Field corn plant phenotype measurement method and system

Abstract

The invention relates to a field corn plant phenotype measurement method and system, and the method comprises the steps: firstly employing an image collector to shoot an RGB image and a depth image of a field corn plant, then extracting a mask image from the depth image based on a clip module added in a YOLOv8 model, carrying out the mask processing of the RGB image, employing the YOLOv8 model to carry out the target recognition of the RGB image after the mask processing, and carrying out the measurement of the phenotype of the field corn plant. Then key point pixel coordinates output by the recognized target are determined, coordinate depths corresponding to the key point pixel coordinates are extracted from the depth image, camera coordinate system coordinates of the key points are calculated and then converted into world coordinate system coordinates, key points belonging to the same plant are determined through coordinate matching and pairing is carried out, and the target is obtained. And finally, according to the world coordinate system coordinates of the paired key points, calculating the ear height and the leaf included angle of the corn plant. According to the method, accurate analysis of the spike height and leaf included angle phenotype of the low-resolution image can be realized.

Description

Technical Field

[0001] The invention belongs to the field of agricultural bioinformatics technology, and in particular relates to a method and system for measuring the phenotype of field corn plants. Background Art

[0002] Acquiring crop phenotypic information is a key link in modern breeding technology. With the development of gene sequencing technology, the huge number of gene combinations has spawned a massive number of crop phenotypic samples. By analyzing the traits of these samples, breeders can screen out high-quality and high-yield crop varieties to meet the rapidly growing world food demand.

[0003] At present, the actual measurement method of corn plant phenotype in the agricultural field is still mainly manual, and the published literature mainly uses binocular RGBD cameras and robot vehicles to analyze phenotypes through images. This method not only uses more complex equipment, but also because the field of view of the RGBD camera is small, it is difficult to obtain full-view information from roots to ears in one picture. It requires a relatively large shooting distance or a large posture pitch angle, or more cameras to meet the viewing angle requirements. The row spacing in the field does not allow a large shooting distance, and a large posture pitch angle will seriously affect the accuracy of the depth value. The high-resolution depth map is still severely compressed in the area far from the lens. The larger the inclination angle, the more severe the compression. Therefore, it is mostly limited to potted scenes and cannot be applied to the field. Summary of the invention

[0004] The purpose of the present invention is to solve the above problems in the prior art and to provide a method and system for measuring the phenotype of field corn plants with simple operation and high precision.

[0005] To achieve the above objectives, the technical solution of the present invention is as follows:

[0006] In a first aspect, the present invention provides a method for measuring the phenotype of corn plants in a field, comprising:

[0007] The method comprises:

[0008] S1. Using an image collector to capture an RGB image and a depth image of corn plants in the field, wherein the RGB image and the depth image both include a near-view image and a far-view image;

[0009] S2, extracting a mask image from the depth image based on the clip module added in the YOLOv8 model, and applying the extracted mask image to the RGB image to extract the RGB pixels covered by the mask image as the RGB image after mask processing;

[0010] S3. Use the YOLOv8 model to perform target recognition on the masked RGB image. For the distant image, the recognition target is the corn aerial root and the segmentation target is the corn ear. For the close-up image, the recognition target is the corn leaf node.

[0011] S4, determine the key point pixel coordinates output by the identified target, wherein the key point pixel coordinates of the corn aerial root are the coordinates of the midpoint of the target frame, the key point pixel coordinates of the corn ear are the coordinates of the lowest end point of the segmented area, and the key point pixel coordinates of the corn leaf node are the coordinates of a point S on the leaf vein that is less than 50 pixels from the leaf node, the intersection point M of the leaf and the stem, and the coordinates of a point C on the stalk that is less than 50 pixels from the leaf node;

[0012] S5, extracting the coordinate depth D corresponding to the pixel coordinates of the key point from the depth image, and calculating the camera coordinate system coordinates of the key point;

[0013] S6, converting the coordinates of the key points into world coordinates according to the camera coordinate system coordinates and the posture data of the image collector, determining the key points belonging to the same plant through coordinate matching and pairing them;

[0014] S7. Calculate the ear height and leaf angle of the corn plant according to the world coordinate system coordinates of the paired key points.

