A method and apparatus for target classification of three-dimensional sonar point clouds based on persistent homology

Abstract

The application discloses a three-dimensional sonar point cloud target classification method and device based on persistent homology, and the method comprises the following steps: performing preprocessing including filtering, downsampling and noise reduction on obtained three-dimensional sonar point cloud data; adopting a clustering algorithm to cluster the three-dimensional sonar point cloud data after preprocessing to obtain point cloud clusters; adopting a persistent homology method to extract topological structure information of the point cloud clusters and to represent the topological structure information in a persistent graph, calculating a Wasserstein distance of the persistent graph, and simultaneously counting multi-feature fusion parameters including density, centroid and average scattering intensity of the point cloud clusters; and inputting the Wasserstein distance of the persistent graph and the multi-feature fusion parameters into a KNN model to obtain a target classification result. The application can further improve the generalization ability of KNN model prediction, so that the model has stronger adaptability and robustness in actual application, thereby effectively improving the accuracy of three-dimensional sonar point cloud target classification.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of three-dimensional sonar detection, and particularly relates to a three-dimensional sonar point cloud target classification method and device based on persistent homology. BACKGROUND

[0002] With the development of marine resources, the demand for underwater detection is increasing, and underwater target recognition has become a topic of concern. Sonar, as a means of underwater acoustic detection, has received widespread attention. Currently, related technologies in the field of sonar have developed to some extent, but there are still many difficulties and challenges in this field.

[0003] Compared with two-dimensional images, the point cloud data obtained by three-dimensional imaging sonar has a significant increase in information due to the increase in dimension, and there are also differences in processing algorithms. Currently, the processing of three-dimensional imaging sonar data can be roughly divided into three methods: two-dimensional image-based, voxel-based, and point cloud-based methods. The two-dimensional image-based method projects three-dimensional point clouds onto a two-dimensional plane and uses mature image processing algorithms for processing, but this method loses part of the spatial information and affects the recognition accuracy. The voxel-based method divides the three-dimensional space into multiple small cubes (voxels) and performs statistical analysis on the point cloud in each voxel. This method retains part of the spatial information, but has a large amount of calculation and is easily affected by noise. The point cloud-based method directly processes the original point cloud data, which can fully utilize the spatial information and structural features of the point cloud, but requires high computing resources.

[0004] In the process of three-dimensional point cloud processing, the following difficult problems need to be paid special attention to:

[0005] (1) Small sample size: Due to the high cost of underwater sonar equipment and the requirement of professional operation skills for technical personnel, the process of sonar data collection is long and costly. Therefore, the number of available sonar samples is often limited, and there is a lack of sufficient public data sets for algorithm verification and performance comparison.

[0006] (2) Target imbalance: The number of target species in the real environment is limited and affected by different experimental environments, which leads to a large difference in the number of samples between different target species in the sonar collected data, and the overall target amount is small. This imbalance not only affects the generalization ability of the target recognition model, but also reduces the recognition accuracy of the model for rare targets.

[0007] (3) Low resolution: Compared with high-precision sensors such as laser radar, the resolution of sonar images is low, and the target features are difficult to identify clearly. The complexity of the underwater environment also makes the analysis of acoustic signals more difficult. Therefore, the imaging results of sonar images often have problems such as blurring, high noise, weak interpretability, etc. These problems not only increase the difficulty of target recognition, but also reduce the accuracy and reliability of the recognition results.

[0008] In summary, three-dimensional imaging sonar technology has broad application prospects and important research value in the field of underwater target recognition. However, faced with the challenges and difficulties of small sample size, target imbalance, and low resolution, researchers need to continuously explore new technical routes and solutions to promote the further development and application of three-dimensional imaging sonar technology. SUMMARY

[0009] In view of the above, the purpose of the present application is to provide a three-dimensional sonar point cloud target classification method and device based on persistent homology, which can accelerate the processing speed through preprocessing and clustering, and significantly improve the accuracy of target classification by introducing the topological structure information of three-dimensional sonar point cloud and multi-feature fusion parameters into the decision of target classification. Further, the K-fold cross-validation method is used to train the KNN model, and the hyperparameters most suitable for sample data are selected, thereby further improving the generalization ability of the model, making the model have stronger adaptability and robustness in actual application.

