Soybean germplasm resource high-throughput measurement method based on image recognition

Abstract

The invention relates to the technical field of general image data processing or generation, in particular to a soybean germplasm resource high-throughput measurement method based on image recognition, and the method comprises the steps: obtaining a correction image of a target measurement object; based on the corrected image, pod contour image parameters and grain contour image parameters are extracted, and size correction is carried out on the image; calculating a pod length parameter, a grain length parameter, a pod width parameter and a grain width parameter based on the corrected image; obtaining a pod width change curve based on the pod width parameter; calculating the pod number based on the pod width change curve; extracting colors based on the corrected image to obtain pod color data and grain color data; and verifying a pod length parameter, a grain length parameter, a pod width parameter, a grain width parameter, pod color data and grain color data based on the decision coefficient and the root mean square error. The problems that in the prior art, time and labor are consumed, and precision is low are solved.

Description

Technical Field

[0001] The present application relates to the general field of image data processing or generation technology, and in particular, to a high-throughput measurement method for soybean germplasm resources based on image recognition. Background Art

[0002] Soybean belongs to the subfamily Papilionatae, genus Glycine and subgenus Soja. It is native to China and is the most important source of vegetable protein and vegetable oil in the world today. In recent years, my country's total soybean imports have continued to increase, and the soybean self-sufficiency rate is only 15%. Faced with this situation, my country's arable land area is very scarce and cannot be solved by large-scale expansion of soybean planting area. Therefore, improving the yield level is the most practical and effective method. How to cultivate high-yield, high-quality, disease- and pest-resistant soybean varieties is a problem that needs to be solved. However, soybean germplasm resources are the basis and prerequisite for cultivating excellent soybean varieties, so researchers are paying more and more attention to the collection, research and application of excellent germplasm resources.

[0003] Phenotypic identification of excellent soybean germplasm resources is the basis for breeding excellent new soybean varieties. One of the important items is to count the morphological and color traits of pods and grains. The traditional way to obtain relevant data mainly relies on manual means to observe plants, fruits or seeds one by one and manually record and manage data. Manual measurement is not only time-consuming and labor-intensive, but also has problems of low measurement accuracy and poor repeatability. In addition, due to the subjectivity of the experimenters, it is easy to cause deviations in visual evaluation. Summary of the invention

[0004] The purpose of this application is to provide a high-throughput measurement method for soybean germplasm resources based on image recognition, which solves the problems of time-consuming, labor-intensive and low-precision in the prior art.

[0005] The technical solution of this application:

[0006] This application provides a high-throughput measurement method for soybean germplasm resources based on image recognition, including:

[0007] Acquire a calibrated image of the target measurement object;

[0008] Based on the corrected image, the pod contour image parameters and the seed contour image parameters are extracted, and the image is size corrected;

[0009] Based on the corrected image, the pod length parameters, grain length parameters, pod width parameters, and grain width parameters are calculated;

[0010] Based on the pod width parameter, a pod width variation curve is obtained;

[0011] Based on the pod width change curve, the number of pods is calculated;

[0012] Extract color based on the corrected image to obtain pod color data and seed color data;

[0013] Based on the determination coefficient and root mean square error, the pod length parameters, grain length parameters, pod width parameters, grain width parameters, pod color data and grain color data were verified.

[0014] Furthermore, the obtaining of the calibrated image of the target object is specifically as follows:

[0015] Arrange the pods and seeds vertically side by side on a white paper background with a scale grid;

[0016] Place the color chart on the side with the pods and kernels;

[0017] Shoot and obtain the original image;

[0018] Perform noise reduction on the original image based on the median filter method;

[0019] Perform color correction on the denoised image based on the colorimetric chart.

