A method for early prediction of rice regenerative force based on X-CT and three-dimensional structured light imaging
By combining CT with three-dimensional structured light imaging technology, non-destructive testing and reconstruction of ratooning rice are performed, solving the problem of damage to rice structure caused by traditional testing methods and enabling early and accurate assessment of the regeneration capacity and yield prediction of ratooning rice.
Patent Information
- Application Number
- CN202610329354.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-03-18
- Publication Date
- 2026-06-16
AI Technical Summary
Existing technologies struggle to accurately assess the regeneration capacity of ratooned rice without damaging its external phenotype. Traditional testing methods require disrupting the rice structure, leading to inconsistent test results.
By combining computed tomography (CT) and three-dimensional structured light imaging, non-destructive testing and three-dimensional reconstruction of ratooning rice were performed. The ratooning buds were segmented and phenotypic parameters were extracted. Combined with panicle segmentation and phenotypic calculation, a yield prediction model for ratooning rice was established.
It enables early, non-destructive assessment of the regeneration capacity of ratooning rice, improves the accuracy and consistency of detection, and allows for early prediction of ratooning season yield.
Smart Images
Figure CN122222973A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to an early prediction method for rice regeneration capacity based on X-CT and three-dimensional structured light imaging. It combines non-destructive testing of early regeneration buds and tillers in the regeneration season of ratooning rice using computed tomography with three-dimensional reconstruction of the external phenotype of ratooning rice using three-dimensional structured light to obtain a method for predicting yield and judging regeneration capacity of ratooning rice at different stages. Background Technology
[0002] Rice is a staple crop for over 3 billion people worldwide and is also my country's most important food crop. Ratoon rice is a planting method that utilizes the regenerative characteristics of rice plants. Through specific cultivation techniques, dormant axillary buds on the mother plant stubble sprout after harvest, grow into panicles, and are harvested again for another season. Ratoon rice has advantages such as a high multiple cropping index, high total yield, labor and fertilizer savings, and high yield and efficiency. The regenerative capacity of ratoon rice determines the yield of the ratoon season. While ensuring the yield of the first season, improving the regenerative capacity will increase the total yield of the ratoon rice. Therefore, research on rice regenerative capacity is of great significance for improving ratoon rice varieties and increasing my country's grain output.
[0003] The regeneration capacity of ratooning rice is mainly reflected in the yield of the ratooning season compared to the first season, as shown in Formula 1. Raised buds generally have one or more dormant buds at each node on the above-ground part of the plant. Under suitable growth conditions, these buds can germinate and develop into panicles. Axillary buds are also called regenerated buds, and the second season's yield of ratooning rice comes from the germination and panicle formation of these axillary buds. Therefore, promoting the germination and growth of regenerated buds and increasing the number of effective panicles per unit area is key to high yields in the ratooning season. After the first harvest of ratooning rice, the germination rate of regenerated buds on the stubble to panicle formation is positively correlated with the yield of the ratooning season. Improving the germination capacity of axillary buds in the first season and promoting their development into panicles is crucial for increasing the number of effective panicles per unit area in the ratooning season and achieving high yields.
[0004] F R =×100% (1) F R Regeneration capacity, representing the first season's output, and the regeneration season's output. CT (Computed Tomography) is a technique that, without destroying the structure of an object, reconstructs a two-dimensional image of a specific layer of the object using projection data of certain physical quantities (such as wave velocity, X-ray intensity, electron beam intensity, etc.) acquired from its surroundings. This data is then processed by a computer using random algorithms. The technique also reconstructs a three-dimensional image from a series of these two-dimensional images. Computed tomography was first applied in the medical field. With the development of CT technology and various professional disciplines, the significant advantages of CT in material detection have led to its widespread application in non-medical fields such as industry, geophysics, engineering, agriculture, and security inspection. For example, Hughes et al. used X-ray micro-computed tomography to perform non-destructive and high-content analysis of wheat grain traits. [1] Yang et al. used X-ray diffraction to perform high-throughput non-destructive measurement of tiller number in rice. [2] Wu et al. analyzed the genetic structure of rice tillering growth through micro-CT and RGB phenotype and genome-wide association studies. [3] Shota Teramoto et al. used X-ray computed tomography to achieve high-throughput three-dimensional visualization of rice root structure. [4] .