[0015] The clip module is located before the backbone module of YOLOv8;

[0016] In S2, extracting a mask image from a depth image based on the clip module added in the YOLOv8 model includes:

[0017] The clip module first determines the cut-off depth range according to the shooting distance, and then extracts the pixels in the cut-off depth range from the depth image as the mask image. The cut-off depth range is [d min ,d min +0.3(d max -d min )],d max d min They are the maximum and minimum depth values ​​that can be read in the depth image.

[0018] In S5, the coordinate depth D corresponding to the pixel coordinates of the key point is extracted from the depth image and the depth confidence of the point is determined. If the depth confidence is lower than a threshold, the depth D is replaced by the depth average of multiple pixel points around the point whose confidence is higher than the threshold. If the confidence of the pixel points around the point are all lower than the threshold, the depth image is discarded.

[0019] In S7, the ear height is calculated according to the following formula:

[0020]

[0021] In the above formula, EH is ear height, are the three-dimensional coordinates of the key points of the corn ear, The three-dimensional coordinates of the key points of corn aerial roots;

[0022] The leaf angle is calculated according to the following formula:

[0023]

[0024] In the above formula, LA is the leaf angle, SM is the distance between point S and point M, MC is the distance between point M and point S, and SC is the distance between point S and point C. These are the three-dimensional coordinates of point S, point M, and point C respectively.

[0025] The image collector is a mobile phone equipped with a TOF sensor. When shooting long-range images, the distance between the image collector and the corn plants is not less than 0.75 meters; when shooting close-range images, the distance between the image collector and the corn plants is 0.2 meters; the pitch angle and roll angle ranges of the image collector are both within (-30°, 30°).

[0026] The S6 determines the key points belonging to the same plant according to the Z value and X value range of the world coordinate system coordinates of the key points, and the Z value and X value range are both less than 0.15 meters.

[0027] In a second aspect, the present invention provides a field corn plant phenotype measurement system, including an image collector, a YOLOv8 model, a key point pixel coordinate determination module, a key point camera coordinate determination module, a key point pairing module, and an ear height and leaf angle calculation module;

[0028] The image collector is used to capture RGB images and depth images of corn plants in the field, wherein the RGB images and depth images both include a close-up image and a distant image;

[0029] The YOLOv8 model is used to:

[0030] A mask image is extracted from the depth image based on the added clip module, and the extracted mask image is applied to the RGB image to extract the RGB pixels covered by the mask image as the RGB image after mask processing;

[0031] The masked RGB image is used for target recognition. For distant images, the target is the aerial root of corn and the segmentation target is the female ear of corn. For close-up images, the target is the leaf node of corn.

[0032] The key point pixel coordinate determination module is used to determine the key point pixel coordinates output by the identified target, wherein the key point pixel coordinates of the corn aerial root are the coordinates of the center point of the target frame, the key point pixel coordinates of the corn ear are the coordinates of the lowest end point of the segmented area, and the key point pixel coordinates of the corn leaf node are the coordinates of a point S on the leaf vein that is less than 50 pixels away from the leaf node, the intersection point M of the leaf and the stem, and the coordinates of a point C on the stem that is less than 50 pixels away from the leaf node;

[0033] The key point camera coordinate determination module is used to extract the coordinate depth D corresponding to the pixel coordinates of the key point from the depth image, and calculate the camera coordinate system coordinates of the key point in combination with the posture and longitude and latitude data of the image collector;

[0034] The key point pairing module is used to convert the coordinates of the key points into the world coordinate system coordinates according to the camera coordinate system coordinates and the posture data of the image collector, and determine the key points belonging to the same plant through coordinate matching and pair them;

[0035] The ear height and leaf angle calculation module is used to calculate the ear height and leaf angle of the corn plant according to the world coordinate system coordinates of the paired key points.

[0036] The clip module is located before the backbone module of YOLOv8;

[0037] The extracting of the mask image from the depth image comprises:

[0038] The clip module first determines the cut-off depth range according to the shooting distance, and then extracts the pixels in the cut-off depth range from the depth image as the mask image. The cut-off depth range is [d min ,d min +0.3(d max -d min )],d max d min They are the maximum and minimum depth values ​​that can be read in the depth image respectively;

[0039] The key point camera coordinate determination module extracts the coordinate depth D corresponding to the pixel coordinates of the key point from the depth image and determines the depth confidence of the point. If the depth confidence is lower than a threshold, the depth D is replaced by the depth average of multiple pixel points around the point whose confidence is higher than the threshold. If the confidence of the pixel points around the point are all lower than the threshold, the depth image is discarded.