[0020] Furthermore, the extraction of pod contour image parameters and seed contour image parameters based on the corrected image specifically includes:

[0021] Convert the image from RGB color space to HSV color space and calculate the maximum and minimum values ​​of hue, saturation, and brightness of the image;

[0022] Set thresholds;

[0023] The image is divided into black and white parts and the noise is removed to obtain a binary image;

[0024] Perform morphological opening operation on the binary image to divide the adhered kernel pixels into independent parts and mark them as seed points;

[0025] Based on the watershed segmentation algorithm, each pod and each grain is segmented into separate patches;

[0026] Based on the edge detection method, the outer boundary of each pod and each grain is extracted to form a frame graphic;

[0027] Based on the segmentation results of the watershed segmentation algorithm, the area and perimeter of each segmented patch are counted, and the trait index S is calculated:

[0028]

[0029] Among them: S>1.1 is pod, S≤1.1 is grain.

[0030] Furthermore, the image size correction is specifically performed as follows:

[0031] Extract features based on the scale-invariant feature transformation algorithm, match the scale standard format with the image, and determine the location of the scale grid in the image;

[0032] Determine the number of pixels NP0 occupied by the side length of the scale square in the image;

[0033] Based on the number of pixels occupied by the scale grid, get the size correction factor of the image:

[0034] a = length of scale grid side / NP0.

[0035] Furthermore, the pod length parameter, grain length parameter, pod width parameter, and grain width parameter are calculated based on the corrected image, specifically:

[0036] Determine the minimum enclosing rectangle minAreaRect of a single pod and seed;

[0037] After determining the direction of the main axis, rotate the image so that the main axis remains vertical;

[0038] Determine the pod length prediction parameter NP based on the length of the long side of minAreaRect 荚长 and particle length prediction parameter NP 粒长 ;

[0039] Calculate the centroid of each pod and kernel and determine the middle dividing line of the pod or kernel;

[0040] The pod width prediction parameter NP is determined by the length of the line segment between the middle dividing line and the two intersection points of the pod or grain outline. 荚宽 and particle width prediction parameter NP 粒宽 ;

[0041] Based on the size correction factor a, calculate the pod length parameter L 荚长 , grain length parameter W 粒长 , pod width parameter L 荚宽 , particle width parameter W 粒宽 :

[0042] L 荚长 =a*NP 荚长

[0043] L 荚宽 =a*NP 荚宽

[0044] W 粒长 =a*NP 粒长

[0045] W 粒宽 =a*NP 粒宽 .

[0046] Furthermore, based on the pod width parameter, a pod width variation curve is obtained;

[0047] Calculate the number of pods based on the pod width change curve; specifically include:

[0048] From one end of the long side of the pod to the other end, determine the value of the pod width at each location to obtain a pod width change curve;

[0049] The curve is smoothed based on the Savitzky-Golay (SG) filter to remove the jagged edges of the curve and obtain the pod width adjustment curve.

[0050] The peak number of the pod width change adjustment curve is obtained, and the pod number is determined based on the peak number.

[0051] Furthermore, the color is extracted based on the corrected image to obtain the pod color data and the seed color data, specifically:

[0052] Taking the color of RGB color space as reference, the average values ​​of R, G, and B within the borders of pods and seeds were extracted;

[0053] Based on the identification standards of germplasm resources, five colors of pods and seeds in germplasm resource identification were determined, and the average values ​​of R, G, and B color channels of 10 pods and seeds in each color category were calculated as standard reference values;

[0054] The pod color data and the kernel color data were determined by comparing the three-channel average values ​​of each image with the set standard reference values ​​based on the Ming's distance method.

[0055] Furthermore, the target measurement objects are selected from pods and grains of normal and fully mature plants.

[0056] The technical solution of the present application has at least the following advantages and beneficial effects:

[0057] The present application provides a high-throughput measurement method for soybean germplasm resources based on image recognition. Based on image recognition, the steps of human participation and the proportion of human operation involved in the whole process are greatly reduced, the result errors caused by human subjective factors are minimized as much as possible, the time for phenotypic identification of high-quality soybean germplasm resources is shortened, the data phenotypic identification results are improved, and the accuracy and precision are improved, thus solving the problems of time-consuming, labor-intensive and low-precision in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 is a flow chart of the measurement method of the present invention;

[0059] Figure 2 The original image obtained by the embodiment of the present invention;

[0060] Figure 3is a diagram of the correction process of the original image according to an embodiment of the present invention;

[0061] Figure 4 This is a diagram of the segmentation process of pods and seeds in a corrected image according to an embodiment of the present invention;

[0062] Figure 5 The process of determining the minimum circumscribed rectangle of a pod according to an embodiment of the present invention;

[0063] Figure 6 The process of determining the minimum circumscribed rectangle of a grain in an embodiment of the present invention;

[0064] Figure 7 It is a pod width change adjustment curve diagram of three experimental samples in the embodiment of the present invention.