[0005] Structured light 3D imaging is a computer-aided digital 3D imaging technique. It involves projecting a carrier frequency fringe onto the surface of an object, recording the deformed fringe image (highly modulated by the object) from another angle using an imaging device, and then digitally demodulating and reconstructing the 3D digital image of the object from the acquired deformed fringe image. Structured light 3D imaging can preserve the object's 3D spatial information (including 3D shape, grayscale, and color information) and can completely recover the object's 3D features during reconstruction. Gong et al. constructed a point cloud scanning device based on a 3D structured light projection module and designed a highly efficient, precise semantic segmentation network for rice spike point clouds called Panicle-3D. [5] Jin Qiyi et al. used the Kinect sensor to reconstruct the surface of plant point clouds and measure parameters such as leaf length. [6] .
[0006] Lin Qiang et al. measured and graded the regeneration capacity by statistically investigating the number of seedlings sprouting from 10 regenerated buds and the number of mother stems in each plot on the 7th and 14th days after the first harvest. [7]Traditional methods for detecting regenerated buds typically involve breaking the stems of growing ratooning rice and manually peeling away the outer leaf sheaths and epidermis to obtain the structural tissue of the regenerated buds. However, this method damages the external phenotype of the rice, and each sample tested is different. Therefore, CT scans and three-dimensional reconstructions can be used to detect second-season tillers and extract regenerated buds in ratooning rice. This allows for a preliminary analysis of regeneration capacity without damage in the early stages. Starting from the heading stage of the second season, regular 3D structured light imaging is performed to segment the panicles and extract their phenotypic characteristics, thereby obtaining dynamic temporal results of ratooning rice. By combining the preliminary regeneration capacity analysis performed by CT with the final phenotypic results calculated by structured light, a complete regeneration capacity analysis result of ratooning rice can be obtained non-destructively.
[0007] References Hughes A, Askew K, Scotson CP, Williams K, Sauze C, Corke F, DoonanJH, Nibau C. Non-destructive, high-content analysis of wheat grain traits using X-ray micro computed tomography. Plant Methods. 2017 Nov 1;13:76. doi:10.1186 / s13007-017-0229-8. Yang W, Xu Combining high-throughput micro-CT-RGB phenotyping and genome-wideassociation study to dissect the genetic architecture of tiller growth inrice.[J]. Journal of experimental botany, 2018. Teramoto S, Takayasu S, Kitomi Y, et al. High-throughput three-dimensional visualization of root system architecture of rice using X-raycomputed tomography. 2020. Gong L, Du X, Zhu K, et al. Panicle-3D: Efficient Phenotyping Tool for Precise Semantic Segmentation of Rice Panicle Point Cloud[J]. Plant Phenomics (English), 2021, 3(1):9. Jin Qiyi, Liu Liu, Xu Mengzhu, Ma Shang, Chen Yingduo. Reconstruction and parameter measurement of structural light point cloud surface of plants [J]. Electronic Measurement Technology, 2021, 44(03):120-124. DOI:10.19651 / j.cnki.emt.2005566. Lin Qiang, Zheng Changlin, Lin Fang, Zheng Li, Wang Hongfei, Jiang Jiahuan, Xie Zhenxing, Jiang Zhaowei, Zhang Jianfu. Creation and genetic analysis of rice germplasm with strong regeneration ability [J]. Chinese Science Bulletin, 2021, 66(02):244-252. Summary of the Invention
[0008] In order to change the existing methods for judging the regeneration ability of ratooning rice varieties, the present invention provides a system for non-destructive testing of the regeneration ability of ratooning rice in the early stage. The system obtains the yield prediction and regeneration ability of ratooning rice by combining the three-dimensional reconstruction of the growth process of ratooning rice by CT with the three-dimensional reconstruction of the external phenotype of ratooning rice by structured light. Technical solution