[0010] To achieve the above-mentioned purpose of the application, the technical solutions provided by the present application are as follows:

[0011] The three-dimensional sonar point cloud target classification method based on persistent homology provided by the embodiment of the present application comprises the following steps:

[0012] The acquired three-dimensional sonar point cloud data is preprocessed, including filtering, downsampling and noise reduction;

[0013] The preprocessed three-dimensional sonar point cloud data is clustered by a clustering algorithm to obtain point cloud clusters;

[0014] The topological structure information of the point cloud cluster is extracted by a persistent homology method and represented by a persistent graph, the Wasserstein distance of the persistent graph is calculated, and the multi-feature fusion parameters including density, centroid and average scattering intensity of the point cloud cluster are counted;

[0015] The Wasserstein distance of the persistent graph and the multi-feature fusion parameters are input into a KNN model to predict the target classification result.

[0016] Specifically, the topological structure information of the point cloud cluster is extracted by the persistent homology method and represented by the persistent graph, and the Wasserstein distance of the persistent graph is calculated, comprising:

[0017] Encode the point cloud in the same point cloud cluster and initialize the filtering parameter ε;

[0018] The complex is formed by the process of adding data points to the simplex, that is, when the diameter between two data points is less than 2ε, an edge is added between them to construct the VR complex;

[0019] Repeat the construction process of the VR complex for different filtering parameters ε to obtain a series of nested sub-complexes of simplexes, which constitute the filter flow, and identify the topological features in the filter flow by calculation;

[0020] During the change of the filtering parameter ε, record the generation and disappearance of the topological features and represent them in the form of persistent diagrams;

[0021] Calculate the Wasserstein distance between different point cloud persistent diagrams to evaluate the similarity of the topological structures of the point clouds, and the smaller the distance value, the higher the similarity of the two point clouds in the topological structure.

[0022] Specifically, the Wasserstein distance of the persistent diagram and the multi-feature fusion parameter are input into the KNN model to predict the target classification result, which comprises:

[0023] The multi-feature fusion parameter of the point cloud cluster including density, centroid and average scattering intensity is weighted with the Wasserstein distance of the persistent diagram and normalized, and then input into the KNN model. In the KNN model, at least one neighbor point of the data point to be classified in the point cloud cluster is found, and the class of the data point to be classified is determined according to the majority voting principle, and finally the target classification result of the point cloud cluster is obtained.

[0024] Specifically, the K-fold cross-validation method is used to train the KNN model based on the point cloud clusters obtained by clustering.

[0025] Specifically, the filtering in the preprocessing includes threshold filtering and maximum filtering. For the beam domain obtained by the beam forming process of the three-dimensional sonar point cloud data, the data with intensity less than the filtering threshold in the beam domain is removed by threshold filtering, and the data with the highest intensity in each beam direction in the beam domain is retained by maximum filtering.

[0026] Specifically, the downsampling in the preprocessing includes voxel method, which divides the sonar detection space into several cubic units, each cubic unit as a voxel, selects the representative point of the three-dimensional sonar point cloud data in the voxel, and calculates the average intensity of all data points in the voxel. The average intensity is used as the intensity of the representative point, and all representative points in the voxel are combined to form the downsampled point cloud data.

[0027] Specifically, the noise reduction in the preprocessing includes adopting a region self-growth algorithm based on the Euclidean distance, selecting a point with the smallest curvature in the filtered and down-sampled three-dimensional sonar point cloud data as a seed point, performing an expansion algorithm with the seed point as the center, searching for a neighbor point of the seed point in the expansion process, and dividing the neighbor point into a valid point if the Euclidean distance from the neighbor point to the seed point is less than a set distance threshold, continuing to iteratively perform seed point selection, expansion, searching, and dividing valid points on the remaining three-dimensional sonar point cloud data except the valid points, and finally saving all the valid points as the three-dimensional sonar point cloud data after preprocessing.