[0065] Figure 8 A reference table of color values ​​after correction in an embodiment of the present invention;

[0066] Fig. 9 This is a schematic diagram of the color of mature bean pods in an embodiment of the present invention;

[0067] Fig.10 This is a schematic diagram of the color of mature seeds in an embodiment of the present invention;

[0068] Fig.11 It is a scatter plot of the measurement results of the pod length parameters in the embodiment of the present invention;

[0069] Fig.12 It is a scatter plot of the measurement results of the pod width parameter in the embodiment of the present invention;

[0070] Fig.13 This is a table of verification results of the number of pods and grains in the embodiments of the present invention. DETAILED DESCRIPTION

[0071] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0072] Example

[0073] Please refer to Figure 1-Figure 13 The present application provides a high-throughput measurement method for soybean germplasm resources based on image recognition, the method comprising:

[0074] Acquire a calibrated image of the target measurement object;

[0075] Based on the corrected image, the pod contour image parameters and the seed contour image parameters are extracted, and the image is size corrected;

[0076] Based on the corrected images, the pod length parameters, grain length parameters, pod width parameters, and grain width parameters are calculated;

[0077] Based on the pod width parameter, a pod width variation curve is obtained;

[0078] Based on the pod width change curve, the number of pods is calculated;

[0079] Extract color based on the corrected image to obtain pod color data and seed color data;

[0080] Based on the determination coefficient and root mean square error, the pod length parameters, grain length parameters, pod width parameters, grain width parameters, pod color data and grain color data were verified.

[0081] It is worth noting that in the prior art, there are many steps of manual operation in the process of phenotypic identification of high-quality soybean germplasm resources, which makes the identification process time-consuming and labor-intensive, and due to the high level of human involvement, human subjectivity has a greater impact on the results. Compared with the traditional way of obtaining soybean pod and grain phenotypic data, the method provided in this embodiment is much more efficient and can provide richer, more objective and accurate data. The method is used for data acquisition, specifically: in terms of efficiency, data of multiple traits such as pod length, pod width, grain length, grain width, pod color, seed coat color, etc. can be obtained simultaneously through a single photo. According to experiments, it only takes 1 second to count and record all traits of a single photo; in terms of data richness, the method can also simultaneously provide parameters such as the area, perimeter, shape index, aspect ratio, etc. of pods and grains that are difficult to obtain through conventional manual measurements, and retain the original image to facilitate further extraction of more parameters (such as re-dividing colors) or correction; in terms of data accuracy, the measurement of length and width can ensure direct correspondence to its original definition, without errors such as measurement position offset or data estimation; for color determination, the influence of a series of factors such as lighting conditions, human eye fatigue, visual deviation, etc. can be effectively removed to ensure objectivity and accuracy; in addition, there will be no data entry errors, which largely avoids waste of resources.

[0082] In detail, the size of the pods, the size of the grains, the number of pods and grains, the color of the pods, and the color of the grains are all important shapes of concern in the identification of soybean germplasm resources. The existing technology requires multiple photographs of the pods and grains to obtain the original images in order to realize the identification and determination of the size of the pods, the size of the grains, the number of pods and grains, the color of the pods, and the color of the grains. The measurement method provided in this embodiment only requires one photograph to complete and realize the determination of the data on the size of the pods, the size of the grains, the number of pods and grains, the color of the pods, and the color of the grains, that is, to solve the problems of high-throughput measurement of the size of pods and grains of soybean germplasm resources, high-throughput measurement of the number of pods and grains of soybean germplasm resources, and high-throughput classification of the color of pods and grains of soybean germplasm resources.