[0009] Step 1: Harvest the first crop of ratooning rice when the first season is ripe; Step 2: Conduct a preliminary CT experiment on ratooning rice to test suitable parameters and select appropriate current and voltage. Perform CT imaging on ratooning rice five days after the first harvest. Step 3: Regenerated Rice CT Image Regeneration Bud Segmentation: First, the tomographic image of the regenerated rice is preprocessed to remove background noise and obtain a CT image containing only the regenerated rice. Second, leaf sheath removal is performed because the regenerated buds in the regenerated rice are located between the leaf sheath and the stem, so the leaf sheath needs to be removed first. The next step is to remove the stem from the image to complete the initial segmentation of the regenerated buds. Finally, non-bud removal is performed to complete the segmentation of the regenerated buds. Step 4: Analyze the phenotypic morphological parameters and tillering of the previously segmented regenerated buds; Step 5: Perform three-dimensional structured light imaging on ratooning rice during the first ripening period and the heading stage of the ratooning season; Step 6: Segmenting rice panicles and calculating phenotypic parameters: The 3D point cloud of ratooning rice after heading is segmented using PointNet++ network to obtain the clustering and segmentation results of single panicle point cloud, and the phenotypic parameters of single panicle are calculated. Step 7: Complete the time-series-based 3D structured light point cloud growth sequence of ratooned rice to obtain the phenotypic traits of ratooned rice panicles.
[0010] Step 8: Combine the CT-extracted regenerated buds and tillers of the regenerated rice with the 3D point cloud of the regenerated rice structured light to obtain the yield prediction and regeneration capacity model of the regenerated rice. Attached Figure Description
[0011] Figure 1 This is a flowchart of the overall technical solution; Figure 2 Flowcharts for CT processing and 3D structured light processing of ratooned rice; Figure 3 Flowchart for CT removal of stems and leaf sheaths in ratooned rice; Figure 4 A diagram illustrating the effect of CT segmentation on regenerated buds in ratooning rice; Figure 5 This is a CT tomographic image of regenerated rice. Figure 6 A time-series rendering of structured light segmentation of rice panicles in regenerated rice. Detailed Implementation
[0012] Step A: First harvest. When the first crop matures, the ratooning rice is harvested for the first time, at a height of about 40cm from the ground. Step B: Conduct a CT pre-experiment on the ratooning rice, complete the control experiment by selecting different current and voltage, select the optimal imaging parameters by comparing with the CT tomographic images, and adjust the appropriate imaging field of view. See steps C-E. Figure 2 Processing flow of CT tomographic images of left-side regenerated rice Step C: A CT scan was performed on the ratooning rice five days after the first harvest. A fast circumferential scan mode was used, with each sample rotated 360° at a step angle of 0.25°, resulting in 1440 projected images at a frame rate of 6 frames / second. After scanning, reconstruction was performed using the system's built-in 3D reconstruction software CERA, taking approximately 10 minutes. The reconstruction results for each sample were saved as 2880 tomographic images, stored in unsigned 16-bit RAW format, and opened as VGL files. The tomographic images can be exported using VGStudio software. The resolution and slice interval were 0.0619 mm x 0.0619 mm x 0.0619 mm. The imaging volume was 177.029 mm x 177.029 mm x 177.0289 mm. See the CT reconstruction tomographic image of the ratooning rice. Figure 5 ; Step D: Segmenting regenerated buds from CT images of regenerated rice, and the flowchart for segmenting regenerated rice stems and leaf sheaths is shown below. Figure 3 ① Image preprocessing: Since there is a large difference in grayscale values between the background and the regenerated rice, the Otsu thresholding method can be used for automatic segmentation to remove the background and retain the target. ② Leaf sheath removal: Since the leaf sheath is relatively small, morphological opening operations are used. The kernel is set to 3×3. First, an opening operation with 2 iterations is performed, then an opening operation with 3 iterations is performed. Then, the area of the small region is removed by setting an area threshold of 300. Noise is removed (the small area of the region can be removed by finding all connected components in the graph, deleting connected components in the vector container that are larger than the area threshold by setting an area threshold, setting the remaining small area outline pixels to 0, and setting the pixels in the outline to 0 by using the drawContours function). Finally, the leaf sheath part can be removed. ③ Stem Removal: The binary image of the stem's internal contour, after dilation, is used as a mask to initially remove the stem. All contours (including the inner and outer contours of the stem, and the contours of specks) are searched on the binary CT image of the ratooning rice. There are two special cases for stem removal: the outer contour of a small stem is smaller than the inner contour of a large stem; and the stem is solid.
