used to identify the year of the sample to be tested 200, and the parameters of the device are adjusted to evaluate the imaging effect, so that the device is in the best state to perform year identification of the sample to be tested 200, so as to improve the accuracy of the year result of the sample to be tested 200; 00, to achieve non-destructive and non-contact detection, and the detection speed is fast, the cost is low, the threshold of use is low, and the scope of application is wide. It has become a convenient and fast means of identifying the age of tangerine peel.

On this basis, another aspect of the embodiment of the present application also discloses a method for quickly identifying the age of tangerine peel, which is applied to the processor in the above-mentioned device for quickly identifying the age of tangerine peel. As shown in Figure 2, the method includes:

S10: Adjust the distance between the multi-spectral camera 100 and the target area, so that the images collected by the multi-spectral camera 100 can cover the target area to meet preset requirements.

The multispectral camera 100 is vertically arranged above the sample to be tested 200, and the distance between the multispectral camera 100 and the sample to be tested 200 is adjusted so that the image of the sample to be tested 200 on the multispectral camera 100 can cover the target area. The target area is a pre-defined area on the stage 102 for placing the sample to be tested 200. The target area is larger than the area of the sample to be tested 200, and the sample to be tested 200 is generally placed in the center of the target area.

During adjustment, manual adjustment or automatic adjustment can be used; for example, the multispectral camera 100 can be driven by an automatic mechanism to move close to or away from the sample 200 to be tested, and the image of the multispectral camera 100 can be used to judge whether the adjustment is in place. Wherein, the picture of the multispectral camera 100 can be fed back to the processor, and the processor compares the preset standard picture with the picture of the multispectral camera 100 for automatic judgment.

S100: Adjust the angle and distance between the light source 101 and the sample to be tested 200 (target area) according to a preset algorithm, so that the light source 101 and the sample to be tested 200 meet the preset distance and angle relationship.

In the process of adjusting the angle and distance between the light source 101 and the sample to be tested 200, a standard reflective whiteboard is placed in the target area to replace the sample to be tested 200, and the standard spectral image of the standard reflective whiteboard in the target area is collected by the multispectral camera 100;

Spectral image and formula based on standard reflective whiteboard Calculate and determine whether the angle and distance between the light source 101 and the sample 200 to be tested satisfy the preset distance and angle relationship.

in, Indicates the variance of the gray value of the image at the i-th wavelength, is the average gray value of the image at the i-th wavelength, W and H are the number of pixels in the width and height of the target area respectively, f _i (x, y) represents the gray value of the image at the coordinate (x, y) at the i-th wavelength.

In a practicable manner of the present application, the adjustment range of the distance is 30 cm to 60 cm, and the step length (the minimum unit of one adjustment) is 5 cm; the adjustment range of the angle (the angle between the light source 101 and the horizontal direction) is 70° to 85°, and the step length is 3°.

The purpose of the adjustment is to uniformly illuminate the target area in the imaging picture. The evaluation method for the uniformity of illumination is: place a standard reflective whiteboard with a size that can cover the target area on the stage 102, and after each adjustment, use the multispectral camera 100 to take a spectral image of the standard reflective whiteboard, and use the formula The illumination uniformity index I is calculated, and the illumination uniformity index I represents the average value of the variance of the gray value of the image at each wavelength in the target area, and the smaller the value of the illumination uniformity index I, the better the illumination uniformity.

After adjusting the angle and distance between the light source 101 and the sample 200 to be tested, adjust the exposure parameters of the multispectral camera 100 so that the grayscale values of no more than 5% of the pixels in the target area of the imaging image reach the maximum or minimum value.

Specifically, adjust the distribution of the light intensity of the light source 101 along with the wavelength, so that the target area is fully illuminated at each wavelength captured by the multispectral camera 100; adjust the exposure parameters of the multispectral camera 100, so that the imaging picture is overexposed or underexposed; overexposure or underexposure respectively means that the gray value of more than 5% of the pixels in the target area of the imaging picture reaches the maximum value (255) or the minimum value (0).

