Intelligent phenotype identification method and system for soybean plants in mature period

Abstract

The invention discloses an intelligent phenotype identification method and system for soybean plants in a mature period. The method comprises the following steps: acquiring an original soybean plant image; performing image preprocessing and target labeling on the acquired image data, and establishing a pod detection and recognition model; obtaining a soybean plant image, obtaining a pod mask image by using the pod detection and recognition model, and obtaining a main stem and side stem contour image; and based on the extracted contour image, separating the main stem from the side stem, and respectively calculating the length of the main stem and the side stem to obtain the stem length of the soybean plant. A recognition device of the phenotype recognition system mainly comprises a light-proof shell, a base, a feeding mechanism, an image acquisition device, a rotary flattening mechanism, a sensor group, a light source group, a single-chip microcomputer controller, an upper computer and the like. According to the device, the problems of illumination interference, plant flattening, automatic control and the like in the phenotype data acquisition process are effectively solved; and meanwhile, the method does not need to determine coordinates of a starting point and an end point in the process of obtaining the stem length of the soybean plant, so that the calculation amount is remarkably reduced, the operation time is shortened, and double advantages of high efficiency and accuracy can be brought to related researches.

Description

Technical Field

[0001] The present invention relates to the technical field of plant phenotype detection, and in particular to a method and system for intelligently identifying the phenotype of mature soybean plants. Background Art

[0002] Soybean is widely cultivated worldwide due to its rich protein and oil content. The high protein and oil content of soybean seeds makes it an important source of human food, animal feed and industrial products. Improving the yield and quality of soybean is the core goal of breeding work. However, due to the complex structure of soybean plants, research on extracting phenotypic parameters and analyzing growth conditions based on plant image models is still relatively limited. At present, traditional manual measurement methods are still the most commonly used means, but this method is destructive, inefficient and costly. Therefore, the development of automated phenotypic characteristic recording technology can quickly and cost-effectively collect genetic diversity data of soybeans, thereby more effectively selecting excellent traits in breeding programs.

[0003] Reliable and accurate estimation of soybean maturity phenotypes is not only crucial for predicting yields and reducing losses, but also the key to evaluating the quality of crop germplasm resources. Therefore, how to accurately and quickly identify the phenotypes of mature soybean plants has become a key issue in modern agricultural scientific research and technological applications.

[0004] With the development of advanced technologies such as computer vision, machine learning, and artificial intelligence, plant phenotyping technology based on image recognition has gradually emerged. However, the existing plant phenotyping recognition technology still has certain technical bottlenecks in dealing with complex phenotypic characteristics of soybean maturity, such as pod type, main and side stem length, and side stem number, and cannot fully meet the actual needs of soybean research and production. In general, the existing related technologies still have certain challenges in extracting specific morphological features of mature soybean plants and stable recognition in complex environments.

[0005] Therefore, designing and developing a method and device that can accurately and efficiently identify the phenotype of mature soybean plants has important theoretical value and practical significance for realizing accurate monitoring of the soybean planting process, improving soybean breeding efficiency, and ensuring the sustainable development of the soybean industry. The present invention aims to provide such a solution to make up for the shortcomings of the existing technology in automatic identification of soybean maturity phenotypes and promote the progress and development of smart agriculture. Summary of the invention

[0006] In view of the above problems that the extraction of phenotypic parameters in the traditional soybean seed testing process is difficult, consumes a lot of human resources and is time-consuming, the present invention provides a method and system for intelligent phenotypic identification of mature soybean plants, which can efficiently collect phenotypic information of mature soybean plants, and innovatively adopts a rotary flattening mechanism to reduce occlusion of soybean plants during collection and obtain more complete phenotypic data. The application of this method and device will significantly improve the efficiency of obtaining phenotypic information and simplify the seed testing process.

[0007] The technical solution adopted by the present invention to solve the technical problem is:

[0008] The phenotypic intelligent identification method of mature soybean plants includes

[0009] Collecting original soybean plant images;

[0010] Perform image preprocessing and target labeling on the collected image data, and establish a pod detection and recognition model;

[0011] Acquire a soybean plant image, use a pod detection and recognition model to obtain a pod mask image, and apply the pod mask image to the soybean plant image to obtain the main stem and side stem contour images;

[0012] Based on the extracted contour image, the main stem and the side stems were separated, and their lengths were calculated to obtain the stem length of the soybean plant.

[0013] As an improvement to the above technical solution, in order to establish a pod detection and recognition model:

[0014] Mark the pods and give them labels with the number of seeds in them;

[0015] The target detection model is selected as the basic model. The model training process includes steps such as data preprocessing, feature extraction, network training and parameter optimization to improve the accuracy of the model, perform model verification and evaluation, and adjust the training parameters and model structure until a satisfactory recognition effect is achieved.