[0013] Normal stems can be removed by statistically analyzing the area of all contours. The area threshold for the outer contour of normal stems is determined to be above 10,000, while the area threshold for the contour of impurities is below 500. By judging the contours within the determined area threshold range, the purpose of removing the outer contour of normal stems and impurities can be achieved.
[0014] For thin stems, since the outer contour of a thin stem contains an inner contour, we can determine whether a stem is the outer contour of a thin stem by checking if any contour meeting the area threshold contains an inner contour, thus achieving the goal of deleting the outer contour of the thin stem. This yields all the inner contours of the stem. Using `drawContours`, the contours are marked in white. By setting the kernel to 3×3 and performing a dilation operation with 16 iterations, a mask image is obtained. This mask is then compared with the original image to obtain a rough tomographic image of the stem after removal. Since there are still noise points, morphological opening and small region removal are required. Small region removal can be achieved by using the connected region area method described above to remove small regions that do not belong to the regenerated bud.
[0015] Since some stems of the whole ratooning rice plant are solid stems, it is impossible to obtain their internal outlines. Therefore, it is impossible to remove such stems. On the basis of roughly removing the stems, it is necessary to perform area threshold removal to remove the outlines of those that do not belong to the regenerated buds, and then complete a masking operation to remove all stems. ④ The above operations are all two-dimensional tomographic image processing. Since noise points in the three-dimensional plane cannot be removed, small-volume connected component removal based on the three-dimensional 26-connected components is required. First, all two-dimensional tomographic images are placed in a vector container and stored as three-dimensional voxel information. For each point, it is determined whether there are non-zero pixels among the surrounding 26 connected components. If so, the connected component of that point is placed in the vector. <vector <point3i>After scanning all tomographic maps, a container with its pixel values set to 0 is created. This yields containers for all 3D connected components. The size of a single container within these containers represents the volume of each connected component. A threshold of 300,000 is used. (Because copying a Mat type vector becomes a shallow copy, after completing all tomographic scans, all voxel pixel values are 0, requiring another 3D voxel information storage of the original image.) The newly created 3D voxel information is processed, and vectors with values less than the threshold are processed. <point3i>The points inside were changed to 0, and the small-volume connected component deletion operation was finally completed within the 3D voxel. The final 3D voxel information of the regenerated bud was obtained. The effect of segmenting the regenerated bud is shown in [link to voxel]. Figure 4 ; Step E: For the voxel information of the regenerated bud segmentation completed above, isosurface extraction and reconstruction are performed using the vtkMarchingCubes face rendering algorithm. Subsequently, the reconstructed regenerated buds are processed by vtkStripper and tkSmoothPolyDataFilter for smoothing filtering. Finally, the voxels are converted to .ply files. First, point cloud downsampling is performed. After downsampling, Euclidean clustering is used for classification. The search radius setClusterTolerance is set to 0.01, and the number of point clouds setMinClusterSize and setMaxClusterSize are set to 3000 and 30000 respectively. Finally, PCA principal component analysis is used to calculate the length, width, and height phenotypic values of each regenerated bud.
[0016] Step F: 3D structured light photography. 3D structured light photography is performed on the ratooning rice panicles after the first crop matures and after harvest, during the heading stage. Only one photography session is taken after maturity, and an average of once every three days is taken after harvest. The target is located approximately to the left and right of the first crop's harvest position. Photography method: The ratooning rice pot is placed on a turntable. The 3D structured light is set to a fixed scanning mode, and the distance between the structured light camera and the ratooning rice is fixed at approximately 30cm. The turntable is rotated by a fixed 15° increment each time. After the ratooning rice panicles come to a complete stop, structured light photography is performed. After one full rotation, structured light photography at a single height is completed. The remaining heights can be completed by adjusting the structured light camera. After photography at all heights is completed, the data is saved as a point cloud file using a non-closed model.