Through the above steps, the parameters of the system are adjusted to the best state, and then the sample 200 to be tested is placed in the target area for identification, so as to obtain the best imaging data of the sample 200 to be tested, and then the result of identification of the year of the sample 200 to be tested is more accurate.

S110: After the light source 101 and the sample to be tested 200 meet the preset distance and angle relationship, make the plurality of LED lamp beads in the light source 101 respectively emit light beams toward the sample to be tested 200 located in the target area.

Step S100 has adjusted the light source 101 and the sample to be tested 200 to meet the preset distance and angle relationship. At this time, the sample to be tested 200 is placed in the target area to replace the standard reflective white board, and the light source 101 emits a light beam toward the sample to be tested 200. At this time, the sample to be tested 200 can be fully illuminated, and the illumination uniformity of the light beam received by the sample to be tested 200 is the best.

S120: The multi-spectral camera 100 respectively receives the light beams reflected by the light beams of multiple LED beads irradiating the sample 200 to be tested, so as to obtain the original spectral image data of the sample 200 to be tested under different wavelengths of light, wherein the distance between the multi-spectral camera 100 and the sample 200 to be tested meets the preset requirements.

After the light source 101 irradiates the test sample 200, the test sample 200 reflects the light beam to the multispectral camera 100, and the processor receives the light beams of the test sample 200 in multiple different bands fed back by the multispectral camera 100, so as to obtain the original spectral image data of the test sample 200 in multiple different bands.

For example, the sample 200 to be tested is placed in the target area on the stage 102, and the original spectral image data of the sample 200 to be tested in ten different bands are collected by the multispectral camera 100, wherein the central wavelengths of the ten bands are: 720nm, 743nm, 766nm, 789nm, 812nm, 835nm, 858nm, 881nm, 904nm and 927nm; The image resolution is 1280*1024 pixels, in other words, ten spectral images of the sample 200 to be tested can be obtained; in addition, the number of bands is determined according to the performance of the multispectral camera 100 itself.

S130A: Perform registration correction on the original spectral image data to obtain correction data. Registration correction includes image registration of different bands and reflectance correction. Since the multispectral camera 100 scans each band sequentially during imaging, when the multispectral camera 100 shakes during the shooting process, the images of different bands will have position deviations, so this deviation needs to be corrected by means of image registration.

The specific method of image registration is: through the ORB feature detection module in the processor, the key point features of the original spectral image data of a plurality of different bands are respectively extracted; the key point features are the edge corner features of the spectral image of the sample 200 to be tested;

Use the brute force matching module to match the key point features, so that the key point features of the original spectral image data of multiple different bands overlap; collect the spectral images of ten bands as described above, superimpose and overlap the ten spectral images, pull the edge corner points of the sample to be tested 200 on the ten spectral images, and align the edge corner points of the sample to be tested 200 on the ten spectral images respectively.

The original spectral image data is converted according to the matched key point features, and the pixel-level alignment of the original spectral image data of multiple different bands is realized, and multiple registered and transformed spectral images are obtained.

Then reflectance correction is performed, and the purpose of reflectance correction is to eliminate the influence of ambient light and camera dark current on multispectral imaging. The specific method for correcting the reflectance is: place a standard reflective whiteboard with known reflectance in the target area to replace the sample 200 to be tested;

Under the same shooting conditions as shooting the sample to be tested 200, a lens cover is used to cover the lens of the multispectral camera 100, and the dark current data of the multispectral camera 100 is obtained by shooting with the multispectral camera 100;

After removing the lens cover, the standard reflective whiteboard is photographed by the multispectral camera 100, and the standard spectral image of the standard reflective whiteboard fed back by the multispectral camera 100 is obtained;

According to the standard spectral image of the standard reflective whiteboard, the reflectance-corrected spectral image is calculated by the formula I ₁ = (I ₀ -B)/((WB)/r); wherein, I ₁ is the corrected spectral image data, I ₀ is the spectral image data after registration conversion, B is the dark current data, W is the standard reflective whiteboard data, and r is the reflectance of the standard reflective whiteboard.