[0083] The following is a detailed description of each step:

[0084] It should be noted that the target measurement objects are pods and seeds of normal and fully mature plants. In this embodiment, multiple varieties are selected for experimental measurement at the same time.

[0085] For further information, please refer to Figure 2-Figure 3 , the obtaining of the corrected image of the target object is specifically as follows:

[0086] Arrange the pods and seeds vertically side by side on a white paper background with a scale grid;

[0087] Place the color chart on the side with the pods and kernels;

[0088] Shoot and obtain the original image;

[0089] Perform noise reduction on the original image based on the median filter method;

[0090] Perform color correction on the denoised image based on the colorimetric chart.

[0091] It should be noted that when obtaining the original image, when placing the experimental samples, the pods and seeds need to be placed directly below the camera, and a position with good light should be selected, and the automatic mode should be used for shooting. Among them, the orientation of the pods and seeds does not affect the identification of morphological parameters, so the placement angle does not need to be strictly consistent, but it should be laid in a single layer without overlapping each other. In order to complete the size correction later, it is necessary to obtain the original image on a white paper background with a scale grid. The scale grid preferably uses 1cm and 2cm squares. After printing on white paper, it is placed on the background board and photographed together under the camera. In order to complete the color correction later, it is necessary to set the colorimetric card when obtaining the original image and shoot it into the image together. In order to standardize the color, a 24-grid color correction plate is placed in the camera field of view during shooting to eliminate the error of color recognition caused by light at different times; if the shooting height of the camera from the experimental sample is constant, there is no need to place the colorimetric card when shooting again after the resolution is calibrated. The color correction plate is the colorimetric card.

[0092] It should be noted that the color traits of the soybean phenotypic traits involved in the experimental samples selected in this embodiment mainly include pod color (gray-brown, yellow-brown, brown, dark brown, black) and seed coat color (yellow, green, black, brown, two-color). The color correction plate is mainly used to restore the standard color in the picture. A 24-color correction plate is selected, and other correction plates such as three primary colors can also be used.

[0093] It is worth mentioning that please refer to Figure 3 , is the process of denoising and correcting the original image. The corrected image is used to complete the next step. Figure 3In the figure, (a) is the original image, (b) is the image after denoising, and (c) is the image after color correction.

[0094] For further information, please refer to Figure 4 The extraction of pod contour image parameters and seed contour image parameters based on the corrected image specifically includes:

[0095] Convert the image from RGB color space to HSV color space and calculate the maximum and minimum values ​​of hue, saturation, and brightness of the image;

[0096] Set thresholds;

[0097] The image is divided into black and white parts and the noise is removed to obtain a binary image;

[0098] Perform morphological opening operation on the binary image to divide the adhered kernel pixels into independent parts and mark them as seed points;

[0099] Based on the watershed segmentation algorithm, each pod and each grain is segmented into separate patches;

[0100] Based on the edge detection method, the outer boundary of each pod and each grain is extracted to form a frame graphic;

[0101] Based on the segmentation results of the watershed segmentation algorithm, the area and perimeter of each segmented patch are counted, and the trait index S is calculated:

[0102]

[0103] Among them: S>1.1 is pod, S≤1.1 is grain.

