Corn plant point cloud organ segmentation method fusing dynamic weight and feature enhancement

Abstract

The invention discloses a corn plant point cloud organ segmentation method fusing dynamic weight and feature enhancement, and the method comprises the following steps: (1) inputting corn plant point cloud data, including the space coordinates and additional features of the point cloud; (2) carrying out point cloud sampling and aggregation on the input point cloud data to obtain sampling point cloud data, and carrying out point cloud grouping by combining multi-scale clustering, namely carrying out sphere query through different radiuses; (3) carrying out feature extraction by using a DWE network; the DWE network is formed by fusing a ScoreNet dynamic weight generation module and a feature enhancement module, a dynamic weight is generated through the ScoreNet, and weighted aggregation is performed on the feature enhanced point cloud according to the dynamic weight; (4) inputting the point cloud and the features corresponding to the point cloud into a full connection layer, and segmenting semantic information of the point cloud; according to the invention, high-quality three-dimensional point cloud data is provided for phenotypic research of corn plants.

Description

Technical Field

[0001] The present invention relates to the technical field of computer vision and point cloud processing, and in particular to a method for segmenting corn plant point cloud organs by integrating dynamic weights and feature enhancement. Background Art

[0002] Corn is one of the most important food crops in the world, and the growth of its plants directly affects the yield and quality in the later stage. Therefore, accurate identification and segmentation of corn plant organs are of great significance. In recent years, point cloud data has been widely used in crop phenotyping due to its rich three-dimensional spatial information. The acquisition of corn plant organ point cloud data is of great significance in agricultural automation, especially in precision agriculture. However, the corn plant point cloud data is not only affected by environmental and noise factors during the acquisition process, but also by the overlap of leaves during the growth of corn plant leaves.

[0003] Traditional point cloud segmentation algorithms have difficulty in accurately extracting the structural features of the target organ, resulting in low segmentation accuracy. Existing point cloud segmentation methods based on deep learning only solve the problem of point cloud segmentation in a general way, and cannot accurately segment the organs of corn plant point clouds. Summary of the invention

[0004] Purpose of the invention: The purpose of the present invention is to provide a method for organ segmentation of corn plant point clouds that integrates dynamic weights and feature enhancement, so as to solve the problem of point cloud segmentation and the problem of being unable to accurately segment organs of corn plant point clouds.

[0005] Technical solution: The corn plant point cloud organ segmentation method integrating dynamic weight and feature enhancement described in the present invention comprises the following steps:

[0006] (1) Input corn plant point cloud data, including the spatial coordinates and additional features of the point cloud;

[0007] (2) Perform point cloud sampling and aggregation on the input point cloud data to obtain sampled point cloud data, and combine multi-scale clustering, that is, perform sphere queries with different radii to group the point clouds;

[0008] (3) Use the DWE network for feature extraction; the DWE network is composed of the ScoreNet dynamic weight generation module and the feature enhancement module. The dynamic weight is generated by ScoreNet, and the feature enhanced point cloud is weighted and aggregated according to the dynamic weight;

[0009] (4) The point cloud and its corresponding features are input into the fully connected layer and the semantic information of the point cloud is segmented.

[0010] Furthermore, in step (2), the sampling points are obtained by using density-based farthest point sampling, including the following steps:

[0011] (21) Receive point cloud data and set the number of sampling points n point The number of neighbors k required for density calculation, an array for storing the sampling point index and an array for recording the distance from each point to the nearest sampling point;

[0012] (22) Calculate the Euclidean distance matrix between any two points in the point cloud, find the k nearest neighbor points and corresponding distances for each point, and use the inverse of the average nearest distance of each point as the density weight;

[0013] (23) After obtaining the density weight of each point, a point in the point cloud is randomly selected as the first sampling point. In order to ensure that all points are likely to be selected when the farthest point is selected for the first time, we initialize the distance from all points in the point cloud to the sampling point to 10 10 ;

[0014] (24) Among the currently unselected points, select the point with the largest weighted distance to the existing sampling point set as the new sampling point, calculate the distances of all points to the current sampling point and update the closest distance of each point to the selected sampling point set, retaining only the smallest distance, and perform n point Iterative sampling; the distance between sampling points is obtained according to the density weight;

[0015] (25) Output the final selected n point The index of the sampling point is used to complete the sampling.