[0016] As an improvement of the above technical solution, to obtain the soybean plant image:

[0017] The original soybean plant image is obtained through image acquisition, image correction is performed, and the ROI area is extracted to represent the useful camera shooting area;

[0018] The preprocessed RGB image was converted into an HSV format image, and the soybean plant image was extracted by setting the threshold.

[0019] As an improvement of the above technical solution, in order to extract the soybean plant image:

[0020] Analyze the images of three channels of HSV image;

[0021] The S channel image is processed by Gaussian filtering and converted into a binary image;

[0022] Through contour detection, the contour of the plant area is identified. In order to exclude pseudo areas with smaller areas, an area threshold is set, and only contours greater than or equal to the threshold are retained. Finally, a binary image of the soybean plant emphasizing the plant area in the ROI area is obtained.

[0023] As an improvement of the above technical solution, in order to separate the main stem from the side stems:

[0024] Applying a convex hull algorithm to the contour image to generate a convex hull shape surrounding the contour;

[0025] Performing convex defect detection on the convex hull of the contour image;

[0026] Connect the detected convex defects with straight lines and visualize the results through binary images of the main stem and side stems;

[0027] The main stem and the side stem are separated based on the binary images of the main stem and the side stem.

[0028] As an improvement of the above technical solution, in order to separate the main stem from the side stems based on the binary images of the main stem and the side stems:

[0029] Starting from the leftmost side of the image, check column by column to find the first non-zero pixel, mark the point as A0, and its coordinates are (W0, H0);

[0030] Increase the horizontal coordinate W0 of A0 by 5 pixels to obtain a new horizontal coordinate W1=W0+5. Then use W1 as the horizontal coordinate and search from top to bottom until the first non-zero pixel point is encountered, marked as A1, and its coordinates are (W1, H1);

[0031] Take A1 as the new starting point, increase the horizontal coordinate W1 of A1 by 2 pixels in the horizontal direction, record it as W2=W1+2, and search for non-zero pixels in the vertical range from H1-3 to H1. If a zero pixel is found in this range, continue to move the horizontal coordinate to the right by 2 pixels until a non-zero pixel is found.

[0032] Once a non-zero pixel is found within this range, the coordinates of the current position are recorded (W i ,H i ), at the same vertical coordinate H i Search from top to right until the first zero pixel point is encountered (W j ,H j ), where H i =H j ;

[0033] Then the white pixels between these two points are changed to black, that is, the pixel value is set to 0. Through this processing, the main stem and the side stems in the image are separated.

[0034] As an improvement of the above technical solution, in order to calculate the length of the main stem and the side stem respectively to obtain the stem length of the soybean plant:

[0035] First, find the vertical coordinates of each column of white pixels in the image and calculate their average to determine the midpoint coordinates;

[0036] Next, the Euclidean distances between these midpoint coordinates were calculated and accumulated to obtain the overall stem length of the soybean plant.

[0037] As an improvement of the above technical solution, the algorithm for obtaining the stem length of the soybean plant is:

[0038] Let Y ij is the ordinate of the jth white pixel in the i-th column, then the ordinate midpoint M of the white pixel in the i-th column i for:

[0039]

[0040] Among them, N i is the total number of white pixels in the i-th column;

[0041] Compute the Euclidean distance between these midpoint coordinates:

[0042] Assume there are two midpoint coordinates M i =(x i ,y i ) and M i+1 =(x i+1 ,y i+1 ) are the coordinates of the midpoints of adjacent columns, then the Euclidean distance D between these two midpoints i,i+1 for:

[0043]

[0044] Among them, x i ,y i and x i+1 ,y i+1 They are the horizontal and vertical coordinate values ​​of two adjacent midpoints respectively;

[0045] The overall stem length of the soybean plant was obtained by calculating the Euclidean distance between each pair of adjacent midpoint coordinates and summing these distances.

[0046] The present invention solves the technical problem and provides a phenotypic intelligent identification system for mature soybean plants. The system comprises

[0047] A phenotypic image acquisition unit, used for acquiring original soybean plant images;

[0048] A pod detection and recognition model building unit is used to perform image preprocessing and target annotation on the collected image data to establish a pod detection and recognition model;

[0049] The phenotypic image processing unit is used to obtain a pod mask image based on the soybean plant image and the pod detection and recognition model, apply the mask image to the soybean plant image to obtain the main stem and side stem contour images, and then separate the main stem and side stems based on the extracted contour image, and calculate their lengths respectively to obtain the soybean plant stem length.