[0017] Step G: Harvest the mature ratooning rice panicles, thresh them, and test the seed yield using a seed testing machine to obtain the actual yield of each ratooning rice plant. Calculate the actual ratooning capacity of the ratooning rice variety based on the first season's yield.
[0018] See steps H-K Figure 2 Right side structured light point cloud processing flow of regenerated rice Step H: Perform pre-segmentation of the rice panicles in the 3D point cloud of ratooning rice captured by structured light. First, convert the ply file to a pcd point cloud file using pcl. Since the number of directly captured ratooning rice point clouds is large, it is necessary to perform VoxelGrid voxel downsampling processing on the pcd point cloud file to reduce the amount of ratooning rice point cloud data while preserving the shape features of the point cloud. Set the setLeafSize parameter in VoxelGrid downsampling to 1.5f. Since there is a color difference between the rice panicles and leaves of ratooning rice, color-based region growing can be used to perform preliminary segmentation of the rice panicles. Color-based region segmentation can identify point clouds with similar colors and within a certain distance. To identify points of the same type, the point cloud is first segmented. Seed points and their surrounding areas with color differences less than a threshold are considered as a single cluster. Then, a merging operation is performed, combining clusters with color differences less than a threshold into a single cluster. The distance threshold `setDistanceThreshold` is set to 0.2, the color threshold between points `setPointColorThreshold` to 8, and the color threshold between clusters `setRegionColorThreshold` to 1. The minimum number of points in each cluster `setMinClusterSize` is set to 800. This yields a preliminary segmentation of the ratooning rice panicle, reducing the cost of manual annotation. After the preliminary segmentation of the ratooning rice panicle point cloud, CloudCompare is used for manual point cloud annotation, creating a 3D ratooning rice panicle point cloud dataset. Using the PonitNet++ network in mmdetection3D and modifying the config file, semantic segmentation of the ratooning rice panicle is performed. The ratooning rice point cloud segmentation image is shown below. Figure 6 As shown.
[0019] Step 1: Process all the point clouds of rice panicles after each ratooning rice plant is segmented. Use the conditional Euclidean clustering algorithm to segment the ratooning rice panicles into individual rice plants and panicle counts. Set the search radius setClusterTolerance to 0.01, and setMinClusterSize and setMaxClusterSize to be set according to the number of point clouds. Use statistical filtering to remove noise points and outliers from the extracted individual rice panicle point clouds.
[0020] Step J: After completing the Euclidean clustering segmentation of each ratooning rice plant, individual panicle point clouds can be obtained. By performing PCA principal component analysis and VTK triangular patch reconstruction on the individual panicle point clouds, phenotypic traits such as panicle biomass, panicle number, and panicle volume are extracted.
[0021] Step K: Arrange the ratooned rice and the separated rice panicles that have been photographed by structured light in chronological order to complete the ratooned rice time series diagram. Arrange the individual rice panicles of the ratooned rice ratooning season in chronological order based on each shooting time point, and calculate the phenotypic values such as the number of panicles per ratooned rice plant, panicle volume, and biomass calculated in the previous steps.
[0022] Step L: Since rice panicles develop from regenerated buds, the phenotype of regenerated buds and the early regeneration ability of regenerated bud tillers obtained by CT can be fused and analyzed with the rice panicle results extracted by 3D structured light point cloud.
