Plant root system multi-dimensional data acquisition method, support and storage medium thereof

Abstract

The invention relates to a plant root system multi-dimensional data acquisition method, a support and a storage medium thereof. The method comprises the following steps: in response to an acquired plant root system data acquisition instruction, acquiring a plant root system full-coverage image sequence; splicing adjacent root system images in the plant root system full-coverage image sequence to generate a rough two-dimensional root system expansion graph; performing root system main distribution area detection on the rough two-dimensional root system expansion diagram to obtain a root system thermodynamic diagram; the standard path is optimized according to the root thermodynamic diagram, and an optimized path and a recollection instruction are generated; and extracting a root system structure according to the received plant root system multi-view image set to obtain plant root system data. By adopting the method, the standard path can be optimized, the camera is ensured to perform more accurate sampling in the root system dense area, and unnecessary sampling is reduced, so that the data precision and efficiency are improved, and the acquisition time and resource consumption are reduced.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of image processing, and particularly relates to a plant root system multi-dimensional data acquisition method, a support and a storage medium thereof. BACKGROUND

[0002] With the continuous development of plant biology and agricultural science technology, the importance of plant root system research in agricultural optimization, ecological environment monitoring and soil health assessment is increasingly prominent. As the main channel for plant water and nutrient absorption, the growth and distribution pattern of root system directly affects the health and productivity of plants.

[0003] In order to better understand the growth rule of root system, in the traditional technology, root system data acquisition usually relies on manual excavation or static image acquisition, which not only consumes time and effort, but also often damages the integrity of plant root system, and the precision is not high, and detailed multi-dimensional information is difficult to obtain. The existing root system imaging system mostly uses a static camera and manually sets a path for shooting, and image stitching and data processing also need to rely on complex manual operation or efficient but expensive computing resources. The existing system is not intelligent enough in automatic recognition of root system area and path planning, which may cause repeated or invalid shooting in the collection process, thereby affecting the data collection efficiency and accuracy. SUMMARY

[0004] Therefore, it is necessary to provide a plant root system multi-dimensional data acquisition method, a support and a storage medium thereof, which can automatically optimize the data acquisition path according to the distribution of plant root system.

[0005] In a first aspect, the application provides a plant root system multi-dimensional data acquisition method, comprising:

[0006] In response to the acquired plant root system data acquisition instruction, a plant root system full coverage image sequence is acquired; the plant root system data acquisition instruction is used to instruct a camera to perform a shooting operation on a plant root system according to a standard path, and to feed back in the form of a plant root system full coverage image sequence; the standard path is a path point parameter set; the path point parameters include a horizontal step length, a vertical step length, a Z-axis height layer number and a point view angle;

[0007] Adjacent root system images in the plant root system full coverage image sequence are spliced to generate a rough two-dimensional root system development map;

[0008] The rough two-dimensional root system development map is subjected to root system main distribution area detection to obtain a root system heat map;

[0009] The standard path is optimized according to the root system heat map to generate an optimized path and a reacquisition instruction; the reacquisition instruction is used to instruct the camera to perform a shooting operation on the plant root system according to the optimized path, and to feed back in the form of a plant root system multi-view image set;

[0010] According to the received plant root system multi-view image set, root system structure extraction is performed to obtain plant root system data.

[0011] In one of the embodiments, adjacent root system images in the plant root system full coverage image sequence are spliced to generate a rough two-dimensional root system development map, including:

[0012] Feature point extraction is performed on each root system image in the plant root system full coverage image sequence to obtain a key point set corresponding to each root system image; the key point set includes key points and corresponding descriptors;

[0013] The descriptor similarity between the key point sets corresponding to adjacent images is calculated, and key point matching is performed to obtain a matching pair set;

[0014] The image change relationship between the matching points is calculated according to the matching pair set to obtain a homography matrix;

[0015] The plant root system full coverage image sequence is mapped to a common development plane according to the homography matrix between adjacent images to splice and generate a rough two-dimensional root system development map.

[0016] In one of the embodiments, the rough two-dimensional root system development map is subjected to root system main distribution area detection to obtain a root system heat map, including:

[0017] Edge detection and connected domain analysis are performed on the rough two-dimensional root system development map to obtain a root region probability map;

[0018] The rough two-dimensional root system development map is subjected to block division to obtain a block root system development map; the block root system development map contains multiple blocks and corresponding shooting poses;

[0019] The root system pixel density of each block in the block root system development map is calculated according to the root region probability map to obtain a root system distribution density matrix;

[0020] The root system distribution weight of each block is calculated according to the root system distribution density matrix to obtain a root system distribution heat map.