[0028] Specifically, in the expansion process, the kd-tree structure is adopted to organize the point cloud data, the point cloud data is divided into two sets as a left subtree and a right subtree, and a neighbor searching algorithm based on the Euclidean distance is adopted to search for the neighbor points based on the left subtree and the right subtree.

[0029] Specifically, the clustering algorithm includes adopting a k-means++ algorithm to cluster the three-dimensional sonar point cloud data after preprocessing to obtain a plurality of point cloud clusters.

[0030] To achieve the above-mentioned purposes, the embodiments of the present application also provide a three-dimensional sonar point cloud target classification device based on persistent homology, which comprises a preprocessing module, a clustering module, a topological feature extraction module, and a target classification module.

[0031] The preprocessing module is used for preprocessing the obtained three-dimensional sonar point cloud data, including filtering, down-sampling, and noise reduction.

[0032] The clustering module is used for clustering the three-dimensional sonar point cloud data after preprocessing by adopting a clustering algorithm to obtain point cloud clusters.

[0033] The topological feature extraction module is used for extracting the topological structure information of the point cloud clusters by adopting a persistent homology method and representing it in a persistent graph, calculating the Wasserstein distance of the persistent graph, and simultaneously calculating a multi-feature fusion parameter of the point cloud cluster, including density, centroid, and average scattering intensity.

[0034] The target classification module is used for inputting the Wasserstein distance of the persistent graph and the multi-feature fusion parameter into a KNN model to predict a target classification result.

[0035] Compared with the prior art, the present application has at least the following beneficial effects:

[0036] (1) The present application introduces the topological structure information of the point cloud into the decision of target classification, fully excavates and utilizes the structure information of the existing data, and improves the resolution of the three-dimensional sonar point cloud data and the interpretability of the target classification.

[0037] (2) The application comprehensively considers the density, centroid and average scattering intensity information of the point cloud, combines the Wasserstein distance weighted calculation between the persistence diagram as the input of the KNN model, and can automatically retrain the parameters in idle time, thereby effectively improving the classification accuracy of the model.

[0038] (3) The application reduces the influence of underwater noise echo by filtering, downsampling and noise reduction processing of point cloud data in the preprocessing stage, and accelerates the processing speed through the kd-tree structure, which can further save the platform computing power.

[0039] (4) The application adopts K-fold cross-validation method to train KNN model parameters, optimizes the parameter selection process, optimizes the small sample scene, and can effectively deal with the problems of small number of underwater acoustic samples and insufficient diversity, further improves the generalization ability and robustness of the model. BRIEF DESCRIPTION OF DRAWINGS

[0040] In order to more clearly illustrate the technical solutions in the embodiments of the application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the application, and those skilled in the art can obtain other drawings according to these drawings without creative labor.

[0041] Figure 1 is a flowchart of the three-dimensional sonar point cloud target classification method based on persistent homology provided by the embodiment of the application;

[0042] Figure 2 is a flowchart of data preprocessing provided by the embodiment of the application;

[0043] Figure 3 is a flowchart of constructing kd-tree provided by the embodiment of the application;

[0044] Figure 4 is a flowchart of persistent homology provided by the embodiment of the application;

[0045] Figure 5 is a schematic diagram of simplex, complex and persistence diagram provided by the embodiment of the application;

[0046] Figure 6 is a schematic diagram of complex growth process provided by the embodiment of the application;

[0047] Figure 7 is a flowchart of KNN model target classification provided by the embodiment of the application;

[0048] Figure 8is a multi-feature fusion parameter construction process schematic diagram provided by the embodiment of the application.

[0049] Figure 9 is a structural schematic diagram of the three-dimensional sonar point cloud target classification device based on persistent homology provided by the embodiment of the application. DETAILED DESCRIPTION

[0050] In order to make the objects, technical solutions and advantages of the present application clearer, further detailed description will be made to the present application in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application, and do not limit the protection scope of the present application.