S130: Perform a region-of-interest extraction operation according to the spectral image data to obtain region-of-interest data.

Region of interest extraction, that is, to filter out a specific region in the multispectral image for analysis, and eliminate the influence of the background on the analysis results. To be precise, the region of interest extraction is to operate on the corrected spectral image. The specific method of region of interest extraction is:

Randomly copy the reflectance-corrected spectral image under any band to make a mask, and use the median filter method to denoise the mask image;

Binarize the mask image by means of adaptive threshold segmentation to divide the background area and the 200 image area of the sample to be tested;

Carry out the morphological processing of the opening operation and the closing operation on the binarized mask image in turn, and eliminate the white spots in the background area and the black spots in the image area of the sample 200 to be tested;

Extract the outer frame of the image area of the sample to be tested 200 from the mask image, select a frame of a specific shape as the outer frame, and frame the image area of the sample 200 to be tested by the outer frame, for example, the outer frame can be a rectangular frame;

Select the area of 40% to 50% of the center of the outer frame as the region of interest in the image area of the sample 200 to be tested; select the center area of the outer frame as the region of interest, and generally take 40% to 50% of the central area as the selection standard;

The above steps are all processing the mask image to obtain the coordinates of the mask image, and then based on the coordinates of the mask image, the reflectance-corrected multiple spectral image data are cut to extract the data of the region of interest. In this step, the aforementioned ten reflectance-corrected spectral images can be substituted into the mask image, and then cut out one by one to obtain ten regions of interest data.

S140: Input the tangerine peel age identification model according to the data of the region of interest to obtain the age of the sample 200 to be tested.

As shown in Figure 3, the age identification model of tangerine peel has a double-branch structure, including image branch and spectral branch. Among them, the image branch structure includes:

The 1*1 convolutional layer is used to reduce the number of input spectral image channels; the ten bands collected above and the data of the ten regions of interest obtained can be regarded as ten spectral image channels, and the ten spectral images can be reduced into three feature maps.

Image features are extracted through the image feature extraction network module, and the image features are image one-dimensional vectors; a set of image one-dimensional vectors is obtained after the image features are extracted from the three feature maps.

The spectral branch structure includes: using an average pooling layer with a kernel size of 9*9 to reduce the number of pixels of the input spectral image;

Adjust the number of image channels through 1*1 convolution. Generally, ten channels can be adjusted to sixty-four channels, and feature extraction and feature combination between image channels represented by different wavelengths are performed to obtain sixty-four three-dimensional feature images;

After the output three-dimensional feature image is taken out in order according to the pixel points, it is rearranged into a two-dimensional feature vector group; for example, it can be arranged into a set of horizontal or vertical feature vector groups;

The two-dimensional feature vector group is input to the spectral feature extraction network module to extract spectral features, and the spectral feature is a spectral one-dimensional vector; among them, the image feature extraction network module and the spectral feature extraction network module are lightweight convolutional neural networks with similar structures, and their features are both depth separable convolution and ECA attention mechanism modules; the difference is that the image feature extraction network module is a two-dimensional convolutional neural network, and the spectral feature extraction network module is a one-dimensional convolutional neural network.

Concatenate the image one-dimensional vector and the spectrum one-dimensional vector;

The year probability distribution of the sample 200 to be tested is obtained through the fully connected layer and the Softmax activation function, and the year with a larger probability distribution can be used as the final identification result of the year of the sample 200 to be tested.

The method for quickly identifying the age of tangerine peel contains the same structure and beneficial effects as the device for quickly identifying the age of tangerine peel in the foregoing embodiments. The structure and beneficial effects of the device for quickly identifying the age of orange peel have been described in detail in the foregoing embodiments, and will not be repeated here.