[0021] In one of the embodiments, the standard path is optimized according to the root system heat map to generate an optimized path and a reacquisition instruction, including:

[0022] A plurality of key attention blocks are obtained according to the root system distribution weight in the root system heat map;

[0023] The path point parameters in the standard path corresponding to the shooting poses of the key attention blocks are obtained;

[0024] The path point parameters of the key attention block are refined according to the root system distribution weight, to obtain an optimized path and a reacquisition instruction; the refinement operation corresponds to shortening a step length, increasing a point view angle, increasing a Z-axis height layer number and increasing a sampling number operation.

[0025] In one of the embodiments, root system structure extraction is performed according to the received plant root system multi-view image set to obtain plant root system data, including:

[0026] Disparity recovery and depth extraction are performed on the plant root system multi-view image set to obtain a root system sparse point cloud model;

[0027] Root system structure skeleton extraction is performed according to the root system sparse point cloud model to obtain a root system skeleton graph; the root system skeleton graph includes a main root and a lateral root;

[0028] A root system skeleton graph curve is fitted according to the root system sparse point cloud model to obtain a root system curve;

[0029] Plant root system data is obtained by calculating parameters of each root system curve based on camera parameters and the optimized path; the plant root system data includes root length, root depth, lateral extension distance, diameter and branch density.

[0030] In one of the embodiments, the descriptor similarity between the key point sets corresponding to adjacent images is calculated, and key point matching is performed to obtain a matching pair set, including:

[0031] The key points between the two key point sets corresponding to adjacent image distributions are paired one by one, and the similarity distance of the descriptors is calculated;

[0032] Based on the principle of minimum similarity distance, a key point matching pair is obtained;

[0033] A preset number of key point matching pairs are extracted to construct a transformation model;

[0034] All key point matching pairs are projected according to the transformation model, and the projection error of each key point matching pair is calculated;

[0035] If the projection error exceeds a preset threshold, the key point matching pair is removed to obtain a maximum consistent matching pair set.

[0036] In one of the embodiments, disparity recovery and depth extraction are performed on the plant root system multi-view image set to obtain a root system sparse point cloud model, including:

[0037] Feature point extraction is performed on the plant root system multi-view image set, and feature point matching is performed to obtain a matching point pair;

[0038] Depth information extraction is performed on the plant root system multi-view image set according to camera calibration and the matching point pair to obtain feature point three-dimensional coordinates;

[0039] The feature point three-dimensional coordinates are converted by parallax to obtain a sparse point cloud;

[0040] The sparse point cloud is subjected to point cloud denoising and point cloud reconstruction processing to obtain a root system sparse point cloud model.

[0041] In a second aspect, the present application also provides a plant root system multi-dimensional data acquisition support, comprising an adjustable camera equipment support assembly, a root system holder and a controller, wherein the controller has a computer program stored thereon, and the computer program is executed by a processor to realize the steps of any of the above plant root system multi-dimensional data acquisition methods.

[0042] In a third aspect, the present application also provides a computer device, comprising a memory and a processor, wherein the memory has a computer program stored thereon, and the processor executes the computer program to realize the steps of any of the above plant root system multi-dimensional data acquisition methods.

[0043] In a fourth aspect, the present application also provides a computer readable storage medium, wherein the computer readable storage medium has a computer program stored thereon, and the computer program is executed by a processor to realize the steps of any of the above plant root system multi-dimensional data acquisition methods.

[0044] The above plant root system multi-dimensional data acquisition method, support and storage medium thereof automatically execute the shooting operation according to the standard path parameters through the automatic path control, ensure the consistency and accuracy of the data acquisition process, reduce the human error, improve the acquisition efficiency, and ensure the quality and coverage range of the image. Compared with the traditional manual operation mode, more complete and accurate root system data can be obtained in a shorter time. The adjacent images are automatically spliced and a rough two-dimensional root system development map is generated, which can effectively show the overall structure of the root system. The process is based on accurate feature point matching and image transformation to ensure high image alignment during the splicing process. The automatic splicing process improves the efficiency of image processing, and the generated two-dimensional development map can accurately present the overall appearance of the root system, providing an intuitive view basis for subsequent analysis. Through the generation of the root system heat map, the high-density area of the root system is intelligently identified, and the acquisition path is optimized based on this information to guide the camera to only finely sample in the key area, avoiding the acquisition of invalid areas. Path optimization greatly saves acquisition time and resources, especially in large-scale root system scanning, which can significantly improve the acquisition efficiency and data quality. The optimized path can ensure higher precision sampling while reducing unnecessary workload, and can accurately extract the three-dimensional structure of the root system, including root length, root depth, lateral expansion distance and other parameters. BRIEF DESCRIPTION OF DRAWINGS

[0045] In order to make the technical solutions in the embodiments of the present application or the related art clearer, the accompanying drawings needed to be used in the embodiments or the related art description will be briefly introduced. Obviously, the accompanying drawings in the following description only constitute some embodiments of the present application, and for those skilled in the art, other drawings can be obtained without creative effort based on these drawings.

[0046] Figure 1 A flowchart of a plant root system multi-dimensional data acquisition method of the present application is shown in the figure.