[0040] The ear height is calculated according to the following formula:

[0041]

[0042] In the above formula, EH is ear height, are the three-dimensional coordinates of the key points of the corn ear, The three-dimensional coordinates of the key points of corn aerial roots;

[0043] The leaf angle is calculated according to the following formula:

[0044]

[0045] In the above formula, LA is the leaf angle, SM is the distance between point S and point M, MC is the distance between point M and point S, and SC is the distance between point S and point C. These are the three-dimensional coordinates of point S, point M, and point C respectively.

[0046] The image collector is a mobile phone equipped with a TOF sensor. When shooting long-range images, the distance between the image collector and the corn plants is not less than 0.75 meters; when shooting close-range images, the distance between the image collector and the corn plants is 0.2 meters; the pitch angle and roll angle ranges of the image collector are both within (-30°, 30°).

[0047] Compared with the prior art, the present invention has the following beneficial effects:

[0048] 1. A field corn plant phenotype measurement method of the present invention first uses an image collector to shoot an RGB image and a depth image of a field corn plant, then extracts a mask image from the depth image based on a clip module added in a YOLOv8 model, and applies the extracted mask image to the RGB image to extract RGB pixels covered by the mask image as the RGB image after mask processing, then uses the YOLOv8 model to perform target recognition on the masked RGB image, then determines the key point pixel coordinates output by the identified target, then extracts the coordinate depth D corresponding to the key point pixel coordinates from the depth image, and calculates the camera coordinate system coordinates of the key point, finally converts the coordinates of the key point into world coordinate system coordinates according to the camera coordinate system coordinates of the key point and the posture data of the image collector, determines the key points belonging to the same plant by coordinate matching and pairs them, and calculates the ear height and leaf angle of the corn plant according to the world coordinate system coordinates of the paired key points. On the one hand, this method can be used for depth maps with very low resolution (such as the depth map of a mobile phone, with a resolution of only 256*192), and can simultaneously analyze phenotypes such as ear height and leaf angle. The analysis effect is accurate, and the relative error is less than 5%. On the other hand, the method has a simple process, does not require the establishment of a point cloud, has low requirements for computing resources, can perform edge computing, and can be directly deployed on mobile terminals (such as mobile phones) for real-time measurement.

[0049] 2. A field corn plant phenotype measurement method of the present invention extracts a coordinate depth D corresponding to the pixel coordinates of the key point from a depth image and determines the depth confidence of the point. If the depth confidence is lower than a threshold, the depth D is replaced by the depth average of multiple pixel points around the point whose confidence is higher than the threshold. If the confidences of the pixel points around the point are all lower than the threshold, the depth image is discarded. This strategy can effectively improve the accuracy of the phenotypic measurement results. In addition, the method is based on the clip module added in the YOLOv8 model, extracts a mask image from the depth image according to the shooting distance, eliminates background interference, and makes the recognition of key points very accurate. The recognition accuracy map50 exceeds 0.93, thereby further ensuring the accuracy of the phenotypic measurement results. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 This is a flowchart of the method described in Example 1.

[0051] Figure 2 Schematic diagram of the structure of the YOLOv8 model in Example 1.

[0052] Figure 3 This is the recognition result of the close-up image in Example 1.

[0053] Figure 4 This is the recognition result of the distant image in Example 1.

[0054] Figure 5 This is a structural diagram of the system described in Example 2. DETAILED DESCRIPTION

[0055] The present invention is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0056] The present invention proposes a method for measuring the phenotype of corn plants in the field. The method is based on a monocular camera and a TOF sensor. Any mobile phone equipped with a TOF sensor is applicable and has very low requirements on the resolution of the depth camera. In addition, the correlation index R of the ear height measurement result of this method is 2 is 0.81, RMSE is 4 cm, MAE is 3 cm, and the correlation index R 2 The average error is 0.798, the RMSE is 6.8°, and the MAE is 4.4°.

