Target clustering method based on multi-type photoelectric detection results of unmanned surface vehicle

Abstract

The invention belongs to the technical field of target clustering, and particularly relates to a target clustering method based on multi-type photoelectric detection results of an unmanned ship, which can cluster detection data according to a clustering radius and a clustering number, can realize manual correction, and can perform clustering calculation on corrected data again, thereby improving the accuracy of a clustering center. In order to realize post-processing on multi-type target data of photoelectric detection and reduce the influence of noise points on a clustering center, after the detection data is classified, the detection data is clustered by adopting a DBSCAN method according to the clustering radius and the clustering number, so that post-processing clustering can be carried out on multi-type targets of photoelectric detection, and the clustering accuracy is improved. Moreover, a mode that a detection picture corresponds to a detection target is designed, manual correction can be realized, and the corrected data is subjected to clustering calculation again, so that the accuracy of the clustering center is improved.

Description

Technical Field

[0001] This invention belongs to the field of target clustering technology, specifically relating to a target clustering method based on multi-type photoelectric detection results from unmanned surface vessels. Background Technology

[0002] The unmanned surface vessel (USV) optoelectronic target clustering technology is mainly applied to autonomous surface target detection missions by USVs in beyond-line-of-sight (BLOS) scenarios. In BLOS situations, the transmission of USV detection data is limited, and the detection results are stored in the onboard control unit. Once the USV reaches its mission withdrawal position or is recovering, communication is restored, and the detection results are downloaded to the control terminal for analysis and processing. During optoelectronic detection, to ensure a high target localization success rate, laser ranging is performed more than once, and the detection area is probed multiple times. This results in multiple localization data points for the same target within the detection results.

[0003] Patent CN119247390A uses the DBSCAN clustering algorithm to process water surface obstacles detected by the lidar of unmanned surface vessels. It filters and clusters multi-frame point cloud data to construct a grid map, providing support for environmental perception of unmanned surface vessels.

[0004] Patent CN116543299A uses the K-Means clustering algorithm to cluster maritime threat targets detected by unmanned surface vessel (USV) formations. It specifies the threat level of targets by location, speed, and heading, thereby driving USVs to move to locations with lower threat levels and reach targets with similar threat levels.

[0005] In existing technologies, when unmanned surface vessels conduct photoelectric detection, there may be a lot of noise points when performing laser ranging on small targets due to the influence of sea conditions. When using the K-Means averaging method for clustering, the cluster centers are easily skewed. Summary of the Invention

[0006] In view of this, the present invention provides a target clustering method based on multi-type photoelectric detection results of unmanned surface vessels. It can cluster detection data according to cluster radius and number of clusters, and can realize manual correction and re-clustering calculation on the corrected data, thereby improving the accuracy of cluster centers.

[0007] To achieve the objectives of this invention, the following technical solutions are provided.

[0008] A target clustering method based on multi-type photoelectric detection results from unmanned surface vessels includes the following steps: S1: Obtain photoelectric detection result files from the unmanned surface vessel. The files contain detection images and target data associated with each detection time. The target data includes at least the target type, target latitude, target longitude, and detection distance. S2: Classify the target data according to the target type, and perform clustering operation on each class of target data to determine the cluster center of that class of targets; S3: Filter the clustering results according to the preset target number, identify and mark abnormal data points; S4: Based on the detection time, associate and match the detected image with the corresponding target data and display it; S5: Based on the displayed information, manually correct the target data; S6: Re-execute the clustering operation in step S2 using the corrected target data.

[0009] The clustering operation employs an iterative method to generate cluster centers, specifically including: a. Select a point from the current target dataset as the initial cluster center; b. Calculate the distance between the cluster center and all other points in the dataset, and find the point with the farthest distance; c. Add the farthest point as a new cluster center; d. For each point in the dataset, calculate its minimum distance to all existing cluster centers, and record the minimum distance value and the corresponding cluster center index; e. Determine whether the maximum value of the minimum distance among all points is greater than the preset threshold. If yes, add the point corresponding to the maximum value as a cluster center and return to step d. If no, end the iteration, complete the clustering based on the recorded cluster center index, and output the final set of cluster centers.

[0010] The threshold is set to half the maximum distance between any two cluster centers in the current cluster center set.

[0011] The calculation of the distance between the two points in step d adopts the geographic coordinate projection method based on the Earth ellipsoid model. Specifically, it includes: calculating the projected distance components in the east-west and north-south directions according to the latitude and longitude of the two points, and calculating the straight-line distance between the two points by using the Pythagorean theorem.

[0012] In step S4, the corresponding detection time is extracted by parsing the naming information of the detection image and matching it with the detection time in the target data to achieve automatic association between the detection image and the target data.

[0013] The manual correction mentioned in step S5 includes at least one of the following operations: correcting erroneous target type labels and deleting distance anomaly data points that are determined to be detection errors.

[0014] The determination of abnormal distance data points prioritizes detection data for targets that are small.