[0047] Figure 2 A flowchart of a sub-step of step S102 is shown in the figure.

[0048] Figure 3 A flowchart of a sub-step of step S103 is shown in the figure.

[0049] Figure 4 A flowchart of a sub-step of step S105 is shown in the figure. DETAILED DESCRIPTION

[0050] In order to make the technical solutions in the embodiments of the present application or the related art clearer, the accompanying drawings needed to be used in the embodiments or the related art description will be briefly introduced. Obviously, the accompanying drawings in the following description only constitute some embodiments of the present application, and for those skilled in the art, other drawings can be obtained without creative effort based on these drawings.

[0051] In one embodiment, as shown in Figure 1 , a plant root system multi-dimensional data acquisition method is provided, and the present embodiment takes the method applied to a terminal as an example. It should be understood that the method can also be applied to a server, and can also be applied to a system including a terminal and a server, and is realized through the interaction of the terminal and the server. In the present embodiment, the method includes the following steps:

[0052] S101, in response to the acquired plant root system data acquisition instruction, a plant root system full coverage image sequence is acquired; the plant root system data acquisition instruction is used to instruct a camera to perform a shooting operation on a plant root system according to a standard path, and the shooting operation is fed back in the form of a plant root system full coverage image sequence; the standard path is a path point parameter set; the path point parameters include a horizontal step length, a vertical step length, a Z-axis height layer number and a point view angle.

[0053] Illustratively, the plant root system data acquisition instruction instructs the camera to scan and shoot the plant root system according to a pre-set standard path. The standard path is composed of a series of path point parameters, each path point including horizontal step, vertical step, Z-axis height layer, and point view angle, to determine how the camera scans along the root system area distribution in three-dimensional space. Specifically, the horizontal step and the vertical step determine the scanning interval of the camera in the horizontal and vertical directions, the Z-axis height layer controls the camera to scan the plant root system at different height levels, and the point view angle ensures that the root system is shot from different angles. The setting of the standard path ensures that the camera can preliminarily cover the entire root system area, maximally reducing the missed area, thereby obtaining full-coverage root system image data. Illustratively, the camera shoots point by point according to the standard path, records the plant root system full-coverage image and the corresponding shooting pose data in real time, and feeds back the image sequence.

[0054] S102, stitch the adjacent root system images in the plant root system full-coverage image sequence to generate a rough two-dimensional root system development map.

[0055] Illustratively, the overlapping part between each pair of adjacent images is analyzed, the common feature points are identified, and the transformation relationship between the images is calculated using the feature points. By transforming and stitching the images in the correct positional relationship, a rough two-dimensional root system development map is finally formed.

[0056] S103, detect the main distribution area of the root system in the rough two-dimensional root system development map to obtain a root system heat map.

[0057] Illustratively, by using image processing algorithms such as threshold segmentation and region growing, the dense area of the root system in the image is identified, and a root system heat map is generated according to the dense area. The highlighted area in the heat map represents the main distribution area of the root system, which can quickly lock the core growth area of the root system to determine whether to pay further attention.

[0058] S104, optimize the standard path according to the root system heat map to generate an optimized path and a re-acquisition instruction; the re-acquisition instruction is used to instruct the camera to shoot the plant root system according to the optimized path, and to feed back in the form of a plant root system multi-view image set.

[0059] Illustratively, according to the root system heat map, the original standard path is optimized to generate a new acquisition path. The high-density root system area in the heat map is analyzed in detail, and the shooting path of the camera is adjusted to ensure that the high-density root system area can be paid more attention to and sampled multiple times. Specifically, the generation process of the optimized path mainly depends on the hot spot area in the heat map, by increasing the number of acquisitions of the hot spot area, changing the angle or moving direction of the camera, thereby improving the acquisition accuracy of the key area.

[0060] Further, while generating the optimized path, a reacquisition instruction is automatically generated to instruct the camera to perform the shooting operation according to the optimized path. The optimized path not only reduces the collection of redundant data, but also improves the shooting efficiency and ensures the accurate recording of the key areas of the root system. The execution of the reacquisition instruction is flexibly adjusted based on the density distribution of the root system area, so as to maximize the efficiency and accuracy of image acquisition while ensuring the comprehensiveness of the data.

[0061] S105, extracting the root system structure according to the received plant root system multi-view image set to obtain plant root system data.

[0062] Illustratively, the root system structure is extracted from the received plant root system multi-view image set, and a three-dimensional reconstruction algorithm and a structure analysis method are used to recover the three-dimensional structure of the plant root system from the multi-view images. Specifically, by using image registration, feature extraction and triangulation techniques, an accurate three-dimensional point cloud model is generated, and the geometric morphological features of the root system, such as the total length of the root system, the branching angle, the depth of the root system and other key parameters, are extracted from the model.