[0016] Furthermore, in step (3), the ScoreNet module constituting the DWE network generates dynamic weights according to the input point cloud data features and additional features for weighted processing of local point cloud features. The specific steps are as follows:

[0017] (31) The ScoreNet module generates a dynamic weight for each point based on the local features of the point cloud input, indicating the importance of the local area in the segmentation process, with a value between 0 and 1;

[0018] (32) Use the generated dynamic weights to perform weighted aggregation on the point cloud data after sampling and grouping, amplifying the contribution of important features to the results, and vice versa;

[0019] Furthermore, in step (3), the feature enhancement module of the DWE network is used to enhance the features of the point cloud data by the following steps:

[0020] (S31) obtaining local aggregated point cloud data from the sampled point cloud data;

[0021] (S32) extracting neighborhood point features from the global point cloud;

[0022] (S33) using the broadcast mechanism, broadcasting the sampling point features, constructing a point cloud feature matrix, forming a broadcast feature with the same dimension as the neighborhood point feature, and calculating the difference gcn between the neighborhood point feature and the broadcast feature, where gcn = groupedpoints-pointstile;

[0023] (S34) splicing the local point feature grouped_points, the multiple copies of the point feature points_tile and the difference feature gcn to obtain an enhanced local point cloud feature grouped_points_agg;

[0024] (S35) The enhanced local point cloud feature grouped_points_agg is concatenated with the local geometric information grouped_xyz to obtain a final feature representation.

[0025] The corn plant point cloud organ segmentation system integrating dynamic weight and feature enhancement described in the present invention comprises:

[0026] Point cloud data module: used to input corn plant point cloud data, including the spatial coordinates and additional features of the point cloud;

[0027] Sampling module: used to sample and aggregate the input point cloud data to obtain sampled point cloud data, and combine multi-scale clustering, that is, perform sphere queries with different radii to group the point clouds;

[0028] DWE network module: used to extract features using the DWE network; the DWE network is a fusion of the ScoreNet dynamic weight generation module and the feature enhancement module. The dynamic weight is generated by ScoreNet, and the feature-enhanced point cloud is weighted and aggregated according to the dynamic weight.

[0029] Segmentation module: used to input the point cloud and its corresponding features into the fully connected layer and segment the semantic information of the point cloud.

[0030] Furthermore, in the sampling module, the sampling points are obtained using density-based farthest point sampling, as follows:

[0031] Receive point cloud data and set the number of sampling points n point The number of neighbors k required for density calculation, an array for storing the sampling point index and an array for recording the distance from each point to the nearest sampling point;

[0032] Calculate the Euclidean distance matrix between any two points in the point cloud, find the k nearest neighbor points and corresponding distances for each point, and use the inverse of the average nearest distance of each point as the density weight;

[0033] After obtaining the density weight of each point, randomly select a point in the point cloud as the first sampling point. In order to ensure that all points are likely to be selected when the farthest point is selected for the first time, we initialize the distance from all points in the point cloud to the sampling point to 10 10 ;

[0034] Among the currently unselected points, select the point with the largest weighted distance to the existing sampling point set as the new sampling point, calculate the distance of all points to the current sampling point and update the shortest distance of each point to the selected sampling point set, retaining only the smallest distance, and perform n point Iterative sampling; the distance between sampling points is obtained according to the density weight;

[0035] Output the final selected n point The index of the sampling point is used to complete the sampling.