[0050] The present invention solves the technical problem and provides a phenotypic image acquisition device for mature soybean plants, comprising:

[0051] Light-proof housing, base, feeding mechanism, image acquisition module, rotary flattening mechanism, sensor group, light source group, control drive module;

[0052] The feeding mechanism comprises two parallel slide rails fixed above the base and connected to the light-proof housing;

[0053] The material platform is installed on the conveying seat of the two slide rails, and the rotary flattening mechanism is arranged behind it;

[0054] The rotary flattening mechanism includes a rotary pressing plate and a rotary driving mechanism, and the rotary driving mechanism is composed of a motor, a worm gear reducer, and a synchronous belt transmission mechanism, and is controlled to rotate slowly through a single-chip microcomputer and a motor drive to achieve a slow rotary flattening effect;

[0055] The sensor group includes two pressure sensors. The first pressure sensor is installed on the material table. When the pressure reaches the set value, the signal is transmitted to the single chip microcomputer to control the motor to rotate in the opposite direction. The second pressure sensor is installed above the rear plate of the light-proof housing. When the rotating pressure plate contacts it, the signal is transmitted to stop the motor until the next operation starts.

[0056] The image acquisition module includes two parallel slide rails installed on the top of the light-proof housing, a camera bracket embedded in the slide rails, a camera, and a host computer connected to the camera;

[0057] The control drive module is connected to the feeding mechanism, the rotary flattening mechanism and the image acquisition module respectively, and includes a control box for placing power wires, a single-chip microcomputer, a motor, a worm gear reducer and a synchronous belt transmission mechanism. The module is used to drive the feeding mechanism to drive the soybean plants to be fed in and out, and when the feeding mechanism moves to a suitable position, control the rotary flattening mechanism to slowly flatten and the movement of the two parallel slide rails on the top to move the camera to a suitable position;

[0058] The host computer synchronously controls the camera to take pictures, and uses the phenotypic intelligent recognition method and system of mature soybean plants to perform real-time recognition and save the recognition results.

[0059] As an improvement of the above technical solution, in order to enhance the light-shielding effect, the inner wall of the light-shielding shell is coated with a light-absorbing coating.

[0060] As an improvement of the above technical solution, in order to facilitate the placement of the soybean plants to be tested, a switch door is provided on the light-proof housing.

[0061] As an improvement of the above technical solution, in order to ensure that the light can act evenly on the plants, a light source group is also provided in the light-proof shell, and the light source group is composed of two groups of LED lights: the first group is an LED light strip embedded in the support rod of the light-proof shell, and the second group is four LED lights installed obliquely on the top.

[0062] As an improvement of the above technical solution, in order to provide a better shooting background, an LED backlight panel is installed inside the material platform.

[0063] As an improvement of the above technical solution, the slide rails are equipped with a driving mechanism, specifically a ball screw pair, to ensure stable transmission and high movement accuracy.

[0064] As an improvement of the above technical solution, the base is hollow inside and is used to place the control drive module.

[0065] The beneficial effects brought by the present invention are:

[0066] The present invention aims to provide a phenotypic intelligent identification method and system for mature soybean plants, which can intelligently identify the type of pods, the number of pods, the length of the main and side stems, and the number of side stems, taking into account both intelligence and practicality, and providing strong technical support for soybean planting and management.

[0067] In the phenotypic identification device, the present invention can collect phenotypic information of mature soybean plants, and innovatively adopts a rotating flattening mechanism to flatten them, thereby reducing obstructions and obtaining relatively complete soybean phenotypic information. Through such a carefully designed workflow, the problems of light interference, plant flattening and automatic control in the process of phenotypic data collection are effectively solved. The device can provide strong technical support for soybean phenotypic research and crop precision management.

[0068] In the phenotypic intelligent identification method and system, the present invention realizes accurate detection of soybean pods by applying target detection and instance segmentation algorithms, so as to more accurately identify soybean plants. At the same time, a more effective stem length measurement method is adopted: first, the main stem and the side stem are processed separately and their lengths are measured respectively. The specific steps are: in each image, the vertical coordinates of each column of white pixels are found and their average values ​​are calculated to determine the midpoint coordinates. Then, the Euclidean distance between these midpoint coordinates is calculated to obtain the overall length of the soybean plant. This method does not need to predetermine the starting and end point coordinates, which greatly reduces the amount of calculation, shortens the running time, and significantly improves the efficiency and accuracy of the research. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

[0070] Figure 1 This is a schematic diagram of the phenotypic intelligent identification method of Example 1;

[0071] Figure 2 This is a schematic diagram of the structure of the phenotypic intelligent identification system of Example 2;

[0072] Figure 3 It is a schematic diagram of the structure of the phenotypic image acquisition device in Examples 3 and 4;

[0073] Figure 4 is a structural schematic diagram of the rotary flattening mechanism;

[0074] Figure 5 for Figure 4 Middle A is a partial enlarged schematic diagram;

[0075] Figure 6 is a schematic diagram of the light-proof housing structure;

[0076] Figure 7 A schematic top view of the structure of the phenotypic image acquisition device;

[0077] Figure 8 It is a simplified side view of the structure of the phenotypic image acquisition device. DETAILED DESCRIPTION

[0078] The following will be combined with the embodiments of the present invention and the accompanying drawings 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.