[0023] The number of regenerated shoots five days after the first harvest of ratooning rice based on CT was analyzed with the actual yield of ratooning rice in the ratooning season, so as to achieve early yield prediction of ratooning rice. Based on 3D structured light, the rice panicles of ratooning rice are segmented at maturity to obtain phenotypic parameters such as the number and volume of panicles at each mature plant. These phenotypic parameters are then used in a regression model analysis with the actual yield of ratooning rice during the ratooning season to predict the yield of ratooning rice in the later stages. Additionally, point clouds of ratooning rice from the first season are segmented to calculate phenotypic parameters such as the volume of the panicles. These parameters are then combined with the panicle phenotypes calculated at maturity in the ratooning season to obtain the later-stage regeneration capacity of ratooning rice, as shown in the following formula: F R =×100% (2) F R Regeneration capacity refers to the volume of the first-season mature rice panicle and the volume of the rice panicle at maturity in the second-season. The early-stage regeneration capacity of ratooning rice is defined as the ratio of the number of regenerated shoots sprouting five days after the first harvest to the total number of stems in the ratooning rice, as shown in the following formula: F R =×100% (3) F R Regeneration capacity of regenerated buds, number of regenerated buds sprouting five days after the first harvest of ratooning rice, and total number of stems. Therefore, the early regeneration capacity of ratooning rice can be obtained by calculating the number of regenerated shoots sprouting five days after the first harvest and the total number of stems per ratooning rice plant. Model analysis of ratooning rice regeneration capacity and non-destructive testing to obtain actual regeneration capacity in the early and late stages of ratooning rice production are then completed.
[0024] More specifically, in step B, the pre-experimental current and voltage are selected as 75KV, 150μA, without adding a filter.
[0025] More specifically, in step B, the CT imaging instrument is a YXLON-FF35CT with a detector size of 43×43 and a high-voltage generator of 225KV.
[0026] More specifically, in step F, the instrument used for 3D structured light imaging of regenerated rice is the Wiiboox-Reeyee Pro 2X (including a texture camera).
[0027] More specifically, in step G, the rice grain testing instrument is the Gufeng Optoelectronics YTS-5D digital rice testing machine. Contents not described in detail in this specification are prior art known to those skilled in the art. Although illustrative specific embodiments of the invention have been described above to facilitate understanding by those skilled in the art, it should be understood that the invention is not limited to the scope of the specific embodiments. Various modifications are readily apparent to those skilled in the art as long as they fall within the spirit and scope of the invention as defined and determined by the appended claims, and all inventions utilizing the concept of this invention are protected.
Claims
1. A method for early prediction of rice regeneration capacity based on X-CT and three-dimensional structured light imaging, characterized by combining non-destructive testing of early regenerated buds and tillers in the regeneration season using computed tomography with three-dimensional reconstruction of the external phenotype of regenerated rice using three-dimensional structured light imaging to obtain a method for judging the regeneration capacity of regenerated rice, wherein... This system includes CT segmentation and non-destructive testing of early-stage regenerated buds in ratooning rice; structured light extraction and segmentation of panicles during the booting stage and after maturity in ratooning rice; and determination of the regeneration capacity of ratooning rice based on phenotypic results extracted by CT and structured light. The system can perform the following methods: Step 1: Harvest the first crop of ratooning rice when the first season is ripe; Step 2: Conduct a preliminary CT experiment on ratooning rice to test suitable parameters and select appropriate current and voltage. Perform CT imaging on ratooning rice five days after the first harvest. Step 3: Regenerated Rice CT Image Regeneration Bud Segmentation: First, the tomographic image of the regenerated rice is preprocessed to remove background noise and obtain a CT image containing only the regenerated rice. Second, leaf sheath removal is performed because the regenerated buds in the regenerated rice are located between the leaf sheath and the stem, so the leaf sheath needs to be removed first. The next step is to remove the stem from the image to complete the initial segmentation of the regenerated buds. Finally, non-bud removal is performed to complete the segmentation of the regenerated buds. Step 4: Analyze the phenotypic morphological parameters and tillering of the previously segmented regenerated buds; Step 5: Perform three-dimensional structured light imaging on ratooning rice during the first ripening period and the heading stage of the ratooning season; Step 6: Segmenting rice panicles and calculating phenotypic parameters: The 3D point cloud of ratooning rice after heading is segmented using PointNet++ network to obtain the clustering and segmentation results of single panicle point cloud, and the phenotypic parameters of single panicle are calculated. Step 7: Complete the time-series-based 3D structured light point cloud growth sequence of ratooning rice to obtain the phenotypic traits of ratooning rice panicles; Step 8: Combine the CT-extracted regenerated buds and tillers of the regenerated rice with the 3D point cloud of the regenerated rice structured light to obtain the yield prediction and regeneration capacity model of the regenerated rice.