[0063] In the above plant root system multi-dimensional data acquisition method, the shooting operation of the camera is controlled according to the standard path in response to the plant root system data acquisition instruction, which improves the efficiency and consistency of root system data acquisition, avoids errors in manual operation, and ensures the comprehensiveness and accuracy of the data. By splicing adjacent root system images, the generated two-dimensional root system expansion diagram provides a clear visual display of the root system structure, which can more intuitively analyze the root system distribution. By detecting the main distribution area of the root system on the rough two-dimensional root system expansion diagram, a root system heat map is obtained, and the generation of the heat map can quickly identify the dense area of the root system as the focus of root system research. According to the information of the key area provided by the root system heat map, the standard path is optimized to generate a new optimized path and a reacquisition instruction, and the optimized path will ensure more accurate sampling of the camera in the dense area of the root system, while reducing unnecessary sampling, thereby improving the accuracy and efficiency of the data and reducing the collection time and resource consumption. The optimized path ensures accurate scanning of the key areas of the root system and improves the sampling quality. According to the plant root system multi-view image set after reacquisition, the structural information of the root system, such as root length, root depth, root lateral expansion and other key parameters, can be further extracted, and the root system structure can be obtained from different angles and depths to ensure the comprehensiveness and detail of the root system data.

[0064] In one embodiment, as shown in FIG. 1, the plant root system full coverage image sequence is obtained by using a camera to capture the root system of a plant from different angles and depths, and the camera is controlled to perform the shooting operation according to a standard path. Figure 2 As shown in FIG. 2, the adjacent root system images in the plant root system full coverage image sequence are spliced to generate a rough two-dimensional root system expansion diagram, including:

[0065] S201, feature point extraction is performed on each root system image in the plant root system full coverage image sequence to obtain a key point set corresponding to each root system image; the key point set includes key points and corresponding descriptors.

[0066] Illustratively, a feature point is a point in an image that has high recognition and can still exist stably under different viewing angles or illumination. In order to ensure the accuracy and stability of image stitching, a SIFT (Scale-Invariant Feature Transform) algorithm is used to extract key points in the image and calculate the descriptor of each feature point. The descriptor is a high-dimensional vector that quantifies the local region information of each feature point, which can effectively express the visual features of the point for subsequent matching process. By extracting the key points of each root system image and their corresponding descriptors, a data structure containing multiple key point sets can be obtained, each set representing the feature information in an image. Specifically, the SIFT algorithm is used to extract the position coordinates of each key point in the image, the rotation angle relative to the image, and the scale information describing the size of the image region corresponding to the key point. Further, the descriptor of the key point is calculated by the gradient direction of the local region.

[0067] S202, the descriptor similarity between the key point sets corresponding to adjacent images is calculated, and key point matching is performed to obtain a matching pair set.

[0068] Illustratively, the Euclidean distance is used to calculate the similarity between the descriptors to determine whether the feature points in different images represent the same physical point or region. Further, by using algorithms such as Brute Force Matching or FLANN (Fast Library for Approximate Nearest Neighbors), the feature point pairs in the same position in each pair of adjacent images are matched to mark the positions and relationships of the corresponding points between images.

[0069] S203, the image change relationship between the matching points is calculated according to the matching pair set to obtain a homography matrix.

[0070] Illustratively, a homography matrix is a mathematical tool that describes the transformation relationship between two planes, which can map a point in one image to another image through geometric transformation. Specifically, RANSAC (Random Sample Consensus) is used to optimize the error of the matching point pairs. The RANSAC algorithm can effectively remove false matches and ensure that the calculated homography matrix reflects the true transformation relationship between the images, rather than noise or false matching points. Illustratively, a mapping equation is constructed using the coordinates of the matching point pairs, and the least squares method is used to solve the equation set to obtain the homography matrix, which represents the mapping of a point in one of the adjacent images to the corresponding position in the other image.

[0071] S204, mapping the plant root system full coverage image sequence to a common unwinding plane according to the homography matrix between adjacent images for splicing to generate a rough two-dimensional root system unwinding map.

[0072] Illustratively, each image is transformed and superimposed according to its spatial positional relationship by the homography matrix, so that they are seamlessly connected together to form a complete and continuous two-dimensional root system unwinding map. Optionally, in the splicing process, a multi-resolution hybrid algorithm is used to consider the overlapping area between images, reduce the seam, and further balance the brightness, color difference, etc. of the images to ensure the visual consistency of the spliced images. The rough two-dimensional root system unwinding map obtained will cover the entire root system area.

[0073] In one embodiment, as shown in Figure 3 The rough two-dimensional root system unwinding map is subjected to root system main distribution area detection to obtain a root system heat map, including:

[0074] S301, edge detection and connected component analysis are performed on the rough two-dimensional root system unwinding map to obtain a root region probability map.