[0051] The inventive concept of the present application is that, in view of the challenges and difficulties such as small sample quantity, unbalanced target types and low resolution faced by the three-dimensional imaging sonar technology in the field of underwater target recognition in the prior art, the embodiment of the present application provides a three-dimensional sonar point cloud target classification method and device based on persistent homology, which can reduce the influence of underwater noise echo and speed up the processing speed by preprocessing and clustering the three-dimensional sonar point cloud data, and can improve the accuracy of target classification by introducing the topological structure information and multi-feature fusion parameters of the three-dimensional sonar point cloud into the decision of KNN model target classification. Further, the K-fold cross-validation method is used to train the KNN model, which not only effectively balances the sample quantity of different target types, but also enhances the diversity and representativeness of sample data, thereby further improving the generalization ability of the model, so that the model has stronger adaptability and robustness in actual application. In the actual deployment process, if the conditions are met, the model supports retraining parameters in idle time to better adapt to different application scenarios.

[0052] Figure 1 is a flowchart of the three-dimensional sonar point cloud target classification method based on persistent homology provided by the embodiment of the application. As shown in Figure 1 , the embodiment provides a three-dimensional sonar point cloud target classification method based on persistent homology, which includes the following steps:

[0053] S1, preprocessing the obtained three-dimensional sonar point cloud data, including filtering, downsampling and noise reduction. The preprocessing process is as shown in Figure 2 .

[0054] S1.1, filtering processing. After the signal received by the three-dimensional sonar is processed by the beam forming process, the data intensity of the target in the beam domain is larger than the underwater environmental noise and is concentrated in a specific frequency band range. In order to speed up the subsequent processing flow, a specific threshold is set for preliminary threshold filtering to remove data with intensity less than μ times the maximum intensity in the beam domain data, thereby achieving the purpose of removing random environmental noise and reducing data volume. At the same time, considering that the signal intensity received by the sonar equipment is generally limited within a certain range, in order not to be affected by strong noise or strong interference and its sidelobes in the environment, resulting in a decline in target observation effect, the data is subjected to maximum filtering, and the n highest intensity points in each beam direction are retained.

[0055] S1.2, down-sampling processing. The point cloud data is down-sampled in a voxel manner, the process including: dividing the sonar detection space into a plurality of cubic units with a specific edge length, each cubic unit serving as a voxel, selecting a representative point of the three-dimensional sonar point cloud data in the voxel, and counting the average intensity of all data points located in the voxel, taking the average intensity as the intensity of the representative point. The size of the cubic unit can directly reflect the point cloud down-sampling ratio. Finally, the representative points in all voxels are combined to form the down-sampled point cloud data. In order to improve the running speed of the algorithm while minimizing information loss, the voxel size in the embodiment is set as a tunable hyperparameter, which is optimized according to the running results of the subsequent algorithm.

[0056] S1.3, noise reduction processing. Considering that the suspended matter in the water body will produce echo interference to form abnormal points, it is necessary to perform noise reduction processing on the point cloud. In the embodiment, a region self-growth algorithm based on Euclidean distance is used for noise reduction, and the steps are as follows:

[0057] S1.3.1, for the three-dimensional sonar point cloud data obtained after filtering and down-sampling, selecting the point with the smallest curvature in the point cloud data as a seed point, and performing a growth algorithm for extension with the seed point as the center;

[0058] S1.3.2, in the extension process, searching for the near neighbor points of the seed point, and if the Euclidean distance from the near neighbor point to the seed point is less than a set distance threshold, the near neighbor point is divided into an effective point;

[0059] S1.3.3, for the remaining three-dimensional sonar point cloud data other than the effective points, selecting the point with the smallest curvature as a seed point;

[0060] S1.3.4, repeating steps S1.3.2-S1.3.3 until the number of remaining points is less than a certain threshold, or the number of iterations reaches an upper limit, and saving all effective points as the pre-processed three-dimensional sonar point cloud data.