[0104] It is worth noting that after the original image is denoised and corrected, the pod outline image parameters and the seed outline image parameters are extracted: the first step is to extract the pod and seed pixels and remove the background noise, specifically converting the image from the RGB color space to the HSV color space. Figure 4 (b) is the image of the experimental sample in the RGB color space after preliminary cropping. Figure 4 (c) in the figure is the image of the experimental sample in the HSV color space. Next, the maximum and minimum values ​​of hue, saturation, and brightness are calculated, and then thresholds are set based on the maximum and minimum values ​​of hue, saturation, and brightness and the specific conditions of the experimental sample. Next, the corrected image is segmented into black and white parts and the noise is removed to obtain the initial segmentation result of binarization, that is, the binary image, and the image with the background other than the pods and seeds removed is obtained. Figure 4(d) in the figure is the mask image initially extracted from the experimental sample based on the HSV color space. The second step is to segment individual pods and grains. The reason is that in binary images, since grains are usually placed in piles, several grains may form the same patch, which affects the counting and size measurement. The segmentation process is to perform morphological opening operations on the binary image, divide the adhered grain pixels into independent parts, and mark them as seed points. Then, the watershed segmentation algorithm is applied to segment each pod and each grain into separate patches, and then the edge detection method is used to extract the outer boundaries of each pod and each grain. Figure 4 (e) is a schematic diagram of the preliminary segmentation results of the experimental samples. Figure 4 (f) is a schematic diagram of the segmentation process of the experimental sample performing the opening operation and the watershed segmentation algorithm. Figure 4 (g) is a schematic diagram of the segmentation of the experimental sample after using the edge detection method; the third step is to distinguish the pods and grains in the image. Specifically, according to the watershed segmentation results, the area and perimeter of each segmented patch are counted, and the trait index S is calculated based on the area and perimeter. The grains are generally round or elliptical, and their shape index is less than 1. The shape of the pod is relatively more irregular, and its shape index is generally greater than 1.3. In order to avoid the segmentation deviation affecting the shape, 1.1 is used as the threshold to distinguish the two. Figure 4 (h) in the figure is the result image of the segmentation of pods and grains of the experimental samples.

[0105] Furthermore, the image size correction is specifically performed as follows:

[0106] Extract features based on the scale-invariant feature transformation algorithm, match the scale standard format with the image, and determine the location of the scale grid in the image;

[0107] Determine the number of pixels NP0 occupied by the side length of the scale square in the image;

[0108] Based on the number of pixels occupied by the scale grid, get the size correction factor of the image:

[0109] a = length of scale grid side / NP0.

[0110] It should be noted that the size correction factor a is the side length a of each pixel, that is, the conversion ratio of the number of pixels to the actual length. Preferably, the number of pixels NP0 occupied by a scale square with a side length of 1 cm in the picture is counted.

[0111] Furthermore, the pod length parameter, grain length parameter, pod width parameter, and grain width parameter are calculated based on the corrected image, specifically:

[0112] Determine the minimum enclosing rectangle minAreaRect of a single pod and seed;

[0113] After determining the direction of the main axis, rotate the image so that the main axis remains vertical;

[0114] Determine the pod length prediction parameter NP based on the length of the long side of minAreaRect 荚长 and particle length prediction parameter NP 粒长 ;

[0115] Calculate the centroid of each pod and kernel and determine the middle dividing line of the pod or kernel;

[0116] The pod width prediction parameter NP is determined by the length of the line segment between the middle dividing line and the two intersection points of the pod or grain outline. 荚宽 and particle width prediction parameter NP 粒宽 ;

[0117] Based on the size correction factor a, calculate the pod length parameter L 荚长 , grain length parameter W 粒长 , pod width parameter L 荚宽 , particle width parameter W 粒宽 :

[0118] L 荚长 =a*NP 荚长

[0119] L 荚宽 =a*NP 荚宽

[0120] W 粒长 =a*NP 粒长

[0121] W 粒宽 =a*NP 粒宽 .