[0075] Illustratively, the Canny edge detection algorithm is used to extract the edges of the image, and the edge points represent the significant difference area of the root system from the background in the image, which can effectively highlight the outline of the root system. Further, the connected component analysis method is applied to connect the root system part area in the image, and each connected component represents a region of the root system, which can distinguish different parts of the root system. Optionally, a root region probability map is generated, in which each pixel value represents the probability that the pixel belongs to the root system.

[0076] S302, the rough two-dimensional root system unwinding map is divided into blocks to obtain a block root system unwinding map; the block root system unwinding map includes a plurality of blocks and their corresponding shooting poses.

[0077] Illustratively, the rough two-dimensional root system unwinding map is divided into a plurality of small blocks, and the size of each block can be flexibly set according to the image resolution, the characteristics of the root system distribution, and the actual collection needs. Illustratively, the blocks can be automatically divided based on the grid division of the image or according to the distribution of the dense root system area. According to the shooting poses of the plant root system full coverage image sequence taken along the standard path, each block obtained after division will contain a certain number of root system pixels and their corresponding shooting pose information, and the image information in each block can be processed individually for local analysis.

[0078] S303, the root system pixel density of each block in the block root system unwinding map is calculated according to the root region probability map to obtain a root system distribution density matrix.

[0079] Illustratively, root pixel density of each block is calculated, which represents the number of root pixels in each block, and further reflects the growth density of the root system in the region. Specifically, the root pixel density is obtained by counting the root region pixels in each block, calculating the number of root pixels in each block, and obtaining the root pixel density of the block according to the pixel number of the block. The higher the density value, the more dense the root system distribution in the region. The root pixel density information of all blocks will be summarized into a root distribution density matrix, wherein each matrix element represents the density value of the corresponding block.

[0080] S304, calculate the root distribution weight of each block according to the root distribution density matrix, and obtain the root distribution heat map.

[0081] Illustratively, the root distribution weight is calculated according to the data in the root distribution density matrix. Illustratively, the root distribution weight is proportional to the pixel density, and the region with higher density will be given greater weight. The higher the weight value, the greater the root density in the region, and vice versa, indicating lower root density. Further, by normalizing the root distribution weight of all blocks, a root heat map is finally generated. Each pixel value in the heat map represents the growth intensity or density of the root system at that position. The high-light area of the image represents the area where the root system grows most densely, while the low-light area represents the area where the root system is sparse or has no root system.

[0082] In one embodiment, the standard path is optimized according to the root heat map to generate an optimized path and a reacquisition instruction, including:

[0083] S41, obtain a plurality of key attention blocks according to the root distribution weight in the root heat map.

[0084] Illustratively, by threshold segmentation of the weight value of each region in the heat map, a plurality of key attention blocks can be identified from the heat map, which are usually represented as high-light areas in the image, representing the place where the root system grows most densely. Optionally, in the root heat map, the position of each key attention block will correspond to a specific two-dimensional coordinate. According to the coordinate, it is determined which regions need to be sampled and which regions can be ignored, thereby guiding the direction of subsequent path optimization.

[0085] S42, obtain the path point parameters in the standard path according to the shooting pose corresponding to the key attention block.

[0086] Illustratively, the shooting pose corresponding to each key focus block will contain multiple parameters, including the horizontal step length, vertical step length, Z-axis height layer, and point of view of the camera. According to the position and size of the focus block, a new set of path point parameters will be assigned to each key block in the optimized path, ensuring that the camera's scanning path in the key area is optimized, avoiding missing important areas, while avoiding repeated sampling of blank areas.

[0087] S43, according to the root distribution weight, the path point parameters of the key focus block are refined to obtain an optimized path and a re-sampling instruction; the refinement operation corresponds to shortening the step length, increasing the point of view, increasing the Z-axis height layer, and increasing the sampling number operation.

[0088] Illustratively, in the area where the root system is dense, the horizontal step length and the vertical step length of the camera are automatically shortened according to the weight value, so that the sampling of the camera in the area where the root system is dense is more intensive. The shortening of the step length can ensure that more detailed images are collected in a short time. In order to obtain more root system information from different angles, the point of view of the camera in the key focus block is increased, and the same area is shot from multiple angles, so as to generate more comprehensive and detailed image data to extract multi-level root system structure. For the area where the root system is deep, the Z-axis height layer of the camera is increased, that is, shooting is performed at different height levels to capture deeper root system information, especially in the more complex root system structure, which can effectively avoid missing the bottom root system. In the area where the root system is particularly dense, the sampling number is increased, and through multiple sampling, the accuracy and coverage of the data are improved, and the incomplete image caused by sampling error or omission is reduced.

[0089] Further, after completing the path optimization and parameter refinement, a re-sampling instruction is generated according to the optimized path point to guide the camera to shoot according to the new optimized path, and the optimized sampling area is collected.