[0061] In the extension process, in order to accelerate the process of searching the near neighbor points in step S1.3.2, the point cloud data is organized in a kd-tree structure. The kd-tree is a data structure for dividing data points in a k-dimensional space, and is commonly used for nearest neighbor search in large high-dimensional data spaces.

[0062] The process of constructing the kd-tree, i.e., the process of generating a binary tree, includes the following steps, as shown in FIG. 2: Figure 3

[0063] (1) Select a coordinate axis i (i is any coordinate axis in three-dimensional coordinates) participating in the division. When the variance of data on a certain coordinate axis is large, it indicates that the data points are more dispersed on this coordinate axis, and the division effect is usually good. Therefore, the coordinate axis with the largest coordinate variance is selected by calculation to divide the coordinate value on the coordinate axis i;

[0064] (2) Select a median point in the coordinate value on the coordinate axis i, and take the data point as the parent node of the divided sub-tree. The vertical axis of the median point and the coordinate axis i is taken as the division plane to divide the three-dimensional space, and the point cloud data is divided into two sets as the left sub-tree and the right sub-tree;

[0065] (3) Repeat steps (1)-(2) for the left sub-tree and the right sub-tree obtained by the above division, until the space cannot be divided, at which time the data point is taken as a leaf node.

[0066] After the construction of the kd-tree is completed, the point cloud data is searched for the nearest neighbor in step S1.3.2 according to the nearest neighbor search algorithm based on the Euclidean distance, and the steps are as follows:

[0067] S1.3.2.1, for a seed point to be searched, starting from the root node of the kd-tree, comparing one by one, calculating the Euclidean distance between the seed point and the current node, if less than the set distance threshold ω, accessing the left sub-tree, otherwise accessing the right sub-tree, searching for the nearest neighbor as an effective point;

[0068] S1.3.2.2, save and update the current nearest neighbor and the nearest neighbor distance during the search process until the leaf node of the kd-tree is reached;

[0069] S1.3.2.3, backtracking, judging whether there is a distance updated point in other nodes on the search path, if so, adding it to the search path and continuing to search until the search path is empty.

[0070] The point cloud data after the above preprocessing is more compact and has less noise interference, which helps to improve the running speed and accuracy of the subsequent algorithm.

[0071] ​S2, the pre-processed three-dimensional sonar point cloud data is clustered by using a clustering algorithm to obtain point cloud clusters.

[0072] The k-means clustering algorithm has the characteristics of being able to minimize the distance between data points in the same category, being simple and fast, but considering the accuracy of the k-means algorithm, it is difficult to guarantee due to the random selection of the initial center point of the clustering, and a bad iteration result may be produced. In order to reduce the potential influence of the random initial center on the clustering result, the improved k-means++ algorithm is used in the embodiment to cluster the pre-processed three-dimensional sonar point cloud data to obtain a plurality of point cloud clusters, which has the characteristics that the selection of the first center point is also completely random, but the selection of the subsequent n+1 center point adopts the D 2 The weight method, that is, the farther the point is from the previous n center points, the greater the probability of being selected as the n+1 center point, and in general cases, the k-means++ algorithm can obtain stable results in various aspects.

[0073] S3, the topological structure information of the point cloud cluster is extracted by using the persistent homology method and is represented by a persistent diagram, the Wasserstein distance of the persistent diagram is calculated, and a plurality of feature fusion parameters including density, centroid and average scattering intensity of the point cloud cluster are counted.

[0074] After clustering, the point clouds belonging to the same category are considered as an integral underwater target. Generally, underwater objects in a fixed scene often have relatively fixed structures and persistent characteristics, so the structure information of the sonar point cloud has an important role in target classification, and the structure information can be expressed by topological structure. Topology is a mathematical tool for studying the invariance of geometric figures or spaces in the process of continuous shape change, which focuses on expressing the geometric characteristics and positional relationship of objects without considering deformation and size. Topological data analysis (TDA) characterizes the global features of high-dimensional data by studying the shape characteristics of the data. In TDA, persistent homology is an important analysis method, which focuses on some topological features that appear or disappear with the change of scale, and can give the number of groups, loops, cavities and other structures in the data set, so as to characterize the overall situation of the data without losing dimension information.