[0036] In the further DWE network module, the ScoreNet module constituting the DWE network generates dynamic weights according to the input point cloud data features and additional features for weighted processing of local point cloud features. The specific steps are as follows:

[0037] The ScoreNet module generates a dynamic weight for each point based on the local features of the point cloud input, indicating the importance of the local area in the segmentation process, with a value between 0 and 1;

[0038] The generated dynamic weights are used to perform weighted aggregation on the sampled and grouped point cloud data, amplifying the contribution of important features to the results, and vice versa;

[0039] Furthermore, in the DWE network module, the feature enhancement module constituting the DWE network performs feature enhancement on the point cloud data through the following steps:

[0040] Obtaining local aggregated point cloud data from the sampled point cloud data;

[0041] Extract neighborhood point features from the global point cloud;

[0042] Using the broadcast mechanism, the sampling point features are broadcasted and transmitted to construct the point cloud feature matrix, forming broadcast features with the same dimension as the neighborhood point features, and calculating the difference gcn between the neighborhood point features and the broadcast features, where gcn = groupedpoints-pointstile;

[0043] The local point feature grouped_points, multiple copies of the point feature points_tile and the difference feature gcn are spliced ​​to obtain the enhanced local point cloud feature grouped_points_agg;

[0044] The enhanced local point cloud feature grouped_points_agg is concatenated with the local geometric information grouped_xyz to obtain the final feature representation.

[0045] An electronic device described in the present invention includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is loaded into the processor, it implements any one of the methods for corn plant point cloud organ segmentation integrating dynamic weights and feature enhancement.

[0046] A storage medium described in the present invention stores a computer program, and when the computer program is executed by a processor, it implements any one of the methods for corn plant point cloud organ segmentation that integrates dynamic weights and feature enhancement.

[0047] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: density-weighted farthest point sampling is used in point cloud sampling to reduce the impact of noise and density unevenness, and effectively improve the coverage effect of sampling points. By fusing dynamic weights and feature enhancement, a DWE feature extraction network that is more suitable for corn plant point cloud segmentation is constructed, which greatly improves the segmentation accuracy of corn plant point clouds. Using this method for corn plant point cloud segmentation can obtain a more accurate corn plant point cloud segmentation example, providing high-quality three-dimensional point cloud data for phenotypic research of corn plants. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 is a flow chart of the present invention;

[0049] Figure 2 It is a comparison diagram of the sampling methods of the present invention;

[0050] Figure 3 This is a structural diagram of the feature enhancement module of the present invention;

[0051] Figure 4 It is a structural diagram of the ScoreNet weighted module of the present invention;

[0052] Figure 5 This is a comparison diagram of the segmentation results of the present invention. DETAILED DESCRIPTION

[0053] The technical solution of the present invention is further described below in conjunction with the accompanying drawings.

[0054] like Figure 1As shown, an embodiment of the present invention provides a three-dimensional point cloud segmentation method integrating dynamic weights and feature enhancement. Based on the input point cloud data, multi-scale group learning is performed through a feature learning network. The learned features are propagated through features to obtain the score of each point, thereby achieving the point cloud segmentation task. The difference between the present invention and the original PointNet++ network lies in the construction of the feature learning network. Through density-based farthest point sampling and DWE feature extraction network, a feature learning network that is more suitable for corn plant point cloud segmentation is constructed.

[0055] a1. Density-based farthest point sampling

[0056] The original PointNet++ network uses the farthest point sampling method in the feature learning network, and only focuses on the farthest distance between points for point cloud downsampling. In complex scenes, it will be affected by the uneven distribution of point cloud density and outliers, resulting in a decrease in sampling quality. The present invention adopts density-based farthest point sampling, and the specific operations are as follows:

[0057] a1.1 Based on a given point cloud P = {p 1 ,p 2 ,···,p N}, the coordinates of each point are p i =(x i ,y i ,z i ), set the number of sampling points n point As well as the number of neighbors k for density calculation, initialize the array controids to store the sampling points and the array distance to record the nearest distance of each point.

[0058] a1.2 Calculate the Euclidean distance between two points. The formula is:

[0059]

[0060] For each point, find its k nearest neighbor points and find the corresponding Euclidean distance set: {d i1 ,d i2 ,···,d ik The inverse of the average nearest distance of each point is taken as the density weight of the point, where the density weight ρ(p i ) is defined as Density weight ρ(p i ) is larger, indicating that p i The higher the density of the area.