[0079] Example 1

[0080] This embodiment describes in detail a method for intelligently identifying the phenotype of mature soybean plants, referring to Figure 1 The method specifically includes

[0081] Collecting original soybean plant images;

[0082] Perform image preprocessing and target labeling on the collected image data, and establish a pod detection and recognition model;

[0083] Acquire a soybean plant image, use a pod detection and recognition model to obtain a pod mask image, and apply the pod mask image to the soybean plant image to obtain the main stem and side stem contour images;

[0084] Based on the extracted contour image, the main stem and the side stems were separated, and their lengths were calculated to obtain the stem length of the soybean plant.

[0085] The specific implementation methods of each step will be described in detail below.

[0086] To collect the original soybean plant image as model data, the phenotypic image acquisition device as described in Example 3 can be used, which does not constitute a limitation to this step.

[0087] In order to establish a pod detection and recognition model, the model data must first be annotated and data prepared.

[0088] This embodiment uses a labeling tool to label the bean pods, covers all the bean pods through semantic segmentation, and assigns a label with the number of pods; then, image correction is performed.

[0089] Subsequently, the target detection model is selected as the basic model. The model training process includes steps such as data preprocessing, feature extraction, network training and parameter optimization to improve the accuracy of the model, conduct model verification and evaluation, and adjust the training parameters and model structure until a satisfactory recognition effect is achieved, thereby constructing a pod detection and recognition model.

[0090] Based on this model, soybean plant pods can be identified and counted, that is, soybean pods can be detected by applying object detection and instance segmentation algorithms. To solve the problem of pod overlap, when taking images of soybean plants, soybean pods can be separated to minimize the degree of overlap.

[0091] When faced with soybean plants for phenotyping:

[0092] The original soybean plant image is obtained through image acquisition;

[0093] Perform image correction and extract the ROI area to represent the useful camera shooting area;

[0094] Convert the preprocessed RGB image into an HSV format image;

[0095] The images of the three channels of the HSV image were analyzed; the S channel image was significantly different from the background, so the S channel image was used to extract the soybean plants by setting a threshold. Specifically, the S channel image was processed with a Gaussian filter and converted into a binary image; the contour of the plant area was identified through contour detection, and an area threshold was set to exclude pseudo areas with smaller areas, and only contours greater than or equal to the threshold were retained. Finally, a binary image of the soybean plant was obtained that emphasized the plant area in the ROI area, while eliminating the influence of other irrelevant areas.

[0096] After obtaining the soybean plant image to be identified, the pods are semantically segmented by the pod detection and recognition model to obtain the mask of the soybean pods. The mask is applied to the soybean plant image to remove the pods and obtain the contour images of the main stem and side stems of the soybean plant.

[0097] After the contour image is obtained, in order to better capture the overall shape of the plant, a convex hull algorithm is applied to the contour image to generate a convex hull shape around the contour, and the convex hull provides global information about the shape of the plant;

[0098] Performing convex defect detection on the convex hull of the contour image; convex defects usually occur at bifurcation points of plants, which will cause gaps between the convex hull and the contour. By detecting these convex defects, the bifurcation points of the plants can be located;

[0099] The detected convex defects are connected by straight lines and the results are visualized through binary images of the main stem and lateral stems; this process provides an intuitive understanding of the location of bifurcation points in the plant structure by tracing the connecting lines in the image.

[0100] The main stem and the side stems can be separated based on the binary image of the main stem and the side stems:

[0101] Starting from the leftmost side of the image, check column by column to find the first non-zero pixel, mark the point as A0, and its coordinates are (W0, H0);

[0102] Increase the horizontal coordinate W0 of A0 by 5 pixels to obtain a new horizontal coordinate W1=W0+5. Then use W1 as the horizontal coordinate and search from top to bottom until the first non-zero pixel point is encountered, marked as A1, and its coordinates are (W1, H1);

[0103] Take A1 as the new starting point, increase the horizontal coordinate W1 of A1 by 2 pixels in the horizontal direction, record it as W2=W1+2, and search for non-zero pixels in the vertical range from H1-3 to H1. If a zero pixel is found in this range, continue to move the horizontal coordinate to the right by 2 pixels until a non-zero pixel is found.

[0104] Once a non-zero pixel is found within this range, the coordinates of the current position are recorded (W i ,H i ), at the same vertical coordinate H i Search from top to right until the first zero pixel point is encountered (W j ,H j ), where H i =H j ;

[0105] Then the white pixels between these two points are changed to black, that is, the pixel value is set to 0. Through this processing, the main stem and the side stems in the image are separated.