[0090] In one embodiment, as shown in Figure 4 , according to the received plant root system multi-view image set, the root system structure is extracted to obtain plant root system data, including:

[0091] S401, disparity recovery and depth extraction are performed on the plant root system multi-view image set to obtain a root system sparse point cloud model.

[0092] Schematically, the depth value of each pixel in the image is calculated by analyzing the difference information between multi-view images, thereby constructing the three-dimensional structure of the root system. Specifically, by analyzing images from different perspectives, the displacement of the same point under different perspectives is calculated to obtain a disparity value map. The three-dimensional structure in the image is restored based on the disparity value. Furthermore, the three-dimensional structure in the disparity information is converted into a depth map, and the depth map is back-projected to generate a sparse point cloud model. The point cloud model consists of a large number of three-dimensional coordinate points, representing the distribution of the root system in space.

[0093] S402 , extracting the root system structure skeleton based on the root system sparse point cloud model to obtain a root system skeleton map; the root system skeleton map includes main roots and lateral roots.

[0094] Schematically, a skeletonization algorithm based on Marching Cubes (voxel-level reconstruction algorithm) was used to extract the main skeleton of the root system. The skeleton extraction process uses connectivity analysis and spatial curve fitting to extract the skeleton lines representing the main root and lateral roots from the point cloud. The extracted skeleton image includes the structural information of the main root and lateral roots, which can effectively simplify and depict the morphological characteristics of the root system.

[0095] S403 , fitting a root system skeleton curve according to the root system sparse point cloud model to obtain a root system curve.

[0096] Furthermore, a fitting algorithm is used to perform curve fitting on each root system in the root skeleton diagram, obtaining a curve description of each root system to accurately calculate the root system's shape and geometric characteristics. Specifically, a B-spline fitting algorithm is applied to each root system to obtain a smooth curve representing the root system's growth path and shape. This curve fitting method can be used to determine the spatial trajectory of each root system, including geometric characteristics such as the root's curvature and branching angles.

[0097] S404. Calculate the parameters of each root curve based on the camera parameters and the optimized path to obtain plant root data; the plant root data includes root length, root depth, lateral extension distance, diameter, and branch density.

[0098] Illustratively, according to the camera parameters and the optimized path, the mapping parameters of the parameters of each root curve in the real space scale are calculated, and finally the plant root data are obtained. The root length refers to the distance from the root base to the farthest end point of the root system. The root length of each root system can be obtained by calculating the total length of the root curve. The root depth refers to the deepest point of the root system along the Z-axis direction. The maximum depth of the root system is determined by analyzing the Z-axis coordinates of the curve. The lateral extension distance represents the extension degree of the root system in the horizontal direction. The maximum distance of the lateral extension is obtained by analyzing the projection range or the horizontal step length of the root curve in the X, Y plane. The diameter is the thickness of the root system. The diameter of the root system is usually calculated at a specific position of the root curve. The branching density represents the number of branches per unit length, reflecting the lush degree of the root system. The branching density is obtained by analyzing the branch points in the root curve and calculating the number of branches per unit length.

[0099] In one embodiment, the descriptor similarity between the key point sets corresponding to adjacent image pairs is calculated, and key point matching is performed to obtain a matching pair set, including:

[0100] S51, the key points between the two key point sets corresponding to the adjacent image distribution are paired one by one, and the similarity distance of the descriptor is calculated.

[0101] Illustratively, the SIFT feature point extraction algorithm is used to obtain the key point set in the adjacent two images. Each key point is accompanied by a descriptor. The similarity of the descriptors between each pair of key points is calculated by Euclidean distance to obtain the similarity distance, which represents the similarity between the two key points. The smaller the similarity distance is, the more likely it is that the two key points are matching points at the same position.

[0102] S52, based on the principle of minimum similarity distance, a key point matching pair is obtained.

[0103] The similarity distances of all key point pairs are calculated, and the final key point matching pair is determined according to the principle of minimum similarity. Specifically, according to the calculated similarity distance, the key point pair with the smallest distance is selected as the matching pair. The principle of minimum similarity distance can ensure that the features of the selected matching point pair are as similar as possible, thereby improving the reliability of the matching. The matching information of all key point pairs is screened, and finally a key point matching pair set is obtained, which contains all the matching point pairs with the smallest similarity and the highest reliability.

[0104] S53, a preset number of key point matching pairs are extracted to construct a transformation model.

[0105] Illustratively, a fundamental matrix is used to construct the transformation model, specifically, the RANSAC algorithm is used to calculate the matching relationship of the coordinate mapping of the plurality of key point matching pairs to obtain the transformation model to describe how one image is mapped to another image through rotation, scaling and translation.

[0106] S54, project all key point matching pairs according to the transformation model, and calculate the projection error of each key point matching pair.