[0061] a1.3 Randomly select a point as the first sampling point c 1 , the initial distance D(p i ) is set to a larger value, for example: D(pi )=∞, In each iteration (n point times), select the point c with the largest (weighted) distance between the current unsampled point and all sampled points k , the formula is:

[0062]

[0063] Where C=(C 1 C 2 ,···,C k-1 ) is the set of selected sampling points. For each unsampled point p i , update its closest distance D(p i ), whose formula is:

[0064]

[0065] a1.4 Output selection n point The index array controids of the sampling points.

[0066] like Figure 2 As shown in the figure, comparing the two different sampling methods, the farthest point sampling based on density can not only effectively avoid the influence of outliers by considering the density, but also optimize the uniformity of sampling, effectively ensuring that more representative points are selected in the area.

[0067] a2.DWE feature extraction network

[0068] The feature extraction module in the feature learning network, in the original PointNet++ network, uses the PointNet layer to extract and aggregate features in the space composed of sampling points and neighborhood points, only considering the relative position between the sampling points and the corresponding neighborhood points, and the extraction of local feature information is not sufficient. The present invention uses the DWE (Dynamic Weighting and Enhancement) module for feature extraction. The DWE module is formed by the fusion of the ScoreNet dynamic weighting network and the feature enhancement network, wherein the feature enhancement module is as follows Figure 2 shown.

[0069] For the input point cloud data, the relationship between the neighborhood point features and the sampling point features is comprehensively considered, the sampling point features are transmitted through the broadcast mechanism to obtain the broadcast features, and the difference between the neighborhood point features and the broadcast features is calculated to obtain the geometric features. The fused features are obtained by splicing the neighborhood point features, broadcast features and geometric features, which expands the feature dimension and increases the feature expression ability. Finally, the fused features are spliced ​​with the relative coordinates to obtain the final feature representation.

[0070] SoreNet is a fully connected (MLP) network used to calculate dynamic weights. This dynamic weight is used to adjust the influence of different neighborhood points to better aggregate neighborhood features. Specifically, the structure of ScoreNet is as follows:

[0071] Input layer: The input point cloud shape is [B, N, K, D], which represents the features of each sampling point and its neighborhood points. B is the batch size, N is the number of sampling points, indicating that N representative points are selected from the point cloud; K is the number of neighborhood points of each sampling point, indicating that there are K neighborhood points around each sampling point; D is the feature dimension of each point.

[0072] Fully connected layer: Feature extraction is performed through the fully connected layer (MLP) of ScoreNet, which includes multiple fully connected layers. The number of layers and the number of neurons in each layer are determined by score_mlp, which is the number of units in the MLP layer specified when the network is defined. In SA1, score_mlp = [64, 128, 256]. In SA2, score_mlp = [128, 256, 512]. Specifically, score_mlp is a list that specifies the number of neurons in each layer. For example, if score_mlp =

[0073] [64,128,256], then ScoreNet will have three fully connected layers, containing 64, 128 and 256 neurons respectively. In each layer, the fully connected layer performs a linear transformation on the input features and a nonlinear transformation through the activation function (ReLU). Output layer: After multiple layers of fully connected layers, the output of ScoreNet is usually the weight of each neighborhood point, which affects the contribution of the neighborhood point in feature aggregation. The output shape is [B,N,K,1], that is, each neighborhood point of each sampling point has a weight value. After feature enhancement, the feature dimension of the input point cloud is increased from the original 3+D in Expanded to 3+3*D in , the corresponding dynamic weight generation process is as follows Figure 3 shown.

[0074] The weights calculated by ScoreNet are the contribution of each neighborhood point to its corresponding sampling point. Each sampling point corresponds to a set of weights, indicating the importance of its neighborhood points to the features of the sampling point. The obtained weights (in the shape of [B, N, K, 1]) are weighted with the features of each neighborhood point in the local features. Specifically, for each sampling point and its neighborhood points, the local features are multiplied by the calculated weight value to obtain the weighted features.

[0075] Assuming that after ScoreNet calculation, a weight tensor of shape [B, N, K, 1] is obtained, the weighted features of each sampling point and its neighborhood points can be expressed as:

[0076] Weighted feature = local feature × weight tensor. (4)

[0077] Among them, the shape of the local feature is [B, N, K, D], and the shape of the weight tensor is [B, N, K, 1], which will weight the features of each neighborhood point.