[0075] In the embodiment, the topological structure information of the point cloud cluster is extracted by using the persistent homology method and is represented by a persistent diagram. The persistent homology process is as shown in Figure 4 The specific steps are as follows:

[0076] S3.1, initialize the filtering parameter ε = 0, and encode the point cloud;

[0077] S3.2, for a given filtering parameter ε, the complex is formed by the process of adding data points to simplex: according to the definition of VR complex, when the diameter between two data points is less than 2ε in the process of simplex growth, an edge is added between the point clouds of data points, so as to construct the VR complex, an example of the definition of simplex, complex and persistent diagram is shown as follows Figure 5 As shown in the figure, the complex growth process is as shown in Figure 6

[0078] S3.3, for different filtering parameters ε at different scales, repeat the construction process of the above VR complex to obtain a series of nested sub-complexes of simplexes, which constitute the filter flow, and the topological features are identified by calculating the homology group at different scales;

[0079] S3.4, record the generation and disappearance of topological features in the process of changing the filtering parameter ε, and take the time when each topological feature is generated as the horizontal coordinate and the time when it disappears as the vertical coordinate, so as to obtain a series of point pairs, which constitute the persistent diagram of the homology group;

[0080] S3.5, calculate the Wasserstein distance between the persistent diagrams of different point clouds to evaluate the similarity of the topological structures of the point clouds, and the smaller the distance value is, the higher the similarity of the two point clouds in topological structure is.

[0081] In addition to the topological features, the centroid, density and average scattering intensity information of different categories of point cloud clusters are also calculated in the process to form multi-feature fusion parameters, which are weighted and involved in decision-making in the KNN model.

[0082] S4, input the Wasserstein distance of the persistent diagram and the multi-feature fusion parameter into the KNN model to obtain the target classification result.

[0083] In the embodiment, the KNN model is constructed based on the improved KNN (K-Nearest Neighbors) algorithm, which is a basic classification and regression algorithm suitable for situations where the initial data samples are insufficient, and can automatically optimize and update the model in actual application. In order to obtain better hyperparameters, the K-fold cross-validation method is used to train the KNN model based on the point cloud clusters obtained by clustering. The data set used in the embodiment is obtained from three-dimensional sonar point cloud data obtained by sea experiment, and the parameter optimization is performed by K-fold cross-validation. In the embodiment, K=5 is selected, that is, the data set is divided into 5 folds for cross-validation. At the same time, due to the limitation of the training set, the generalization ability of the obtained model may be limited. With the increase of test data set and the addition of prior information, the performance of the original model decreases, therefore, in the embodiment, when a certain condition is met, the parameter model is automatically retrained in idle time to automatically adapt to more data and improve the generalization ability of the model. ​

[0084] In an embodiment, the process of target classification using the KNN model is as shown in Figure 7 The multi-feature fusion parameters of the point cloud cluster including density, centroid and average scattering intensity are weighted and normalized and then input into the KNN model (as shown in Figure 8 In the KNN model, the class of the data point to be classified is determined by finding the k nearest neighbors of the data point to be classified in the point cloud cluster and according to the majority voting principle, and finally the target classification result of the point cloud cluster is obtained.

[0085] So far, through the algorithm process of point cloud data preprocessing, clustering, topological feature extraction and target classification, the implementation scheme designed by the present application completes the target classification of three-dimensional sonar point cloud data, and the classification result obtained can be observed by the user or applied to other data post-processing processes.

