Target positioning method applied to lobster eye optical detection system

Abstract

The invention provides a target positioning method applied to a lobster eye optical detection system, and the method comprises the steps: converting a target positioning problem of the lobster eye optical detection system into a key point detection problem of a cross focal spot from the image characteristics, and proposing a positioning thought that a key region is screened firstly and then a target source point is extracted. And the calculation principle of the azimuth angle of the target is explored through optical path analysis, so that the accurate position information of the target including the position coordinates and the relative azimuth angle is obtained step by step, the optical detection requirement is better met, the missing detection rate and the false detection rate caused by cross focal spot shape interference and staggered clutter and mutual shielding are reduced, and the detection and positioning effects are better.

Description

Technical Field

[0001] The invention belongs to the technical field of novel photoelectric system detection and target recognition and tracking, and in particular relates to a target positioning method applied to a lobster eye optical detection system. Background Art

[0002] The optical detection system is a system that detects, identifies and monitors objects and environments by collecting, processing and analyzing optical signals. It is a key tool for accurate perception and intelligent decision-making in many fields because of its convenient deployment, wide application range and ability to provide intuitive and detailed information. In modern high-tech warfare, it is an indispensable means of obtaining battlefield situation information; in civilian use, it has also been widely used in many fields such as social security, resource exploration, remote sensing navigation and environmental protection.

[0003] In order to meet complex detection needs, optical detection systems have high requirements in terms of working distance, field of view and image clarity. The lobster eye optical structure has the characteristics of strong adaptability, high directional sensitivity and large working field of view. Compared with traditional optical detection systems, optical detection systems using lobster eye structures are small in size, low in cost and easy to deploy, and have great application potential in the fields of aerial remote sensing, large-scale monitoring and warning, search and navigation. However, the point spread function of the lobster eye optical system is relatively unique, and the image formed has a significant cross feature. Therefore, it is necessary to obtain its position and azimuth information through the cross focal spot formed by the target to achieve the detection purpose.

[0004] In early studies, Zhan et al. proposed an image enhancement and region of interest extraction method for lobster-eye optical images, which consists of three modules: "de-gridding", "low-brightness image reconstruction" and "region of interest (ROI) extraction", providing a new method for the field of lobster-eye image processing; Sawano et al. proposed a lobster-eye source detection algorithm based on one-dimensional projection, which greatly optimized the source detection effect through double condition screening; Liu et al. proposed a lobster-eye telescope target detection idea based on machine learning algorithm, advocated combining multiple image processing methods to construct a lobster-eye target detection framework, and compared and analyzed the performance of different methods; Subsequently, Liu et al. improved the above research and proposed a lobster-eye telescope target detection pipeline algorithm framework based on machine learning algorithm, dividing the source detection of celestial targets into three processes: super-bright, medium-brightness and tiny-brightness detection, and applying different schemes respectively.

[0005] In other words, the current target positioning method based on the lobster eye optical system still has many shortcomings as follows: 1. The algorithm is easily interfered by the cross shape, the detection effect is poor, and the false detection rate is high. 2. It can only determine the target range and approximate position, and it is difficult to accurately locate the target source point. 3. It is limited to the target position information and lacks the target azimuth judgment. 4. It is difficult to solve the problem of cross focus interlacing and occlusion.

[0006] It can be seen that there is currently little research on image processing methods for lobster eye target positioning. Not only does it face the problem of lobster eye cross focal spots being intertwined and blocked from each other, resulting in poor detection effect, but it also cannot meet the demand for accurate positioning of the target source point of the lobster eye optical detection system, greatly limiting its wide application in detection and warning and other fields. Summary of the invention

[0007] In order to solve the problems that the target recognition method of lobster eye image in the prior art cannot solve the problems of cross focal spot distribution being cluttered and occluding each other, accurate positioning of target source point and target orientation judgment, the present invention provides a target positioning method applied to a lobster eye optical detection system. Firstly, the key area is screened, and then the positioning idea of ​​the target point light source is extracted, which can effectively solve the false detection problem caused by target aliasing. The recognition and positioning algorithm has high accuracy and good detection and positioning effect.

[0008] A target positioning method applied to a lobster eye optical detection system comprises the following steps:

[0009] A deep learning-based target detection model is used to extract the cross-focus key area of ​​each point light source in the target image obtained by the lobster-eye optical detection system.

[0010] The centroid of each point light source in the cross focus key area is used as the position coordinate of each point light source on the target image;

[0011] Obtaining the image height of each point light source on the target image according to the position coordinates of each point light source on the target image;

[0012] The azimuth angle of each point light source relative to the lobster eye optical detection system is obtained according to the image height of each point light source on the target image, where the calculation method of the azimuth angle ω' corresponding to any point light source is: Where h is the image height and R is the radius of curvature of the lobster eye optical detection system.