[0106] This embodiment uses the following algorithm to calculate the length of the main stem and the side stem respectively to obtain the stem length of the soybean plant.

[0107] First, find the vertical coordinates of each column of white pixels in the image and calculate their average to determine the midpoint coordinates;

[0108] Next, the Euclidean distances between these midpoint coordinates were calculated and accumulated to obtain the overall stem length of the soybean plant.

[0109] As an improvement of the above technical solution, the algorithm for obtaining the stem length of the soybean plant is:

[0110] Let Y ij is the ordinate of the jth white pixel in the i-th column, then the ordinate midpoint M of the white pixel in the i-th column i for:

[0111]

[0112] Among them, N i is the total number of white pixels in the i-th column;

[0113] Compute the Euclidean distance between these midpoint coordinates:

[0114] Assume there are two midpoint coordinates M i =(x i ,y i ) and M i+1 =(x i+1 ,y i+1 ) are the coordinates of the midpoints of adjacent columns, then the Euclidean distance D between these two midpoints i,i+1 for:

[0115]

[0116] Among them, x i ,y i and x i+1 ,y i+1They are the horizontal and vertical coordinate values ​​of two adjacent midpoints respectively;

[0117] By calculating the Euclidean distance between each pair of adjacent midpoint coordinates and summing these distances, the overall stem length of the soybean plant is obtained. This algorithm does not need to determine the starting and ending point coordinates, which can significantly reduce the amount of calculation and improve efficiency.

[0118] Example 2

[0119] Reference Figure 2 , a phenotypic intelligent identification system for mature soybean plants, the system comprising

[0120] A phenotypic image acquisition unit, used for acquiring original soybean plant images;

[0121] A pod detection and recognition model building unit is used to perform image preprocessing and target labeling on the collected image data to establish a pod detection and recognition model;

[0122] The phenotypic image processing unit is used to obtain a pod mask image based on the soybean plant image and the pod detection and recognition model, apply the mask image to the soybean plant image to obtain the main stem and side stem contour images, and then separate the main stem and side stems based on the extracted contour image, and calculate their lengths respectively to obtain the soybean plant stem length.

[0123] The specific implementation means of each unit can refer to Example 1.

[0124] Example 3

[0125] Reference Figure 3 , a phenotypic image acquisition device for mature soybean plants, comprising

[0126] A light-proof housing 8, a base 7, a feeding mechanism, an image acquisition module, a rotary flattening mechanism, a sensor group, a light source group, and a control driving module; the base 7 is hollow inside and is used to place the control driving module.

[0127] The feeding mechanism includes two slide rails 10 fixed above the base 7 and connected to the light-proof housing 8;

[0128] The material platform 11 is installed on the conveying seat of the two slide rails 10, and the rotary flattening mechanism is arranged behind it;

[0129] The rotary flattening mechanism includes a rotary pressing plate 6 and a rotary driving mechanism, such as Figure 4 , 5 As shown, the rotation drive mechanism is composed of a motor 14, a worm gear reducer, and a synchronous belt transmission mechanism, and is controlled to rotate slowly through a single chip microcomputer 20 and a motor driver 19 to achieve a slow rotation flattening effect;

[0130] The sensor group includes two pressure sensors. The first pressure sensor is installed on the material table 11. When the pressure reaches the set value, the signal is transmitted to the single chip microcomputer 20 to control the motor 14 to rotate in the opposite direction. The second pressure sensor is installed above the rear plate of the light-proof housing 8. When the rotating pressure plate 6 contacts it, the signal is transmitted to stop the motor 14 until the next operation starts.

[0131] The image acquisition module includes two parallel slide rails 3 installed on the top of the light-proof housing 8, a camera bracket 2 embedded in the slide rails, a camera 1, and a host computer 13 connected to the camera 1;

[0132] The control drive module is connected to the feeding mechanism, the rotary flattening mechanism and the image acquisition module respectively, and includes a control box for placing power wires, a single-chip microcomputer 20, a motor 14, a worm gear reducer and a synchronous belt transmission mechanism. The module is used to drive the feeding mechanism to drive the soybean plants to be fed in and out, and when the feeding mechanism moves to a suitable position, control the rotary flattening mechanism to slowly flatten and the movement of the two parallel slide rails 3 on the top to move the camera 1 to a suitable position;

[0133] The host computer 13 synchronously controls the camera 1 to take pictures, and performs real-time recognition and saves the recognition results through the phenotypic intelligent recognition method and system of mature soybean plants.

[0134] In order to improve the light-shielding effect, the inner wall of the light-shielding housing 8 is coated with a light-absorbing coating.

[0135] In order to facilitate the placement of the soybean plants to be tested, a switch door is provided on the light-proof housing 8 .