[0057] Embodiment 1:

[0058] A method for measuring phenotype of corn plants in the field, such as Figure 1 As shown, the specific steps are as follows:

[0059] 1. Use a mobile phone equipped with a TOF sensor to collect data on corn plants in the field. The main camera of the mobile phone collects RGB images, including close-up and long-range images. The ToF sensor collects the required depth images, including close-up and long-range images. When taking images, the roll angle of the mobile phone relative to the X-axis, the pitch angle of the mobile phone relative to the Y-axis, the rotation angle yaw relative to the Z-axis, and the current latitude and longitude coordinates and elevation of the mobile phone recorded by the RTK device are recorded. When taking long-range images, the distance between the image collector and the corn plant is not less than 0.75 meters; when taking close-up images, the distance between the image collector and the corn plant is 0.2 meters; the pitch angle and roll angle range of the image collector are limited to (-30°, 30°).

[0060] 2. The acquired depth image is filled using the nearest neighbor interpolation method to make its resolution the same as that of the RGB image.

[0061] 3. A clip module is added to the YOLOv8 model. This module is located before the backbone module. The RGB image and the depth image are used as the input of YOLOv8. The RGB image is then truncated and cropped by the clip module as the input of the entire detection process, such as Figure 2 shown.

[0062] The truncation and cropping include:

[0063] The clip module first determines the truncated depth range according to the shooting distance, and then extracts the pixels in the truncated depth range from the depth image as the mask image, where the truncated depth range is [d min ,d min +0.3(d max -d min )],d max d min They are the maximum and minimum depth values ​​that can be read in the depth image.

[0064] 4. Apply the mask image to the RGB image, and through image overlay analysis, extract the RGB pixels covered by the mask image as the RGB image after mask processing.

[0065] 5. The YOLOv8 model is used to perform target recognition on the masked RGB image. For distant images, the recognition target is the corn aerial root and the segmentation target is the corn ear. For close-up images, the recognition target is the corn leaf node.

[0066] The recognition results of distant image and distant image are as follows: Figure 3 , Figure 4 shown.

[0067] 6. Determine the pixel coordinates of the key points output by the identified target, among which the pixel coordinates of the key points of the corn aerial root are the coordinates of the center point of the target frame, the pixel coordinates of the key points of the corn ear are the coordinates of the lowest end point of the segmented area, and the pixel coordinates of the key points of the corn leaf node are the coordinates of a point S on the leaf vein that is less than 50 pixels away from the leaf node, the intersection point M of the leaf and the stem, and the coordinates of a point C on the stem that is less than 50 pixels away from the leaf node.

[0068] The pixel coordinates of the key points of the near and far images are shown in Table 1 and Table 2 respectively:

[0069] Table 1 Pixel coordinates of key points in close-up images

[0070] key_point Mx key_point Mx key_point Cx 672 672 693 key_point Cx key_point Sx key_point Sy 693 694 720

[0071] Table 2 Pixel coordinates of key points in distant images

[0072] ear_x ear_y root_x root_y 740 928 673 1869

[0073] 7. Extract the coordinate depth D corresponding to the pixel coordinates of the key point from the depth image, and determine the depth confidence of the point. If the depth confidence is lower than the threshold, replace the depth D with the depth average of the 8 pixel points around the point whose confidence is higher than the threshold. If the confidence of the pixel points around the point are all lower than the threshold, discard the depth image.

[0074] 8. Convert the pixel coordinates of the key points into world coordinate system coordinates.

[0075] Since there is a gap between the pixel coordinates of the image and the actual spatial coordinates, a coordinate system conversion is required. The conversion involves three processes: pixel coordinate system → image coordinate system → camera coordinate system → world coordinate system.

[0076] The conversion from pixel coordinate system to image coordinate system is performed according to the following formula:

[0077] P image (u, v) = P pixel (x, y)-P center (C x , C y )

[0078] In the above formula, P pixel (x, y) is the coordinate of the pixel point, P center (C x , C y ) is the coordinate of the center of the image, P image (u, v) is the image coordinate obtained by transformation;

[0079] The transformation from image coordinates to world coordinates is achieved through the rotation matrix R, translation vector t, and intrinsic parameter matrix K, which are specifically defined as follows:

[0080]

[0081] In the above formula, (u, v) is the coordinate of the pixel in the image coordinate system, (x w ,y w ,z w ) is the three-dimensional coordinate of the pixel point in the camera coordinate system, Z c is the depth from the three-dimensional coordinate to the center of the image, f x 、f y is the focal length of the camera, (X c ,Y c ,Z c ) is the position of the center point of the camera in the world coordinate system, α, β, and γ are the yaw, pitch, and roll angles of the phone, respectively.