[0015] In step S2, the output of the clustering center includes at least the geographical coordinates of each clustering center and the total number of targets in that cluster.

[0016] Beneficial effects 1. To achieve post-processing of multi-type target data from photoelectric detection and reduce the impact of noise points on cluster centers, this invention classifies the detection data and then uses the DBSCAN method to cluster the data based on cluster radius and number of clusters. This allows for post-processing clustering of multi-type targets in photoelectric detection. Furthermore, a method for mapping detection images to detection targets is designed, enabling manual correction and re-clustering of the corrected data, thereby improving the accuracy of cluster centers. This method provides a complete post-processing solution for beyond-line-of-sight photoelectric detection by unmanned surface vessels. By classifying and then clustering multi-type detection data, combined with image association display and manual interactive correction, it can systematically process datasets containing repeated measurements and noise. This method is not only highly automated but also significantly improves the accuracy of the final clustering results by introducing a mechanism of manual verification and iterative re-clustering. In particular, it reduces the interference of isolated noise points on the overall cluster center location, making the cluster centers more representative of the characteristics of real targets, thus providing a more reliable data foundation for subsequent target identification, situational analysis, and other tasks.

[0017] 2. In this invention, distance-based clustering of target data is employed, offering a simple implementation. The cluster centers generated by the clustering algorithm originate from the detected target data, eliminating the influence of outliers. The specific algorithmic flow for clustering operations is defined, differing from the traditional K-Means method's iterative generation of cluster centers based on the maximum-minimum distance principle. This algorithm naturally generates cluster centers from the dataset, independent of the selection of initial random center points, avoiding clustering bias caused by poor initial points. By dynamically determining and adding new cluster centers, it can adaptively determine the number of potential categories in the data, making it particularly suitable for scenarios with unknown category numbers or noise. This method has a clear flow and rigorous logic, effectively handling the complex data distributions commonly encountered in unmanned surface vessel (USV) photoelectric detection due to multiple measurements and noise.

[0018] 3. In this invention, the distance calculation adopts a projection method based on the Earth ellipsoid model. Compared with simple planar Euclidean distance or great circle distance approximations, this method fully considers the Earth's curvature and the actual distance differences at different latitudes, especially for application scenarios such as unmanned surface exploration where the geographical span may be large. By introducing the Earth's polar radius and equatorial radius for correction calculation, the calculated distance between targets is closer to the actual geographical distance, thus providing a higher precision input for distance-based clustering algorithms and further ensuring the physical spatial accuracy of the clustering results.

[0019] 4. In this invention, images and data are automatically associated through time information, employing an efficient and reliable image-text association mechanism. Through a unified and resolvable naming rule, automated matching between detection images and structured detection data is achieved, eliminating the need for manual lookup of correspondences and significantly improving post-processing efficiency. This association method allows operators to intuitively compare the target's numerical information with its visual representation in the image during the manual correction stage, providing crucial visual evidence for determining the accuracy of the target type and whether there are any anomalies in the detection distance. It is an important link in the human-machine collaborative processing closed loop.

[0020] 5. In this invention, the specific application scenarios of data deletion operations in manual correction provide clear judgment criteria and operational targets for manual intervention. Considering that small targets are more susceptible to laser ranging jumps (noise) caused by factors such as sea conditions, operators are allowed to proactively delete abnormal distance data points that significantly deviate from their respective groups, combining visual information from images and preliminary clustering results. This targeted correction operation can accurately eliminate bad points caused by instantaneous detection errors, preventing these bad points from adversely affecting the calculation of center points during re-clustering, thus making the final clustering results purer and more reliable. Attached Figure Description

[0021] Figure 1 This is a schematic diagram illustrating the implementation steps of the method of the present invention.

[0022] Figure 2 This is a schematic diagram of the clustering operation algorithm of the method of the present invention.

[0023] Figure 3 This is a schematic diagram showing the positional relationship of four targets in the detection results of the method embodiment of the present invention. Detailed Implementation

[0024] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0025] This invention provides a target clustering method based on multi-type photoelectric detection results from unmanned surface vessels, the implementation steps of which are as follows: Figure 1 As shown. The method of the present invention includes the following steps: Step S1: Obtain photoelectric detection results from the unmanned surface vessel, including detection images and detection time, target type, target latitude, target longitude and detection distance, and import them as files.

[0026] Step S2 provides a post-processing method for detecting target locations and images in beyond-line-of-sight situations. Specifically, it classifies the detection results according to target type and performs clustering operations. The clustering algorithm flowchart is shown below. Figure 2 As shown, the details are as follows: After importing the probe data, the target data of different types is first classified. For the classified data, a point is selected as the hypothetical cluster center. Based on this center, the distance to all other points is calculated, and the point farthest from the center is obtained, with a distance of D. This point is then used as the new cluster center, and the distance is calculated again. Based on the two calculation results, the minimum value of the distance from all target points to the two centers is compared, and the minimum value and its corresponding center number are recorded. It is then determined whether the maximum value in the minimum value list is greater than D / 2. If it is greater than D / 2, the point corresponding to that distance value is used as the new cluster center, and the loop continues. If it is less than D / 2, the target data is classified into different cluster centers according to the cluster center number.