The above are only examples of the present application, and are not intended to limit the protection scope of the present application. For those skilled in the art, the present application may have various modifications and changes. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of this application shall be included within the protection scope of this application.

Claims (11)

1. A device for rapid identification of the year of dried orange peel, comprising: the device comprises a multispectral camera, a light source and a processor electrically connected with the multispectral camera and the light source respectively, wherein the light source comprises a plurality of LED lamp beads with different wavelengths and adjustable intensity; the multi-spectrum camera receives the light beams reflected after the light beams of the LED lamp beads irradiate the sample to be measured respectively and feeds back detection information under different wavelength illumination to the processor, and the processor analyzes, calculates and outputs a year result of the sample to be measured, wherein the distance between the multi-spectrum camera and the sample to be measured meets the preset requirement.

2. The device for rapid identification of year of dried orange peel according to claim 1, further comprising a camera box, wherein the multispectral camera and the light source are both disposed in the camera box; an objective table is further arranged in the camera bellows, the objective table is provided with the target area in advance, and the sample to be measured is placed in the target area.

3. A method for rapid identification of years of dried orange peel, applied to a processor in an apparatus for rapid identification of years of dried orange peel as claimed in claim 1 or 2, the method comprising:

adjusting the angle and the distance between the light source and the sample to be tested according to a preset algorithm so that the light source and the sample to be tested meet the preset distance and angle relation;

after the light source and the sample to be detected meet the preset distance and angle relation, enabling a plurality of LED lamp beads in the light source to respectively face to the sample to be detected in the target area to emit light beams;

respectively receiving the light beams reflected after the light beams of the LED lamp beads irradiate the sample to be detected through a multispectral camera so as to obtain original spectrum image data of the sample to be detected under different wavelength illumination, wherein the distance between the multispectral camera and the sample to be detected meets the preset requirement;

performing region of interest extraction operation according to the original spectrum image data to obtain region of interest data;

and inputting a dried orange peel year identification model according to the data of the region of interest to obtain the year of the sample to be detected.

4. A method for rapid identification of year of dried orange peel according to claim 3, further comprising:

and adjusting the distance between the multispectral camera and the target area so that the picture acquired by the multispectral camera can cover the target area to meet the preset requirement.

5. The method for rapid identification of year of dried orange peel according to claim 3, wherein the adjusting the angle and distance between the light source and the sample to be measured according to a predetermined algorithm to satisfy a predetermined distance and angle relationship between the light source and the sample to be measured comprises:

in the process of adjusting the angle and the distance between the light source and the sample to be measured, acquiring a standard spectrum image of a standard reflection white board in the target area through the multispectral camera;

according to the spectral image of the standard reflective whiteboard, formula i= (Σ) _i ^N _＝1 V _i ) N, calculating and determining the light source and the to-be-measured light sourceWhether the angle and the distance between the samples meet the preset distance and angle relation or not; wherein,,representing the gray value variance of the image at the i-th wavelength,for the average gray value of the image at the ith wavelength, W, H is the number of pixels of the target area width and height, f _i (x, y) represents the gray value of the image at the coordinates (x, y) at the ith wavelength.

6. The method according to claim 5, wherein after adjusting the angle and distance between the light source and the sample to be measured according to a predetermined algorithm so that the light source and the sample to be measured satisfy a predetermined distance and angle relationship, the method further comprises:

and adjusting the exposure parameters of the multispectral camera to enable the gray value of no more than 5% of pixel points in the target area of the imaging picture to reach the maximum value or the minimum value.

7. The method for rapid identification of year of dried orange peel according to any one of claims 3 to 6, wherein the receiving, by a multispectral camera, the light beams reflected after the light beams of the plurality of LED lamp beads irradiate the sample to be detected, respectively, to obtain the original spectral image data of the sample to be detected under different wavelength illumination, includes:

and receiving the light beams of the sample to be detected in a plurality of different wave bands fed back by the multispectral camera so as to acquire the original spectrum image data of the sample to be detected in a plurality of different wave bands.