[0082] In this embodiment, the coordinates of the key points of the close-up image in the world coordinate system are as follows:

[0083] S point: [-0.27973470966848185, -0.2699330817189083, 1.466796875]

[0084] Point C: [-0.1595048574602088, -0.05785499396763191, 0.9013672471046448]

[0085] Point M: [-0.08165246985004983, -0.0009133816748723191, 0.463134765625];

[0086] The coordinates of the key points of the distant image in the world coordinate system are as follows:

[0087] Female ear: [11.159053080437113, -9.793132763669858, 3.2738947643526237]

[0088] Aerial roots: [11.873333014660819, -9.936938614938233, 3.4883662015631045].

[0089] 9. Determine the key points belonging to the same plant according to the Z value and X value range of the world coordinate system coordinates of the key points, and pair them, wherein the Z value and X value range are both less than 0.15 meters.

[0090] 10. The ear height and leaf angle of the corn plant are calculated based on the coordinates of the paired key points in the world coordinate system, where the ear height is calculated according to the following formula:

[0091]

[0092] In the above formula, EH is ear height, is the three-dimensional coordinate of the key point of the corn ear in the world coordinate system, is the three-dimensional coordinates of the key points of corn aerial roots in the world coordinate system;

[0093] The leaf angle is calculated according to the following formula:

[0094]

[0095] In the above formula, LA is the leaf angle, SM is the distance between point S and point M, MC is the distance between point M and point S, and SC is the distance between point S and point C. They are the three-dimensional coordinates of point S, point M, and point C in the world coordinate system.

[0096] The ear height calculated in this embodiment is 0.759522182, the actual measured value is 0.76, and the relative error is 0.06%; the leaf angle is 12.99°, the actual measured value is 14.7°, and the relative error is 1.03%.

[0097] Embodiment 2:

[0098] A field corn plant phenotyping system, such as Figure 5 As shown, it includes an image collector, a YOLOv8 model, a key point pixel coordinate determination module, a key point camera coordinate determination module, a key point pairing module, and an ear height and leaf angle calculation module.

[0099] The image collector is a mobile phone equipped with a TOF sensor, which is used to capture RGB images and depth images of corn plants in the field. The RGB images and depth images both include close-up images and distant-view images.

[0100] The YOLOv8 model is added with a clip module, which is located before the backbone module of YOLOv8;

[0101] The YOLOv8 model is used for:

[0102] Based on the clip module, the truncated depth range is determined according to the shooting distance, and the pixels of the truncated depth range are extracted from the depth image as the mask image. The truncated depth range is [d min ,d min +0.3(d max -dmin )],d max d min They are the maximum and minimum depth values ​​that can be read in the depth image respectively;

[0103] Apply the extracted mask image to the RGB image, and extract the RGB pixels covered by the mask image as the RGB image after mask processing through image superposition analysis;

[0104] The masked RGB image is used for target recognition. For distant images, the recognition target is the corn aerial root and the segmentation target is the corn ear. For close-up images, the recognition target is the corn leaf node.

[0105] The key point pixel coordinate determination module is used to determine the key point pixel coordinates output by the identified target, wherein the key point pixel coordinates of the corn aerial root are the coordinates of the center point of the target frame, the key point pixel coordinates of the corn female ear are the coordinates of the lowermost end point of the segmented area, and the key point pixel coordinates of the corn leaf node are the coordinates of a point S on the leaf vein whose distance from the leaf node is less than 50 pixels, the intersection point M of the leaf and the stem, and the coordinates of a point C on the stem whose distance from the leaf node is less than 50 pixels.