[0107] Illustratively, according to the constructed transformation model, the key point coordinates of one image in each matching pair are projected into another image through the transformation model. Specifically, the key points of one image are mapped to the spatial position of another image through known geometric relationships to obtain the projected position. Further, the actual distance between the projected position and the original matching point is calculated to obtain the projection error, which is an important basis for verifying whether the matching pair is consistent. The smaller the error, the more accurate the projection result, and the higher the quality of the matching pair.

[0108] S55, if the projection error exceeds the preset threshold, the key point matching pair is removed to obtain the maximum consistent matching pair set.

[0109] Illustratively, if the projection error exceeds the preset threshold, it is determined that the matching pair is unreliable, and it is removed from the matching pair set. The corresponding matching points of the key points are re-found, and the maximum consistent matching pair set is obtained through multiple iterations. The threshold can be determined according to the resolution of the image, the quality of the feature points and the accuracy requirement of the actual application.

[0110] In one embodiment, the multi-view image set of the plant root system is subjected to parallax recovery and depth extraction to obtain a sparse point cloud model of the root system, comprising:

[0111] S61, feature point extraction is performed on the multi-view image set of the plant root system, and feature point matching is performed to obtain a matching point pair.

[0112] Similarly, the multi-view image set of the plant root system is subjected to feature point matching again, that is, a feature point extraction algorithm is used to extract representative feature points from each image. The feature points are usually located at the edges, corner points or texture-rich areas in the image. By calculating the similarity between the feature point descriptors in different images, the feature points from different perspectives are matched to obtain a matching point pair.

[0113] S62, according to the camera calibration and the matching point pair, depth information is extracted from the multi-view image set of the plant root system to obtain three-dimensional coordinates of the feature points.

[0114] Illustratively, the camera calibration is the internal and external parameters of the camera, including focal length, optical center position, camera rotation and translation matrix, etc. Through the camera calibration process, the projection matrix of the camera is obtained, and the matrix is used for depth information calculation. Further, through the triangulation method, the coordinates of the feature points in the three-dimensional space are calculated according to the coordinate difference of the matching point pair in two views, i.e. parallax. Specifically, the three-dimensional coordinates of each feature point are calculated using the camera internal and external parameters and the pixel coordinates of the matching point pair.

[0115] S63, the three-dimensional coordinates of the feature points are converted into parallax to obtain a sparse point cloud.

[0116] Illustratively, parallax refers to the horizontal offset of corresponding points in two different view images. The parallax value is inversely proportional to the depth. The parallax value of each matching point pair is converted into a depth value, and all three-dimensional coordinate data are represented in the form of a point cloud to generate a sparse point cloud containing all the extracted feature points.

[0117] S64, the sparse point cloud is subjected to point cloud denoising and point cloud reconstruction processing to obtain a root system sparse point cloud model.

[0118] Illustratively, in order to improve the quality of the point cloud, statistical filtering, neighborhood averaging method and other point cloud denoising algorithms are applied to remove mis-matched points or abnormal points. The point cloud is interpolated or fitted to supplement and reconstruct the point cloud, thereby improving the density and accuracy of the point cloud.

[0119] It should be understood that although each step in the flowchart involved in each of the above embodiments is displayed in sequence according to the arrow, these steps are not necessarily executed in sequence according to the arrow. Unless otherwise specified herein, the execution of these steps is not strictly limited in sequence, and these steps can be executed in other sequences. Moreover, at least part of the steps in the flowchart involved in each of the above embodiments can include multiple steps or stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution sequence of these steps or stages is not necessarily sequential, but can be executed in rotation or alternation with at least part of other steps or steps or stages in other steps.

[0120] Based on the same inventive concept, the plant root system multi-dimensional data acquisition support for implementing the above-mentioned plant root system multi-dimensional data acquisition method is also provided, which includes an adjustable camera equipment support assembly, a root system clamp and a controller, the controller has a computer program stored therein, and the computer program is executed by a processor to implement the steps in each of the above method embodiments.

[0121] In one embodiment, a computer device is provided, comprising a memory and a processor, the memory stores a computer program, and the processor implements the steps in the above method embodiments when executing the computer program.

[0122] In one embodiment, a computer readable storage medium is provided, which stores a computer program, and the computer program is executed by a processor to implement the steps in the above method embodiments.

[0123] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts are described in the method embodiments. The above described device embodiments are only illustrative, wherein the components described as separate components can or can not be physically separated, and the components displayed as units can or can not be physical units, i.e., they can be located in one place or distributed on multiple network units. Part or all of the modules can be selected to achieve the purpose of the present disclosure according to actual needs. Those skilled in the art can understand and implement it without creative labor.

[0124] The above described embodiments only express several implementation manners of the present application, and the description is more specific and detailed, but it should not be understood as a limitation on the patent scope of the application. It should be pointed out that, for those skilled in the art, without departing from the concept of the present application, several modifications and improvements can be made, which are all within the protection scope of the present application.