8. The method for rapid identification of year of dried orange peel according to claim 7, wherein after receiving the light beams reflected after the light beams of the plurality of LED lamp beads irradiate the sample to be detected by the multispectral camera respectively to obtain the original spectral image data of the sample to be detected under different wavelength illumination, the method further comprises:

respectively extracting key point features of the original spectrum image data of a plurality of different wave bands through an ORB feature detection module;

matching the key point features by adopting a brute force matching module to enable the key point features of the original spectrum image data of a plurality of different wave bands to coincide;

and converting the original spectrum image data according to the matched key point characteristics to realize pixel level alignment of the original spectrum image data of a plurality of different wave bands, thereby obtaining a plurality of spectrum images after registration conversion.

9. The method for rapid identification of year of dried orange peel according to claim 8, wherein a standard reflective white board with known reflectivity is placed in the target area to replace the sample to be tested; the original spectrum image data is converted according to the matched key point characteristics, so that pixel level alignment of the original spectrum image data with a plurality of different wave bands is realized, and after a plurality of converted images are obtained, the method further comprises the steps of:

under the same shooting condition as that of shooting the sample to be detected, covering a lens of the multispectral camera by adopting a lens cover to acquire dark current data of the multispectral camera;

after removing the lens cover, acquiring a standard spectrum image of a standard reflection white board fed back by the multispectral camera;

according to the standard spectrum image of the standard reflection white board, the standard spectrum image is represented by the formula I ₁ ＝(I ₀ -B)/((W-B)/r) calculating to obtain a reflectance corrected spectral image; wherein I is ₁ Is corrected spectral image data, I ₀ Is the spectrum image data after registration conversion, B is dark current data, W is the standard reflection white board data, and r is the reflectivity of the standard reflection white board.

10. The method for rapid identification of year of dried orange peel according to claim 9, wherein the extracting the region of interest according to the original spectral image data to obtain the region of interest data comprises:

randomly copying the spectral image corrected by the reflectivity under any wave band, making a mask, and adopting a median filtering method to reduce noise of the mask image;

binarizing the mask image by a self-adaptive threshold segmentation method to divide a background area and an image area of a sample to be detected;

performing morphological processing of open operation and close operation on the binarized mask image in sequence to eliminate white points in the background area and black points in the image area of the sample to be detected;

extracting an outer frame of the image area of the sample to be detected from the mask image;

selecting a region of 40% -50% of the center of the outer frame as an interested region of the image region of the sample to be detected;

and cutting the plurality of spectral image data after the reflectivity correction based on the coordinates of the mask image, and extracting the data of the region of interest.

11. The method for rapid identification of year of pericarpium Citri Tangerinae of claim 10, wherein inputting the year identification model according to the region of interest data to obtain the year of the sample to be tested comprises:

adopting a 1*1 convolution layer to reduce the dimension of the input spectral image channel number;

extracting image features by an image feature extraction network module, wherein the image features are image one-dimensional vectors;

adopting an average pooling layer with a kernel size of 9*9 to reduce the dimension of the pixel number of the input spectrum image;

the number of the image channels is adjusted through 1*1 convolution, and feature extraction and feature combination among the image channels represented by different wavelengths are carried out, so that a three-dimensional feature image is obtained;

sequentially taking out the output three-dimensional characteristic images according to pixel points, and rearranging the three-dimensional characteristic images into a two-dimensional characteristic vector group;

inputting the two-dimensional feature vector group into a spectrum feature extraction network module to extract spectrum features, wherein the spectrum features are spectrum one-dimensional vectors;

splicing the image one-dimensional vector and the spectrum one-dimensional vector;

and obtaining the annual probability distribution of the sample to be tested through the full connection layer and the Softmax activation function.