[0122] Please note that, please refer to Figure 5 and Figure 6 When obtaining the pod length parameters, grain length parameters, pod width parameters, and grain width parameters, it is necessary to determine the minimum bounding rectangle. The pod length parameters and grain length parameters are determined by the minimum bounding rectangle. When determining the minimum bounding rectangle, first determine the minimum bounding rectangle minAreaRect of a single pod and grain. After the principal axis direction is determined by principal component analysis, rotate the image so that the principal axis remains vertical. Finally, the length of the long side of minAreaRect is used as the pod length prediction parameter NP. 荚长 and particle length prediction parameter NP 粒长 Next, we need to determine the pod width prediction parameter NP 荚宽 and particle width prediction parameter NP 粒宽 The specific process requires calculating the centroid of each pod and seed. The straight line passing through the centroid and perpendicular to the main axis is the middle dividing line of the pod or seed. The length of the line segment between the middle dividing line and the two intersection points of the pod and seed outline is the pod width prediction parameter NP.荚宽 and particle width prediction parameter NP 粒宽 ; Finally, we need to use the long prediction parameter NP 荚长 , Grain length prediction parameter NP 粒长 , pod width prediction parameter NP 荚宽 , particle width prediction parameter NP 粒宽 Combined with the size correction factor a, it is converted into the pod length parameter L 荚长 , grain length parameter W 粒长 , pod width parameter L 荚宽 , particle width parameter W 粒宽 , pod length parameter L 荚长 , grain length parameter W 粒长 , pod width parameter L 荚宽 , particle width parameter W 粒宽 That is the actual length data.

[0123] Furthermore, the method of obtaining a pod width variation curve based on the pod width parameter and calculating the number of pod grains based on the pod width variation curve specifically includes:

[0124] From one end of the long side of the pod to the other end, determine the value of the pod width at each location to obtain a pod width change curve;

[0125] The curve is smoothed based on the Savitzky-Golay (SG) filter to remove the jagged edges of the curve and obtain the pod width adjustment curve.

[0126] The peak number of the pod width change adjustment curve is obtained, and the pod number is determined based on the peak number.

[0127] Please note that, please refer to Figure 7 , the process of determining the number of pod grains is the process of determining the number of grains contained in a single pod. First, it is necessary to count the pod width change curve. Specifically, it is necessary to count the value of the pod width at each location from one end of the long side of the pod to the other end, and then obtain the pod width change curve: In this embodiment, a two-grain pod containing two grains is taken as an example. The length of all line segments from one end of the pod to the other end will show a distribution characteristic of increasing from 0 (the line segment length at one end of the pod) to a peak value (mid-grain segment), then decreasing to a valley value (between grains), then increasing to a peak value (mid-grain segment), and then decreasing to 0 (the other end of the pod). Then it is necessary to smooth the pod width change curve obtained above to obtain its overall shape, that is, to obtain the pod width change adjustment curve. The specific process is to select a Savitzky-Golay (SG) filter, smooth the curve, remove the jagged edges of the curve caused by the hairs on the pod, and obtain the overall shape of the pod width change curve, that is, the pod width change adjustment curve. Then, the number of peaks is found based on the pod width change adjustment curve, and the number of peaks of the pod width change adjustment curve is obtained, and the number of peaks is the number of pods, wherein the number of peaks in the curve is found by searching for the local maximum value. Figure 7 Three experimental samples are shown in the figure. Through SG filtering, the false peaks caused by burrs on the pod tip are effectively removed, ensuring the accuracy of pod and grain count statistics.

[0128] Furthermore, the color is extracted based on the corrected image to obtain the pod color data and the seed color data, specifically:

[0129] Taking the color of RGB color space as reference, the average values ​​of R, G, and B within the borders of pods and seeds were extracted;

[0130] Based on the identification standards of germplasm resources, five colors of pods and seeds in germplasm resource identification were determined, and the average values ​​of R, G, and B color channels of 10 pods and seeds in each color category were calculated as standard reference values;

[0131] The pod color data and kernel color data were determined by comparing the three-channel average values ​​of each image with the set standard reference values ​​based on the Ming's distance method.

[0132] It is worth noting that when determining the pod color data and the seed color data, the accuracy of the data is ensured by eliminating the influence of factors such as ambient lighting and shadows during shooting. Figure 3 The corrected image shown is an image that eliminates the above-mentioned influences, and the acquisition of pod color data and grain color data is performed based on the corrected image. First, the color parameters are based on the commonly used RGB color space, and the average values ​​of R, G, and B within the cropped pods and grain borders are extracted respectively to represent the color level of the entire pods and grains, and the pod color data and grain color data are calculated based on this. Next, it is necessary to specify standard colors, refer to the identification standards of germplasm resources, screen the five colors of pods and grains in germplasm resource identification, and calculate the average values ​​of R, G, and B color channels of 10 pods and grains in each color category as standard reference values, as shown below. Figure 8 As shown; please refer to Fig. 9 When mature, the pods have five different colors: gray-brown, yellow-brown, brown, dark brown, and black; please refer to Fig.10 The seed coat colors of mature soybean seeds are divided into five types: yellow, green, black, brown, and two-color. Finally, color classification is required. The three-channel average of each image is compared with the set standard reference value through the Ming distance method, and it is accurately classified into different color sets to determine the pod color data and grain color data.