[0078] The weighted features are further processed by convolution operations. Finally, after maximum pooling aggregation, a feature representation with a higher level of abstraction is obtained for each sampling point.

[0079] Segmentation experiments were conducted on the corn 3D point cloud instance segmentation dataset (syau single maize) and compared with PointNet and PointNet++. Figure 5 As shown, (a), (b), (c), and (d) are the segmentation effect diagrams of PointNet, PointNet++_SSG, PointNet++_MSG, and the present invention, respectively. It can be seen that PointNet has the worst segmentation effect, with incorrect segmentation areas on all three leaves, and both versions of PointNet++ have deficiencies in details in the overlapping areas of leaves and areas of uneven density. The method of the present invention has the most perfect processing of the junctions between leaves and leaves, and leaves and stems. The specific segmentation accuracy is shown in Table 1. The experimental results show that compared with PointNet and PointNet++, the method proposed in the present invention has significantly excellent performance in organ segmentation of corn plant point clouds.

[0080] Table 1 Segmentation results on the Syau Single Maize dataset

[0081]

[0082] In summary, the present invention improves on the original PointNet++ model by using density-based farthest point sampling and a DWE feature extraction module that integrates dynamic weights and feature enhancement in the feature learning network, thereby improving the network performance and making it more suitable for organ segmentation of corn plant point clouds.

[0083] The corn plant point cloud organ segmentation system integrating dynamic weight and feature enhancement described in the present invention comprises:

[0084] Point cloud data module: used to input corn plant point cloud data, including the spatial coordinates and additional features of the point cloud;

[0085] Sampling module: used to sample and aggregate the input point cloud data to obtain sampled point cloud data, and combine multi-scale clustering, that is, to group point clouds by sphere queries with different radii; use density-based farthest point sampling to obtain sampling points, as follows:

[0086] Receive point cloud data and set the number of sampling points n point The number of neighbors k required for density calculation, an array for storing the sampling point index and an array for recording the distance from each point to the nearest sampling point;

[0087] Calculate the Euclidean distance matrix between any two points in the point cloud, find the k nearest neighbor points and corresponding distances for each point, and use the inverse of the average nearest distance of each point as the density weight;

[0088] After obtaining the density weight of each point, randomly select a point in the point cloud as the first sampling point. In order to ensure that all points are likely to be selected when the farthest point is selected for the first time, we initialize the distance from all points in the point cloud to the sampling point to 10 10 ;

[0089] Among the currently unselected points, select the point with the largest weighted distance to the existing sampling point set as the new sampling point, calculate the distance of all points to the current sampling point and update the shortest distance of each point to the selected sampling point set, retaining only the smallest distance, and perform n point Iterative sampling; the distance between sampling points is obtained according to the density weight;

[0090] Output the final selected n point The index of the sampling point is used to complete the sampling.

[0091] DWE network module: used to extract features using the DWE network; the DWE network is a fusion of the ScoreNet dynamic weight generation module and the feature enhancement module. The dynamic weight is generated by ScoreNet, and the feature-enhanced point cloud is weighted and aggregated according to the dynamic weight. The specific steps are as follows:

[0092] The ScoreNet module generates a dynamic weight for each point based on the local features of the point cloud input, indicating the importance of the local area in the segmentation process, with a value between 0 and 1;

[0093] The generated dynamic weights are used to perform weighted aggregation on the sampled and grouped point cloud data, amplifying the contribution of important features to the results, and vice versa;

[0094] Furthermore, in the DWE network module, the feature enhancement module constituting the DWE network performs feature enhancement on the point cloud data through the following steps:

[0095] Obtaining local aggregated point cloud data from the sampled point cloud data;

[0096] Extract neighborhood point features from the global point cloud;

[0097] Using the broadcast mechanism, the sampling point features are broadcasted and transmitted to construct the point cloud feature matrix, forming broadcast features with the same dimension as the neighborhood point features, and calculating the difference gcn between the neighborhood point features and the broadcast features, where gcn = groupedpoints-pointstile;

[0098] The local point feature grouped_points, multiple copies of the point feature points_tile and the difference feature gcn are spliced ​​to obtain the enhanced local point cloud feature grouped_points_agg;

[0099] The enhanced local point cloud feature grouped_points_agg is concatenated with the local geometric information grouped_xyz to obtain the final feature representation.