[0136] In order to ensure that the light can act evenly on the plants, a light source group is also provided in the light-proof shell 8, and the light source group is composed of two groups of LED lights: the first group is an LED light strip embedded in the support rod of the light-proof shell 8, and the second group is four LED lights 4 installed obliquely on the top.

[0137] In order to provide a better shooting background, an LED backlight panel 12 is installed inside the material platform 11.

[0138] In this embodiment, the slide rails 3 / 10 are equipped with a driving mechanism, specifically a ball screw pair, to ensure stable transmission and high movement accuracy.

[0139] Example 4

[0140] like Figures 3 to 8 As shown, a plant phenotype image acquisition device of the present embodiment includes an RGB camera 1, a camera bracket 2, a parallel slide rail 3, a rotating pressure plate 6, a base 7, a light-proof shell 8, a slide rail 10, a crop table 11, an LED backlight panel 12, a host computer 13, a light source group, a sensor group, and a control drive module.

[0141] The device of this embodiment aims to provide an accurate and efficient tool for obtaining plant phenotypic information, especially suitable for collecting morphological characteristics of mature soybean plants. The device mainly includes the following key technical structures:

[0142] like Figure 3 As shown, the image acquisition module of the device includes an RGB camera 1, a camera bracket 2 embedded in two parallel slide rails 3 on the top, and a host computer 13 connected to the RGB camera 1; after the camera 1 acquires the picture, the picture is transmitted to the host computer 13, and the image is detected in real time by the phenotypic intelligent recognition method recorded in Example 1 and the phenotypic intelligent recognition system recorded in Example 2 to obtain the type of pods, the number of pods, the length of the main lateral stems and the number of lateral stems, and the results are saved.

[0143] Furthermore, if Figure 3 and Figure 7 As shown, the feeding mechanism includes two slide rails 10 installed on the base 7, a crop table 11 embedded in the slide rails 10, and an LED backlight panel 12 located in the crop table 11. When the soybean plants are transported, the front door 801 is first opened, and the crop table 11 is driven by a ball screw to move in a straight line along the slide rails 10. When it reaches the limit block 28 in front of the slide rail 10, it stops moving. After the soybean plants are placed, it starts to move in the opposite direction. When it reaches the limit block 29 behind the slide rail 10, it stops moving. The upper computer 13 is ready to control the rotary flattening mechanism to start the flattening operation.

[0144] Furthermore, if Figure 3 , Figure 4 , Figure 5 As shown, the rotary flattening mechanism includes a rotating pressing plate 6, a rotating shaft bracket 23, a motor 14, a synchronous pulley 15, a synchronous belt 16, a coupling 17, a worm gear reducer 18, a motor drive 19, an STM32 single-chip microcomputer 20, and a power supply 22. When operating, it first needs to determine that the material platform 11 has reached the working position, and then the host computer 13 issues a command, or through Figure 8 The start button 26 shown is manually activated, and the rotating plate 6 starts to rotate slowly. When the rotating plate 6 contacts the first pressure sensor 9 on the material platform 11 and reaches a preset pressure value, the single chip microcomputer 20 controls the motor 14 to rotate in the opposite direction; when the rotating plate 6 contacts the second pressure sensor 5 on the rear plate, the motor 14 stops operating and waits for the next operation to start.

[0145] Furthermore, the light source group is provided with two groups of LED lamps, the first group is four LED light strips embedded on the support rods 804 of the light-proof housing 8, and the second group is four LED lamps 4 placed obliquely on the top.

[0146] Furthermore, in order to ensure the lighting effect, Figure 6 The light-proof housing 8 shown includes: a front panel 801, a top boss 802, a side panel 803, a support rod 804 and a rear panel, the inner wall of which is coated with a light-absorbing material coating, which significantly improves the light-proof performance of the device, ensures that it is not affected by external light during the image acquisition process, and improves the quality and consistency of the phenotypic image.

[0147] The device of this embodiment also includes a base 7 and four casters 25 installed under the base 7. The base 7 is hollow inside and is used to place control modules such as a control box 21, a power supply 22, a motor 14, and a single-chip microcomputer 20. The design of the base 7 extending outward at the top and shortening at the bottom is to facilitate the replacement of plants and reduce the volume.

[0148] The specific workflow of this embodiment is as follows:

[0149] First, when unloaded, the material table 11 is located in the working position. When there are soybean plants that need to be imaged, the front door 801 is opened first, and then the host computer 13 controls the servo drive system to move using the ball screw pair. When it reaches the limit block 28 in front of the slide rail 10, the host computer 13 issues a command to stop moving. After the plant is placed on the material table 11, the host computer 13 sends a command to the servo motor to rotate in the opposite direction, and the material table 11 starts to move in the opposite direction. When it reaches the limit block 29 at the rear of the slide rail 10, that is, the working position, it stops moving.