Claims

1. A method for collecting multi-dimensional data of plant roots, characterized in that: The method comprises: In response to the acquired plant root data acquisition instruction, a plant root system full coverage image sequence is acquired; the plant root system data acquisition instruction is used to instruct the camera to photograph the plant root system along a standard path and provide feedback in the form of the plant root system full coverage image sequence; the standard path is a set of path point parameters; the path point parameters include a horizontal step length, a vertical step length, a number of Z-axis height layers, and a point viewing angle; splicing adjacent root system images in the plant root system full coverage image sequence to generate a rough two-dimensional root system expansion map; Detecting the main distribution areas of the root system on the rough two-dimensional root system expansion map to obtain a root system heat map; Optimizing the standard path according to the root system heat map to generate an optimized path and a re-capture instruction; the re-capture instruction is used to instruct the camera to photograph the plant root system according to the optimized path and provide feedback in the form of a multi-view image set of the plant root system; Root system structure extraction is performed based on the received plant root system multi-view image set to obtain plant root system data.

2. The method according to claim 1, characterized in that The step of splicing adjacent root system images in the plant root system full coverage image sequence to generate a rough two-dimensional root system expansion map includes: Extracting feature points from each root image in the plant root full coverage image sequence to obtain a key point set corresponding to each root image; the key point set includes key points and their corresponding descriptors; Calculating the descriptor similarity between the key point sets corresponding to adjacent images, and performing key point matching to obtain a set of matching pairs; Calculating the image change relationship between the matching points according to the matching pair set to obtain a homography matrix; The plant root system full coverage image sequence is mapped onto a common expansion plane according to the homography matrix between adjacent images, and then spliced ​​to obtain the rough two-dimensional root system expansion map.

3. The method according to claim 2, characterized in that The detecting of the main root distribution areas on the rough two-dimensional root system expansion diagram to obtain a root system heat map includes: Performing edge detection and connected domain analysis on the rough two-dimensional root system expansion map to obtain a root area probability map; Dividing the rough two-dimensional root system expansion map into blocks to obtain a block root system expansion map; the block root system expansion map includes a plurality of blocks and their corresponding shooting postures; Calculate the root pixel density of each block in the block root system expansion map according to the root area probability map to obtain a root distribution density matrix; The root distribution weight of each block is calculated according to the root distribution density matrix to obtain the root distribution heat map.

4. The method according to claim 3, characterized in that The step of optimizing the standard path according to the root system heat map and generating an optimized path and a re-collection instruction includes: Obtaining a plurality of key focus blocks according to the root distribution weights in the root system heat map; Obtaining the path point parameters corresponding to the standard path according to the shooting posture corresponding to the key focus block; The path point parameters of the key focus block are refined according to the root distribution weight to obtain the optimized path and re-collection instructions; the refinement operation corresponds to shortening the step size, increasing the point viewing angle, increasing the number of Z-axis height layers and increasing the number of sampling times.

5. The method according to claim 1, wherein The step of extracting the root structure based on the received multi-view image set of the plant root system to obtain the plant root system data includes: Performing parallax restoration and depth extraction on the multi-view image set of the plant root system to obtain a sparse point cloud model of the root system; Extracting the root system structure skeleton according to the root system sparse point cloud model to obtain a root system skeleton diagram; the root system skeleton diagram includes main roots and lateral roots; Fitting the root system skeleton curve according to the root system sparse point cloud model to obtain a root system curve; Based on the camera parameters and the optimized path, the parameters of each root curve are calculated to obtain the plant root data; the plant root data includes root length, root depth, lateral extension distance, diameter and branch density.

6. The method according to claim 2, characterized in that The calculating the descriptor similarity between the key point sets corresponding to adjacent images and performing key point matching to obtain a matching pair set includes: Pairing the key points between the two key point sets corresponding to the adjacent image distributions one by one, and calculating the similarity distance of the descriptors; Based on the principle of minimum similarity distance, key point matching pairs are obtained; Extract a preset number of key point matching pairs to build a transformation model; Projecting all the key point matching pairs according to the transformation model, and calculating the projection error of each key point matching pair; If the projection error exceeds a preset threshold, the key point matching pairs are eliminated to obtain the most consistent matching pair set.

7. The method according to claim 4, characterized in that The performing of parallax restoration and depth extraction on the multi-view image set of the plant root system to obtain a sparse point cloud model of the root system includes: Extracting feature points from the multi-view image set of the plant root system and performing feature point matching to obtain matching point pairs; Extracting depth information from the multi-view image set of the plant root system according to the camera calibration and the matching point pairs to obtain three-dimensional coordinates of feature points; Performing parallax transformation on the three-dimensional coordinates of the feature points to obtain a sparse point cloud; Point cloud denoising and point cloud reconstruction are performed on the sparse point cloud to obtain the root system sparse point cloud model.

8. A plant root multi-dimensional data acquisition bracket, comprising an adjustable camera support assembly, a root clamp, and a controller, on which a computer program is stored. When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images