[0100] Segmentation module: used to input the point cloud and its corresponding features into the fully connected layer and segment the semantic information of the point cloud.

[0101] An embodiment of the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the computer program is loaded into the processor, the method for segmenting corn plant point cloud organs by integrating dynamic weights and feature enhancement is implemented.

[0102] An embodiment of the present invention further provides a storage medium storing a computer program, wherein the computer program, when executed by a processor, implements any one of the methods for corn plant point cloud organ segmentation integrating dynamic weights and feature enhancement.

Claims

1. A corn plant point cloud organ segmentation method integrating dynamic weights and feature enhancement, characterized in that: The following steps are involved: (1) Input corn plant point cloud data, including the spatial coordinates and additional features of the point cloud; (2) Perform point cloud sampling and aggregation on the input point cloud data to obtain sampled point cloud data, and combine multi-scale clustering, that is, perform sphere queries with different radii to group the point clouds; (3) Use the DWE network for feature extraction; the DWE network is composed of the ScoreNet dynamic weight generation module and the feature enhancement module. The dynamic weight is generated by ScoreNet, and the feature enhanced point cloud is weighted and aggregated according to the dynamic weight; (4) The point cloud and its corresponding features are input into the fully connected layer and the semantic information of the point cloud is segmented.

2. The corn plant point cloud organ segmentation method integrating dynamic weight and feature enhancement according to claim 1, characterized in that: In step (2), the sampling points are obtained by using density-based farthest point sampling, including the following steps: (21) Receive point cloud data and set the number of sampling points n point The number of neighbors k required for density calculation, an array for storing the sampling point index and an array for recording the distance from each point to the nearest sampling point; (22) Calculate the Euclidean distance matrix between any two points in the point cloud, find the k nearest neighbor points and corresponding distances for each point, and use the inverse of the average nearest distance of each point as the density weight; (23) After obtaining the density weight of each point, a point in the point cloud is randomly selected as the first sampling point. In order to ensure that all points are likely to be selected when the farthest point is selected for the first time, we initialize the distance from all points in the point cloud to the sampling point to 10 10 ; (24) Among the currently unselected points, select the point with the largest weighted distance to the existing sampling point set as the new sampling point, calculate the distances of all points to the current sampling point and update the closest distance of each point to the selected sampling point set, retaining only the smallest distance, and perform n point Iterative sampling; the distance between sampling points is obtained according to the density weight; (25) Output the final selected n point The index of the sampling point is used to complete the sampling.

3. The corn plant point cloud organ segmentation method integrating dynamic weight and feature enhancement according to claim 1, characterized in that: In step (3), the ScoreNet module constituting the DWE network generates dynamic weights according to the input point cloud data features and additional features for weighted processing of local point cloud features. The specific steps are as follows: (31) The ScoreNet module generates a dynamic weight for each point based on the local features of the point cloud input, indicating the importance of the local area in the segmentation process, with a value between 0 and 1; (32) The generated dynamic weights are used to perform weighted aggregation on the sampled and grouped point cloud data, thereby amplifying the contribution of important features to the results and vice versa.

4. The corn plant point cloud organ segmentation method integrating dynamic weight and feature enhancement according to claim 1, characterized in that: In step (3), the feature enhancement module of the DWE network is used to enhance the features of the point cloud data through the following steps: (S31) obtaining local aggregated point cloud data from the sampled point cloud data; (S32) extracting neighborhood point features from the global point cloud; (S33) using the broadcast mechanism, broadcasting the sampling point features, constructing a point cloud feature matrix, forming a broadcast feature with the same dimension as the neighborhood point feature, and calculating the difference gcn between the neighborhood point feature and the broadcast feature, where gcn = grouped_points-points_tile; (S34) splicing the local point feature grouped_points, the multiple copies of the point feature points_tile and the difference feature gcn to obtain an enhanced local point cloud feature grouped_points_agg; (S35) The enhanced local point cloud feature grouped_points_agg is concatenated with the local geometric information grouped_xyz to obtain a final feature representation.