[0133] In detail, in the process of converting the bean pod image to RGB values, taking the pod color as an example, for all pixels inside the entire pod, the average values ​​of the three channels R, G, and B (r, g, b) are calculated to represent the color of the pod. There are five standard pod colors (gray brown, yellow brown, brown, dark brown, black). Representative pods are selected and their average RGB values ​​are calculated to represent the five standard pod colors, such as Figure 8 As shown. The seed coat color is obtained in the same way. After each pod and seed is represented by (r, g, b), it can be classified by the classifier and divided into its closest standard color class.

[0134] It should be noted that based on the determination coefficient and the root mean square error, the pod length parameters, grain length parameters, pod width parameters, grain width parameters, pod color data and grain color data were verified. Specifically, the determination coefficient R 2 The pod length parameter, grain length parameter, pod width parameter, grain width parameter, pod color data and grain color data are evaluated for accuracy using the root mean square error (RMSE). Preferably, the accuracy, precision, recall and F1 score can also be used to verify the measurement accuracy of the method. Specifically, this embodiment is verified on 172 germplasms and 714 images, and the specific accuracy analysis is as follows:

[0135] First, in the morphological feature extraction, the pod length parameter L is extracted 荚长 , grain length parameter W 粒长 , pod width parameter L 荚宽 , particle width parameter W 粒宽 By comparing the true value with the image analysis result, it can be verified that all pods and seeds in the image can be accurately identified with clear outlines, and the results of multiple tests are stable and unchanged. The pod length parameter L obtained by image analysis 荚长 , pod width parameter L 荚宽 There is no significant difference compared with the true value, p>0.05, p is a common indicator of statistical test, paired t test, p greater than 0.05 means accepting the null hypothesis, there is no significant difference between the two groups of data at the 0.05 level, its accuracy is more accurate than the existing technology, please refer to Fig.11 and Fig.12 , with the pod length parameter L 荚长 , pod width parameter L 荚宽 For example, the measurement root mean square error is only 0.48mm, the maximum error does not exceed 2mm, the root mean square error of the pod width is 1.39mm, and the maximum error does not exceed 6mm;

[0136] Next, in the process of determining the number of pods, the pod number recognition method can effectively distinguish single-, double-, and triple-pods, with a recognition accuracy of 94%, a precision of 96.7%, and a recall rate of 91.2% in an independent validation set. Please refer to Fig.13 ;

[0137] Finally, in the process of color trait classification, that is, when determining the pod color data and grain color data, the distribution results of pods and grains of different varieties are almost consistent with the results of naked eye identification, which is consistent with the color difference changes judged by the naked eye. The color classification accuracy reaches 98.61%, indicating that image analysis methods can realize the quantitative analysis of color differences between pods and grains of different varieties, and provide technical support for the rapid and accurate identification of color shape of soybean germplasm resources.

[0138] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A high-throughput measurement method for soybean germplasm resources based on image recognition, characterized in that: include: Acquire a calibrated image of the target measurement object; Based on the corrected image, the pod contour image parameters and the seed contour image parameters are extracted, and the image is size corrected; Based on the corrected image, the pod length parameters, grain length parameters, pod width parameters, and grain width parameters are calculated; Based on the pod width parameter, a pod width variation curve is obtained; Based on the pod width change curve, the number of pods is calculated; Extract color based on the corrected image to obtain pod color data and seed color data; The pod length parameters, grain length parameters, pod width parameters, grain width parameters, pod color data and grain color data were verified based on the determination coefficient and root mean square error.