[0086] In summary, the three-dimensional sonar point cloud target classification method based on persistent homology provided by the embodiment of the present application introduces the topological structure information of the point cloud into the decision of target classification, fully mines and utilizes the structure information of the existing data, and improves the resolution of the three-dimensional sonar point cloud data and the interpretability of target classification. The density, centroid and average scattering intensity information of the point cloud are comprehensively considered, the weighted calculation of the Wasserstein distance between the persistent graphs is combined as the input of the KNN model, and the parameters can be automatically retrained at leisure, thereby effectively improving the classification accuracy of the model. Through filtering, downsampling and noise reduction processing of the point cloud data in the preprocessing stage, the influence of underwater noise echoes is reduced, and the processing speed is accelerated through the kd-tree structure, which can further save platform computing power. The K-fold cross-validation method is used to train the KNN model parameters, the parameter selection process is optimized, and optimization is made for small sample scenarios, which can effectively solve the problems of small number of underwater acoustic samples and insufficient diversity, and further improve the generalization ability and robustness of the model.

[0087] Based on the same inventive concept, as shown in Figure 9 The embodiment of the present application also provides a three-dimensional sonar point cloud target classification device 900 based on persistent homology, which comprises a preprocessing module 910, a clustering module 920, a topological feature extraction module 930 and a target classification module 940.

[0088] The preprocessing module 910 is used for preprocessing the acquired three-dimensional sonar point cloud data, including filtering, downsampling and noise reduction.

[0089] The clustering module 920 is used for clustering the preprocessed three-dimensional sonar point cloud data to obtain a point cloud cluster by using a clustering algorithm.

[0090] The topology feature extraction module 930 is configured to extract the topology structure information of the point cloud cluster by using the persistent homology method and represent the topology structure information in a persistent diagram, calculate the Wasserstein distance of the persistent diagram, and meanwhile, count the multi-feature fusion parameters of the point cloud cluster, including the density, the centroid and the average scattering intensity.

[0091] The target classification module 940 is configured to input the Wasserstein distance of the persistent diagram and the multi-feature fusion parameters into a KNN model to obtain a target classification result.

[0092] It should be noted that the three-dimensional sonar point cloud target classification device based on persistent homology provided in the above embodiments and the three-dimensional sonar point cloud target classification method based on persistent homology belong to the same inventive concept, and the specific implementation process is detailed in the three-dimensional sonar point cloud target classification method based on persistent homology, which will not be described here.

[0093] The specific embodiments described above have explained the technical solutions and beneficial effects of the present application in detail. It should be understood that the above description is only the most preferred embodiments of the present application and is not intended to limit the present application. Any modifications, supplements and equivalent replacements made within the principle range of the present application shall be included in the protection scope of the present application.

Claims

1. A persistent coherence-based three-dimensional sonar point cloud target classification method, characterized in that, The method comprises the following steps: The acquired three-dimensional sonar point cloud data is preprocessed, including filtering, downsampling and noise reduction; The preprocessed three-dimensional sonar point cloud data is clustered by using a clustering algorithm to obtain point cloud clusters; The topological structure information of the point cloud clusters is extracted by using a persistent homology method and is represented by a persistence diagram, the Wasserstein distance of the persistence diagram is calculated, and the multi-feature fusion parameters of the point cloud clusters, including density, centroid and average scattering intensity, are counted; The Wasserstein distance of the persistence diagram and the multi-feature fusion parameters are input into a KNN model to obtain a target classification result.

2. The persistent coherence-based three-dimensional sonar point cloud target classification method according to claim 1, characterized in that, The topological structure information of the point cloud clusters is extracted by using a persistent homology method and is represented by a persistence diagram, and the Wasserstein distance of the persistence diagram is calculated, and the multi-feature fusion parameters of the point cloud clusters, including density, centroid and average scattering intensity, are counted. The points in the same point cloud cluster are encoded, and the filtering parameter ε is initialized; A complex is formed through the process of adding data points, that is, when the diameter between two data points is less than 2ε, an edge is added between them to construct a VR complex; For different filtering parameters ε, the construction process of the VR complex is repeated to obtain a series of nested sub-complexes of simple complexes, and these sub-complexes constitute a filter flow, and the topological features are identified by calculation; During the change of the filtering parameter ε, the generation and disappearance of the topological features are recorded and represented in the form of a persistence diagram; The Wasserstein distance between different point clouds is calculated to evaluate the similarity of the topological structures of the point clouds, and the smaller the distance value is, the higher the similarity of the two point clouds in the topological structure is.