2. The CT segmentation and non-destructive testing of early regenerated buds in the regeneration season of ratooning rice according to claim 1, characterized in that: The CT scan used a current and voltage of 75KV, 150μA without a filter. After circumferential scanning of the rice, projection images were obtained. XY-plane tomographic images were then reconstructed from these projection images. Automatic segmentation using the OTSU thresholding method with OpenCV was performed to remove the background and retain the regenerated rice. Morphological operations were then used to remove small-area impurities. Functions such as `drawContours` were then used to remove leaf sheaths, solid stems, and thin stems. Finally, the tomographic images were stacked into a 3D model, and 3D connected component analysis was performed to remove connected components smaller than a threshold, obtaining the final 3D voxel information of the regenerated shoots. Isosurface extraction and reconstruction were performed using a surface rendering algorithm. The reconstructed regenerated shoots were then processed using smoothing filtering. Finally, the voxels were converted to .ply files, and point cloud downsampling was performed. After downsampling, Euclidean clustering was used for classification. Finally, PCA principal component analysis was used to calculate the length, width, and height phenotypes of each regenerated shoot.
3. The structured light method for extracting and segmenting rice panicles during the booting stage and after maturity in ratooning rice according to claim 1, characterized in that: Three-dimensional structured light imaging of ratooning rice panicles was conducted after the first crop matured and after harvest, during the heading stage. After maturity, only one imaging session was taken; after harvest, the imaging cycle was averaged once every three days. The imaging target was located approximately 30cm from the first crop harvest location. After imaging, the data was saved as a point cloud file using a non-closed model. The ply file was converted to a pcd point cloud file using pcl, and then the point cloud was downsampled. Initial segmentation of the ratooning rice panicles was performed using color-based region segmentation, followed by manual point cloud annotation using CloudCompare to create a three-dimensional ratooning rice panicle point cloud dataset. The ratooning rice panicle segmentation was trained and inferred using the 3D deep learning network PonitNet++. Finally, Euclidean clustering was performed on each plant to obtain individual panicle point clouds, along with data on volume, number of panicles, and other ratooning characteristics.
4. The method for determining the regeneration ability of regenerated rice based on phenotypic extraction using CT and structured light as described in claim 1, characterized in that: For the harvested first-season and ratooning season ratooning rice panicles, threshing and seed testing were performed to obtain the actual yield per ratooning rice plant. A regression model analysis was conducted using CT-based data on the number of ratooning shoots five days after the first-season harvest and the actual yield of the ratooning season ratooning rice, thus achieving early yield prediction for ratooning rice. The early-stage regeneration capacity of ratooning rice is defined as the ratio of the number of ratooning shoots sprouting five days after the first-season harvest to the total number of ratooning rice stalks, as shown in the following formula: F R = ×100% （1） F R Regeneration ability of regenerated buds, The number of regenerated rice buds sprouting five days after the first harvest. Total number of stems Based on three-dimensional structured light, the rice panicles of ratooning rice are segmented at the maturity stage of the ratooning season. Phenotypic parameters such as the number and volume of rice panicles at each maturity stage are obtained. The phenotypic parameters are then used to perform regression model analysis with the actual yield of ratooning rice in the ratooning season to complete the subsequent prediction of ratooning rice yield. Point clouds captured from the first crop of ratooning rice were segmented into panicles to calculate phenotypic parameters such as panicle volume. These parameters were then combined with the panicle phenotypes calculated at maturity in the ratooning season to obtain the later-stage regeneration capacity of ratooning rice. The formula is as follows: F R = ×100% （2） F R regeneration ability, This refers to the volume of the first mature rice ear. This refers to the volume of rice panicles at the ripening stage of the regeneration season.