[0027] The distance calculation formula is as follows, given the latitude and longitude of the two points. , Earth's polar radius is Equatorial radius The distance between the two points is L, and the distance calculation formula is as shown in formulas (1) to (5).

[0028]

[0029] Approximating the Earth as an ellipsoid, we establish a rectangular coordinate system with due east and due north. First, we calculate the radius and projected distance of a point at that latitude and longitude. Then, we calculate the difference in latitude and longitude between two points and determine the corresponding distance. Finally, based on the side lengths of a right triangle, we obtain the distance between the two points.

[0030] For example, if there are four target locations in the detection results, numbered 1 to 4, their positional relationships are as follows: Figure 3 As shown, the result list is input into the clustering algorithm. First, taking point 1 as the center, the distances to other points are calculated. Based on the positional relationship of the target points, the maximum distance is 1-3, so point 3 is taken as the new cluster center, and the distances to other points are calculated. Comparing the two sets of distances, the minimum distance is obtained in the min array, and the index of the minimum value is recorded in the Minindex array. The maximum value of the minimum value, 4-1, is taken. It is determined that it is greater than half of the distance 3-1, so point 4 is added as the new cluster center, and the distance to other points is calculated. Entering the next loop, comparing the three sets of data, a new min array and Minindex array are obtained, and point 2 is added as the new cluster center. Entering the loop again, the maximum value of the new min array is no longer greater than half of the distance 3-1, and the loop exits. Based on the latest Minindex array, data is extracted from the cluster centers, which are the complete cluster centers. The entire process is listed in Table 1.

[0031] Table 1 lists the results and the entire process of obtaining all cluster centers using the clustering algorithm.

[0032] Step S3: Based on the input constraints and the predetermined target quantity, filter out the outliers.

[0033] Step S4: The naming format of the photoelectric detection images is consistent with the time in the detection target results, and the detection images corresponding to the target are displayed according to the time.

[0034] Step S5: Based on the target data and corresponding images, the target type can be modified. For small targets with large distance deviations, target data with abnormal distances can be deleted. Then, re-clustering is performed.

[0035] This invention includes, but is not limited to, the above embodiments. Any equivalent substitutions or partial improvements made under the spirit and principles of this invention shall be considered within the scope of protection of this invention.

Claims

1. A target clustering method based on multi-type photoelectric detection results from unmanned surface vessels, characterized in that, Includes the following steps: S1: Obtain photoelectric detection result files from the unmanned surface vessel. The files contain detection images and target data associated with each detection time. The target data includes at least the target type, target latitude, target longitude, and detection distance. S2: Classify the target data according to the target type, and perform clustering operation on each class of target data to determine the cluster center of that class of targets; S3: Filter the clustering results according to the preset target number, identify and mark abnormal data points; S4: Based on the detection time, associate and match the detected image with the corresponding target data and display it; S5: Based on the displayed information, manually correct the target data; S6: Re-execute the clustering operation in step S2 using the corrected target data.

2. The method according to claim 1, characterized in that, The clustering operation employs an iterative method to generate cluster centers, specifically including: a. Select a point from the current target dataset as the initial cluster center; b. Calculate the distance between the cluster center and all other points in the dataset, and find the point with the farthest distance; c. Add the farthest point as a new cluster center; d. For each point in the dataset, calculate its minimum distance to all existing cluster centers, and record the minimum distance value and the corresponding cluster center index; e. Determine whether the maximum value of the minimum distance among all points is greater than the preset threshold. If yes, add the point corresponding to the maximum value as a cluster center and return to step d. If no, end the iteration, complete the clustering based on the recorded cluster center index, and output the final set of cluster centers.

3. The method according to claim 2, characterized in that, The threshold is set to half the maximum distance between any two cluster centers in the current cluster center set.

4. The method according to claim 2 or 3, characterized in that, The calculation of the distance between the two points in step d adopts the geographic coordinate projection method based on the Earth ellipsoid model. Specifically, it includes: calculating the projected distance components in the east-west and north-south directions according to the latitude and longitude of the two points, and calculating the straight-line distance between the two points by using the Pythagorean theorem.

5. The method according to claim 1, characterized in that, In step S4, the corresponding detection time is extracted by parsing the naming information of the detection image and matched with the detection time in the target data to realize the automatic association between the detection image and the target data.

6. The method according to claim 1, characterized in that, The manual correction described in step S5 includes at least one of the following operations: correcting erroneous target type labels and deleting distance anomaly data points that are determined to be detection errors.

7. The method according to claim 6, characterized in that, The determination of abnormal distance data points prioritizes detection data for targets that are small.

8. The method according to claim 1, characterized in that, In step S2, the output of the cluster center includes at least the geographical coordinates of each cluster center and the total number of targets in that class.

Images