[0106] The key point camera coordinate determination module is used for:

[0107] Extract the coordinate depth D corresponding to the pixel coordinates of the key point from the depth image, and determine the depth confidence of the point at the same time. If the depth confidence is lower than a threshold, replace the depth D with the depth average of multiple pixel points around the point whose confidence is higher than the threshold. If the confidence of the pixel points around the point is lower than the threshold, discard the depth image;

[0108] Calculate the camera coordinate system coordinates of the key points;

[0109] The camera coordinate system coordinates of the key points and the posture data of the image collector are converted into the world coordinate system, and the key points belonging to the same plant are determined and paired by matching the Z value and X value coordinates of the world coordinate system coordinates, wherein the Z value and X value range are both less than 0.15 meters.

[0110] The ear height and leaf angle calculation module is used to calculate the ear height and leaf angle of the corn plant according to the world coordinate system coordinates of the paired key points, wherein:

[0111] The ear height is calculated according to the following formula:

[0112]

[0113] In the above formula, EH is ear height, are the three-dimensional coordinates of the key points of the corn ear, The three-dimensional coordinates of the key points of corn aerial roots;

[0114] The leaf angle is calculated according to the following formula:

[0115]

[0116] In the above formula, LA is the leaf angle, SM is the distance between point S and point M, MC is the distance between point M and point S, and SC is the distance between point S and point C. These are the three-dimensional coordinates of point S, point M, and point C respectively.

Claims

1. A method for measuring the phenotype of field corn plants, characterized in that: The method comprises: S1. Using an image collector to capture an RGB image and a depth image of corn plants in the field, wherein the RGB image and the depth image both include a near-view image and a far-view image; S2, extracting a mask image from the depth image based on the clip module added in the YOLOv8 model, and applying the extracted mask image to the RGB image to extract the RGB pixels covered by the mask image as the RGB image after mask processing; S3. Use the YOLOv8 model to perform target recognition on the masked RGB image. For the distant image, the recognition target is the corn aerial root and the segmentation target is the corn ear. For the close-up image, the recognition target is the corn leaf node. S4, determine the key point pixel coordinates output by the identified target, wherein the key point pixel coordinates of the corn aerial root are the coordinates of the midpoint of the target frame, the key point pixel coordinates of the corn ear are the coordinates of the lowest end point of the segmented area, and the key point pixel coordinates of the corn leaf node are the coordinates of a point S on the leaf vein that is less than 50 pixels from the leaf node, the intersection point M of the leaf and the stem, and the coordinates of a point C on the stalk that is less than 50 pixels from the leaf node; S5, extracting the coordinate depth D corresponding to the pixel coordinates of the key point from the depth image, and calculating the camera coordinate system coordinates of the key point; S6, converting the coordinates of the key points into world coordinates according to the camera coordinate system coordinates and the posture data of the image collector, determining the key points belonging to the same plant through coordinate matching and pairing them; S7. Calculate the ear height and leaf angle of the corn plant according to the world coordinate system coordinates of the paired key points.

2. A method for measuring the phenotype of field corn plants according to claim 1, characterized in that: The clip module is located before the backbone module of YOLOv8; In S2, extracting a mask image from a depth image based on the clip module added in the YOLOv8 model includes: The clip module first determines the cut-off depth range according to the shooting distance, and then extracts the pixels in the cut-off depth range from the depth image as the mask image. The cut-off depth range is [d min ,d min +0.3(d max -d min )],d max ,d min They are the maximum and minimum depth values ​​that can be read in the depth image.

3. A method for measuring the phenotype of field corn plants according to claim 1 or 2, characterized in that: In S5, the coordinate depth D corresponding to the pixel coordinates of the key point is extracted from the depth image and the depth confidence of the point is determined. If the depth confidence is lower than a threshold, the depth D is replaced by the depth average of multiple pixel points around the point whose confidence is higher than the threshold; if the confidence of the pixel points around the point are all lower than the threshold, the depth image is discarded.

4. A method for measuring the phenotype of field corn plants according to claim 1 or 2, characterized in that: In S7, the ear height is calculated according to the following formula: In the above formula, EH is ear height, are the three-dimensional coordinates of the key points of the corn ear, The three-dimensional coordinates of the key points of corn aerial roots; The leaf angle is calculated according to the following formula: In the above formula, LA is the leaf angle, SM is the distance between point S and point M, MC is the distance between point M and point S, and SC is the distance between point S and point C. These are the three-dimensional coordinates of point S, point M, and point C respectively.