5. A corn plant point cloud organ segmentation system integrating dynamic weights and feature enhancement, characterized in that: Includes point cloud data module: used to input corn plant point cloud data, including spatial coordinates and additional features of the point cloud; Sampling module: used to sample and aggregate the input point cloud data to obtain sampled point cloud data, and combine multi-scale clustering, that is, perform sphere queries with different radii to group the point clouds; DWE network module: used to extract features using the DWE network; the DWE network is a fusion of the ScoreNet dynamic weight generation module and the feature enhancement module. The dynamic weight is generated by ScoreNet, and the feature-enhanced point cloud is weighted and aggregated according to the dynamic weight. Segmentation module: used to input the point cloud and its corresponding features into the fully connected layer and segment the semantic information of the point cloud.

6. The corn plant point cloud organ segmentation system integrating dynamic weight and feature enhancement according to claim 5, characterized in that: In the sampling module, the sampling points are obtained using density-based farthest point sampling, as follows: Receive point cloud data and set the number of sampling points n point The number of neighbors k required for density calculation, an array for storing the sampling point index and an array for recording the distance from each point to the nearest sampling point; Calculate the Euclidean distance matrix between any two points in the point cloud, find the k nearest neighbor points and corresponding distances for each point, and use the inverse of the average nearest distance of each point as the density weight; After obtaining the density weight of each point, randomly select a point in the point cloud as the first sampling point. In order to ensure that all points are likely to be selected when the farthest point is selected for the first time, we initialize the distance from all points in the point cloud to the sampling point to 10 10 ; Among the currently unselected points, select the point with the largest weighted distance to the existing sampling point set as the new sampling point, calculate the distance of all points to the current sampling point and update the shortest distance of each point to the selected sampling point set, retaining only the smallest distance, and perform n point Iteration sampling; The distance of the sampling points is obtained according to the density weight; Output the final selected n point The index of the sampling point is used to complete the sampling.

7. The corn plant point cloud organ segmentation system integrating dynamic weight and feature enhancement according to claim 5, characterized in that: In the DWE network module, the ScoreNet module that constitutes the DWE network generates dynamic weights based on the input point cloud data features and additional features for weighted processing of local point cloud features. The specific steps are as follows: The ScoreNet module generates a dynamic weight for each point based on the local features of the point cloud input, indicating the importance of the local area in the segmentation process, with a value between 0 and 1; The generated dynamic weights are used to perform weighted aggregation on the sampled and grouped point cloud data, amplifying the contribution of important features to the results, and vice versa.

8. The corn plant point cloud organ segmentation system integrating dynamic weight and feature enhancement according to claim 5, characterized in that: In the DWE network module, the feature enhancement module that constitutes the DWE network performs feature enhancement on the point cloud data through the following steps: Obtaining local aggregated point cloud data from the sampled point cloud data; Extract neighborhood point features from the global point cloud; Using the broadcast mechanism, the sampling point features are broadcasted and transmitted to construct the point cloud feature matrix, forming broadcast features with the same dimension as the neighborhood point features, and calculating the difference gcn between the neighborhood point features and the broadcast features, where gcn = grouped_points-points_tile; The local point feature grouped_points, multiple copies of the point feature points_tile and the difference feature gcn are spliced ​​to obtain the enhanced local point cloud feature grouped_points_agg; The enhanced local point cloud feature grouped_points_agg is concatenated with the local geometric information grouped_xyz to obtain the final feature representation.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the computer program is loaded into a processor, a method for corn plant point cloud organ segmentation integrating dynamic weights and feature enhancement according to any one of claims 1 to 4 is implemented.

10. A storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, a method for corn plant point cloud organ segmentation integrating dynamic weights and feature enhancement is implemented according to any one of claims 1 to 4.

Images