[0150] At this time, it is determined that the material platform 11 has reached the working position, and the rotary flattening mechanism starts to work. The command can be issued through the host computer 13 or directly by rotating the start button 26, and the rotating pressing plate 6 starts to rotate slowly. When the rotating pressing plate 6 contacts the first pressure sensor 9 located on the material platform 11 and reaches the set pressure value, the sensor transmits the generated pulse signal to the single-chip microcomputer CPU, and the single-chip microcomputer 20 issues a command to control the motor 14 to start reversing. When it contacts the second pressure sensor 5 on the rear plate, the sensor generates a pulse signal and sends it to the single-chip microcomputer 20, which controls the motor 14 to stop rotating until the next operation.

[0151] After the image acquisition module starts working, the front door 801 is first closed to prevent light leakage, and the LED light source group is turned on at the same time. Then, the host computer 13 controls the camera 1 to start, and adjusts the camera 1 to the center position through the upper parallel slide rail 3 according to the position of the soybean plant in the camera's field of view. The camera 1 then starts shooting, collects images of one side of the soybean plant, and transmits the images to the intelligent recognition system of the host computer 13 in real time, and displays and saves the detection results in real time.

[0152] After the acquisition and recognition are completed, the front door 801 is opened, and the host computer 13 sends a command, and the single-chip microcomputer 20 controls the ball screw mechanism to send the material table 11 out, and then flips the plant to re-acquire the image of the other side. If the soybean plant is not obviously blocked or the blockage is small, the rotation flattening step can be omitted and the image acquisition can be performed directly. When the images of the front and back sides of a soybean plant are acquired, the above steps are repeated to perform the acquisition operation of the next soybean plant.

[0153] The mature soybean plant phenotypic image acquisition device of this embodiment effectively solves the problems of light interference, plant flattening and automatic control in the process of phenotypic data acquisition through a carefully designed workflow. It can provide strong technical support for soybean phenotypic research and precise crop management, and significantly improve the efficiency and accuracy of data acquisition.

[0154] It should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments or to make equivalent substitutions for some of the technical features therein. 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 method for intelligently identifying the phenotype of mature soybean plants, characterized in that: include Collecting original soybean plant images; Perform image preprocessing and target labeling on the collected image data, and establish a pod detection and recognition model; Acquire a soybean plant image, use a pod detection and recognition model to obtain a pod mask image, and apply the pod mask image to the soybean plant image to obtain the main stem and side stem contour images; Based on the extracted contour image, the main stem and the side stems were separated, and their lengths were calculated to obtain the stem length of the soybean plant.

2. The method for intelligently identifying the phenotype of mature soybean plants according to claim 1, characterized in that: To build a pod detection and recognition model: Mark the pods and give them labels with the number of seeds in them; The target detection model is selected as the basic model. The model training process includes steps such as data preprocessing, feature extraction, network training and parameter optimization to improve the accuracy of the model, perform model verification and evaluation, and adjust the training parameters and model structure until a satisfactory recognition effect is achieved.

3. The method for intelligently identifying the phenotype of mature soybean plants according to claim 1, characterized in that: To obtain an image of a soybean plant: The original soybean plant image is obtained through image acquisition, image correction is performed, and the ROI area is extracted to represent the useful camera shooting area; The preprocessed RGB image was converted into an HSV format image, and the soybean plant image was extracted by setting the threshold.

4. The method for intelligently identifying the phenotype of mature soybean plants according to claim 3, characterized in that: To extract the soybean plant image: Analyze the images of three channels of HSV image; The S channel image is processed by Gaussian filtering and converted into a binary image; Through contour detection, the contour of the plant area is identified. In order to exclude pseudo areas with smaller areas, an area threshold is set, and only contours greater than or equal to the threshold are retained. Finally, a binary image of the soybean plant emphasizing the plant area in the ROI area is obtained.

5. The method for intelligently identifying the phenotype of mature soybean plants according to claim 1, characterized in that: To separate the main stem from the side stems: Applying a convex hull algorithm to the contour image to generate a convex hull shape surrounding the contour; Performing convex defect detection on the convex hull of the contour image; Connect the detected convex defects with straight lines and visualize the results through binary images of the main stem and side stems; The main stem and the side stem are separated based on the binary images of the main stem and the side stem.