3. The persistent coherence-based 3D sonar point cloud target classification method according to claim 1, wherein, The Wasserstein distance of the persistence diagram and the multi-feature fusion parameters are input into a KNN model to obtain a target classification result. The multi-feature fusion parameters of the point cloud clusters, including density, centroid and average scattering intensity, and the Wasserstein distance of the persistence diagram are weighted and normalized and then input into the KNN model, at least one neighbor point of the data point to be classified in the point cloud cluster is found in the KNN model, and the class of the data point to be classified is determined according to the majority voting principle, and finally the target classification result of the point cloud cluster is obtained.

4. The persistent coherence-based 3D sonar point cloud target classification method according to claim 3, characterized in that, The KNN model is trained by using the K-fold cross-validation method based on the point cloud clusters obtained by clustering.

5. The persistent coherence-based 3D sonar point cloud target classification method according to claim 1, wherein, The filtering in the preprocessing includes threshold filtering and maximum filtering, for the beam domain obtained by the beam forming process of the three-dimensional sonar point cloud data, the data with an intensity less than the filtering threshold in the beam domain is removed by threshold filtering, and the data with the highest intensity in each beam direction in the beam domain is retained by maximum filtering.

6. The persistent coherence-based three-dimensional sonar point cloud target classification method according to claim 1, wherein, The downsampling in the preprocessing includes voxelization, the sonar detection space is divided into a plurality of cubic units, each cubic unit is used as a voxel, a representative point of the three-dimensional sonar point cloud data in the voxel is selected, the average intensity of all data points located in the voxel is counted, the average intensity is used as the intensity of the representative point, and all representative points in the voxels are combined to form the downsampled point cloud data.

7. The persistent coherence-based 3D sonar point cloud target classification method according to claim 1, wherein, The noise reduction in the preprocessing includes using a region self-growth algorithm based on the Euclidean distance. For the three-dimensional sonar point cloud data obtained after filtering and downsampling, the point with the smallest curvature in the point cloud data is selected as a seed point. The growth algorithm is executed for expansion with the seed point as the center. The neighboring points of the seed point are searched during the expansion process. If the Euclidean distance from the neighboring point to the seed point is less than a set distance threshold, the neighboring point is divided into an effective point. The seed point selection, expansion, search, and division of effective points are iteratively executed for the remaining three-dimensional sonar point cloud data except the effective points. Finally, all the effective points are saved as the three-dimensional sonar point cloud data after preprocessing.

8. The persistent coherence-based three-dimensional sonar point cloud target classification method according to claim 7, characterized in that, In the expansion process, for the three-dimensional sonar point cloud data obtained after filtering and downsampling, the kd-tree structure is used to organize the point cloud data. The point cloud data is divided into two sets as a left subtree and a right subtree. The nearest neighbor search algorithm based on the Euclidean distance is used to search the neighboring points based on the left subtree and the right subtree.

9. The persistent coherence-based 3D sonar point cloud target classification method according to claim 1, wherein, The clustering algorithm includes using the k-means++ algorithm to cluster the three-dimensional sonar point cloud data after preprocessing to obtain a plurality of point cloud clusters.

10. A persistent coherence based three-dimensional sonar point cloud target classification apparatus, characterized in that, It includes: a preprocessing module, a clustering module, a topological feature extraction module, and a target classification module; the preprocessing module is used for preprocessing the obtained three-dimensional sonar point cloud data, including filtering, downsampling, and noise reduction; the clustering module is used for clustering the three-dimensional sonar point cloud data after preprocessing using a clustering algorithm to obtain point cloud clusters; the topological feature extraction module is used for extracting the topological structure information of the point cloud cluster using the persistent homology method and representing it with a persistent graph, calculating the Wasserstein distance of the persistent graph, and simultaneously calculating the multi-feature fusion parameters of the point cloud cluster, including density, centroid, and average scattering intensity; the target classification module is used for inputting the Wasserstein distance of the persistent graph and the multi-feature fusion parameters into a KNN model to predict the target classification result.

Images