2. The high-throughput measurement method according to claim 1, characterized in that: The obtaining of the calibrated image of the target object is specifically as follows: Arrange the pods and seeds vertically side by side on a white paper background with a scale grid; Place the color chart on the side with the pods and kernels; Shoot and obtain the original image; Perform noise reduction on the original image based on the median filter method; Perform color correction on the denoised image based on the colorimetric chart.

3. The high-throughput measurement method according to claim 2, characterized in that: The extracting of pod contour image parameters and seed contour image parameters based on the corrected image specifically includes: Convert the image from RGB color space to HSV color space and calculate the maximum and minimum values ​​of hue, saturation, and brightness of the image; Set thresholds; The image is divided into black and white parts and the noise is removed to obtain a binary image; Perform morphological opening operation on the binary image to divide the adhered kernel pixels into independent parts and mark them as seed points; Based on the watershed segmentation algorithm, each pod and each grain is segmented into separate patches; Based on the edge detection method, the outer boundary of each pod and each grain is extracted to form a frame graphic; Based on the segmentation results of the watershed segmentation algorithm, the area and perimeter of each segmented patch are counted, and the trait index S is calculated: Among them: S>1.1 is pod, S≤1.1 is grain.

4. The high-throughput measurement method according to claim 3, characterized in that: The image size correction is specifically as follows: Extract features based on the scale-invariant feature transformation algorithm, match the scale standard format with the image, and determine the location of the scale grid in the image; Determine the number of pixels NP0 occupied by the side length of the scale square in the image; Based on the number of pixels occupied by the scale grid, get the size correction factor of the image: a = length of scale grid side / NP0.

5. The high-throughput measurement method according to claim 4, characterized in that: The pod length parameter, grain length parameter, pod width parameter, and grain width parameter are calculated based on the corrected image, specifically: Determine the minimum enclosing rectangle minAreaRect of a single pod and seed; After determining the direction of the main axis, rotate the image so that the main axis remains vertical; Determine the pod length prediction parameter NP based on the length of the long side of minAreaRect 荚长 and particle length prediction parameter NP 粒长 ; Calculate the centroid of each pod and kernel and determine the middle dividing line of the pod or kernel; The pod width prediction parameter NP is determined by the length of the line segment between the middle dividing line and the two intersection points of the pod or grain outline. 荚宽 and particle width prediction parameter NP 粒宽 ; Based on the size correction factor a, calculate the pod length parameter L 荚长 , grain length parameter W 粒长 , pod width parameter L 荚宽 , particle width parameter W 粒宽 : L 荚长 =a*NP 荚长 L 荚宽 =a*NP 荚宽 W 粒长 =a*NP 粒长 W 粒宽 =a*NP 粒宽 。 6. The high-throughput measurement method according to claim 3, characterized in that: The method of obtaining a pod width variation curve based on the pod width parameter and calculating the number of pod grains based on the pod width variation curve specifically includes: From one end of the long side of the pod to the other end, determine the value of the pod width at each location to obtain a pod width change curve; The curve is smoothed based on the Savitzky-Golay (SG) filter to remove the jagged edges of the curve and obtain the pod width adjustment curve. The peak number of the pod width change adjustment curve is obtained, and the pod number is determined based on the peak number.

7. The high-throughput measurement method according to claim 3, characterized in that: The color extraction based on the correction image to obtain the pod color data and the seed color data is specifically as follows: Taking the color of RGB color space as reference, the average values ​​of R, G, and B within the borders of pods and seeds were extracted; Based on the identification standards of germplasm resources, five colors of pods and seeds in germplasm resource identification were determined, and the average values ​​of R, G, and B color channels of 10 pods and seeds in each color category were calculated as standard reference values; The pod color data and kernel color data were determined by comparing the three-channel average values ​​of each image with the set standard reference values ​​based on the Ming's distance method.

8. The high-throughput measurement method according to claim 1, characterized in that: The target measurements selected were pods and kernels of normal and fully mature plants.

Images