5. A method for measuring the phenotype of field corn plants according to claim 1 or 2, characterized in that: The image collector is a mobile phone equipped with a TOF sensor. When shooting long-range images, the distance between the image collector and the corn plants is not less than 0.75 meters; when shooting close-range images, the distance between the image collector and the corn plants is 0.2 meters; the pitch angle and roll angle ranges of the image collector are both within (-30°, 30°).

6. A method for measuring the phenotype of field corn plants according to claim 1 or 2, characterized in that: The S6 determines the key points belonging to the same plant according to the Z value and X value range of the world coordinate system coordinates of the key points, and the Z value and X value range are both less than 0.15 meters.

7. A field corn plant phenotype measurement system, characterized in that: The system includes an image collector, a YOLOv8 model, a key point pixel coordinate determination module, a key point camera coordinate determination module, a key point pairing module, and an ear height and leaf angle calculation module; The image collector is used to capture RGB images and depth images of corn plants in the field, wherein the RGB images and depth images both include a close-up image and a distant image; The YOLOv8 model is used to: A mask image is extracted from the depth image based on the added clip module, and the extracted mask image is applied to the RGB image to extract the RGB pixels covered by the mask image as the RGB image after mask processing; The masked RGB image is used for target recognition. For distant images, the target is the aerial root of corn and the segmentation target is the female ear of corn. For close-up images, the target is the leaf node of corn. The key point pixel coordinate determination module is used to determine the key point pixel coordinates output by the identified target, wherein the key point pixel coordinates of the corn aerial root are the coordinates of the center point of the target frame, the key point pixel coordinates of the corn ear are the coordinates of the lowest end point of the segmented area, and the key point pixel coordinates of the corn leaf node are the coordinates of a point S on the leaf vein that is less than 50 pixels away from the leaf node, the intersection point M of the leaf and the stem, and the coordinates of a point C on the stem that is less than 50 pixels away from the leaf node; The key point camera coordinate determination module is used to extract the coordinate depth D corresponding to the pixel coordinates of the key point from the depth image, and calculate the camera coordinate system coordinates of the key point in combination with the posture and longitude and latitude data of the image collector; The key point pairing module is used to convert the coordinates of the key points into the world coordinate system coordinates according to the camera coordinate system coordinates and the posture data of the image collector, and determine the key points belonging to the same plant through coordinate matching and pair them; The ear height and leaf angle calculation module is used to calculate the ear height and leaf angle of the corn plant according to the world coordinate system coordinates of the paired key points.

8. A field corn plant phenotype measurement system according to claim 7, characterized in that: The clip module is located before the backbone module of YOLOv8; The extracting of the mask image from the depth image comprises: The clip module first determines the cut-off depth range according to the shooting distance, and then extracts the pixels in the cut-off depth range from the depth image as the mask image. The cut-off depth range is [d min ,d min +0.3(d max -d min )],d max ,d min They are the maximum and minimum depth values ​​that can be read in the depth image respectively; The key point camera coordinate determination module extracts the coordinate depth D corresponding to the pixel coordinates of the key point from the depth image and determines the depth confidence of the point. If the depth confidence is lower than a threshold, the depth D is replaced by the depth average of multiple pixel points around the point whose confidence is higher than the threshold. If the confidence of the pixel points around the point are all lower than the threshold, the depth image is discarded.

9. A field corn plant phenotype measurement system according to claim 7 or 8, characterized in that: The ear height is calculated according to the following formula: In the above formula, EH is ear height, are the three-dimensional coordinates of the key points of the corn ear, The three-dimensional coordinates of the key points of corn aerial roots; The leaf angle is calculated according to the following formula: In the above formula, LA is the leaf angle, SM is the distance between point S and point M, MC is the distance between point M and point S, and SC is the distance between point S and point C. These are the three-dimensional coordinates of point S, point M, and point C respectively.

10. A field corn plant phenotype measurement system according to claim 7 or 8, characterized in that: The image collector is a mobile phone equipped with a TOF sensor. When shooting long-range images, the distance between the image collector and the corn plants is not less than 0.75 meters; when shooting close-range images, the distance between the image collector and the corn plants is 0.2 meters; the pitch angle and roll angle ranges of the image collector are both within (-30°, 30°).

Images