6. The method for intelligently identifying the phenotype of mature soybean plants according to claim 5, characterized in that: To separate the main stem from the side stems based on the binary images of the main stem and the side stems: Starting from the leftmost side of the image, check column by column to find the first non-zero pixel, mark the point as A0, and its coordinates are (W0, H0); Increase the horizontal coordinate W0 of A0 by 5 pixels to obtain a new horizontal coordinate W1=W0+5. Then use W1 as the horizontal coordinate and search from top to bottom until the first non-zero pixel point is encountered, marked as A1, and its coordinates are (W1, H1); Take A1 as the new starting point, increase the horizontal coordinate W1 of A1 by 2 pixels in the horizontal direction, record it as W2=W1+2, and search for non-zero pixels in the vertical range from H1-3 to H1. If a zero pixel is found in this range, continue to move the horizontal coordinate to the right by 2 pixels until a non-zero pixel is found. Once a non-zero pixel is found within this range, the coordinates of the current position are recorded (W i ,H i ), at the same vertical coordinate H i Search from top to right until the first zero pixel point is encountered (W j ,H j ), where H i =H j ; Then the white pixels between these two points are changed to black, that is, the pixel value is set to 0. Through this processing, the main stem and the side stems in the image are separated.

7. The method for intelligently identifying the phenotype of mature soybean plants according to claim 6, characterized in that: To calculate the length of the main stem and side stems separately to obtain the stem length of soybean plants: First, find the vertical coordinates of each column of white pixels in the image and calculate their average to determine the midpoint coordinates; Next, the Euclidean distances between these midpoint coordinates were calculated and accumulated to obtain the overall stem length of the soybean plant. As an improvement of the above technical solution, the algorithm for obtaining the stem length of the soybean plant is: Let Y ij is the ordinate of the jth white pixel in the i-th column, then the ordinate midpoint M of the white pixel in the i-th column i for: Among them, N i is the total number of white pixels in the i-th column; Compute the Euclidean distance between these midpoint coordinates: Assume there are two midpoint coordinates M i =(x i ,y i ) and M i+1 =(x i+1 ,y i+1 ) are the coordinates of the midpoints of adjacent columns, then the Euclidean distance D between these two midpoints i,i+1 for: Among them, x i ,y i and x i+1 ,y i+1 They are the horizontal and vertical coordinate values ​​of two adjacent midpoints respectively; The overall stem length of the soybean plant was obtained by calculating the Euclidean distance between each pair of adjacent midpoint coordinates and summing these distances.

8. A phenotypic intelligent recognition system for mature soybean plants, characterized in that: The system comprises A phenotypic image acquisition unit, used for acquiring original soybean plant images; A pod detection and recognition model building unit is used to perform image preprocessing and target annotation on the collected image data to establish a pod detection and recognition model; The phenotypic image processing unit is used to obtain a pod mask image based on the soybean plant image and the pod detection and recognition model, apply the mask image to the soybean plant image to obtain the main stem and side stem contour images, and then separate the main stem and side stems based on the extracted contour image, and calculate their lengths respectively to obtain the soybean plant stem length.

9. A phenotypic image acquisition device for mature soybean plants, characterized in that: include Light-proof housing, base, feeding mechanism, image acquisition module, rotary flattening mechanism, sensor group, light source group, control drive module; in The feeding mechanism comprises two slide rails fixed above the base and connected to the light-proof housing; The material platform is installed on the two slide rails, and the rotary flattening mechanism is arranged behind it; The rotary flattening mechanism includes a rotary pressing plate and a rotary driving mechanism, which is controlled to rotate slowly by a single chip microcomputer and a motor to achieve a slow rotary flattening effect; The sensor group includes two pressure sensors. The first pressure sensor is installed on the material table. When the pressure reaches the set value, the signal is transmitted to the single chip microcomputer to control the motor to rotate in the opposite direction. The second pressure sensor is installed above the rear plate of the light-proof housing. When the rotating pressure plate contacts it, the signal is transmitted to stop the motor until the next operation starts. The image acquisition module includes two slide rails installed on the top of the light-proof housing, a camera bracket embedded in the two slide rails, a camera, and a host computer connected to the camera; The control driving module is respectively connected with the feeding mechanism, the rotating flattening mechanism and the image acquisition module, and is used to drive the feeding mechanism to drive the soybean plants to be fed in and out, and when the feeding mechanism moves to a suitable position, controls the rotating flattening mechanism to slowly flatten and the movement of the two top slide rails to move the camera to a suitable position; The host computer is used to control the camera to take pictures, and to perform real-time recognition and save the recognition results through the phenotypic intelligent recognition method of mature soybean plants described in any one of claims 1 to 7 and the phenotypic intelligent recognition system of mature soybean plants described in claim 8.

10. The phenotypic image acquisition device of mature soybean plants according to claim 9, characterized in that: In order to improve the light-shielding effect, the inner wall of the light-shielding shell is coated with a light-absorbing coating; In order to facilitate the placement of the plants to be tested, a switch door is provided on the light-proof housing; In order to ensure that the light can act evenly on the plants, a light source group is also provided in the light-proof shell, and the light source group consists of two groups of LED lights: the first group is an LED light strip embedded in the support rod of the light-proof shell, and the second group is four LED lights installed obliquely on the top.

Images