[0013] Furthermore, the method for obtaining the image height h of any point light source on the target image is:

[0014] h=KL

[0015]

[0016] Among them, K is the pixel size of the detector receiving image in the lobster eye optical detection system, and L is the distance between the position coordinates (xi, yi) of the current point light source on the target image and the geometric center point (x0, y0) of the detector receiving image.

[0017] Furthermore, the target detection model is a YOLOv5s target detection model, and the YOLOv5s target detection model includes a CSPDarknet53 structure backbone network, a feature pyramid network, a path aggregation network, a detection head, and a non-maximum suppression network;

[0018] The method for obtaining the cross focus key area of ​​any point light source is as follows:

[0019] The CSPDarknet53 structure backbone network of the YOLOv5s target detection model performs initial feature extraction on the target image;

[0020] The feature pyramid network and the path aggregation network are used in turn to fuse the extracted initial features at different levels to obtain fused features;

[0021] The detection head in the YOLOv5s target detection model predicts the target bounding box of the fused features to obtain the target bounding box corresponding to each point light source and the point light source category score of each target bounding box;

[0022] The category corresponding to the maximum value of the point light source category score of each target bounding box is taken as the point light source category of each target bounding box;

[0023] A non-maximum suppression network is used to remove overlapping target bounding boxes belonging to the same point light source category in each target bounding box. For each point light source, only the target bounding box corresponding to the maximum score of the point light source category is retained to obtain the cross focus key area corresponding to each point light source in the target image.

[0024] Furthermore, the method for obtaining the position coordinates of each point light source on the target image is:

[0025] Gaussian filtering is performed on the key area of ​​the cross focus spot of each point light source to obtain a filtered image that removes scattered noise points and highlights the central bright spot;

[0026] Perform an erosion operation on the filtered image to obtain a connected image with disconnected areas removed;

[0027] Binarization operation is performed on the connected image according to the set threshold value to obtain a binary image highlighting the candidate areas of each point light source;

[0028] Perform a dilation operation on the binary image to obtain a dilated image that merges isolated bright spots;

[0029] Perform contour detection on the dilated image to obtain the contour of each point light source;

[0030] The centroid of each contour is calculated by means of zero-order moment, and the centroid of each contour is used as the position coordinate of the point light source corresponding to each contour on the target image.

[0031] Furthermore, the deep learning-based target detection model is a YOLOv5s target detection model, a YOLOv5x target detection model, a YOLOv8s target detection model, or a YOLOv8n target detection model.

[0032] Furthermore, the working distance of the lobster eye optical detection system is 2 km and the working field angle is 42°.

[0033] Furthermore, a target positioning method applied to the lobster eye optical detection system further includes the following steps:

[0034] A rectangular frame is used to select and a solid circle is used to mark each point light source on the target image. At the same time, additional data labels (xi, yi, ω') are displayed around the rectangular frame of each point light source on the target image.

[0035] Beneficial effects:

[0036] 1. The present invention provides a target positioning method applied to a lobster-eye optical detection system. Starting from image characteristics, the target positioning problem of the lobster-eye optical system is converted into a key point detection problem of a cross-focus spot. A positioning idea of ​​first screening the key area and then extracting the target source point is proposed, and the calculation principle of the target azimuth is explored through optical path analysis, and the acquisition of the target's precise position information including position coordinates and relative azimuth is gradually realized, which is more in line with the needs of optical detection, reduces the missed detection and false detection rate caused by the interference of the cross-focus spot shape and its interlaced, disordered and mutually occluded, and has a better detection and positioning effect.

[0037] 2. The present invention provides a target positioning method applied to the lobster eye optical detection system, which uses a series of processing methods such as morphological operations and contour detection, with small calculation amount and fast calculation speed.

[0038] 3. The present invention provides a target positioning method applied to the lobster eye optical detection system, which selects the YOLOv5s algorithm model to detect the key area of ​​the cross focus spot. It has high detection efficiency, small algorithm size, and high detection efficiency, and can better meet the requirements of lightweight and real-time deployment of the supporting algorithm of the lobster eye optical detection system on the mobile terminal. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 This is a flow chart of the target positioning method of the lobster eye optical detection system described in the present invention.

[0040] Figure 2 This is a flow chart of the target source point positioning module described in the present invention.

[0041] Figure 3 This is a focusing principle diagram of the lobster eye optical system described in the present invention.

[0042] Figure 4 This is a schematic diagram of the parallel light incident focusing principle of the lobster eye optical system described in the present invention.

[0043] Figure 5 This is an imaging simulation diagram of the lobster eye optical detection system in the optical simulation analysis software of the present invention.

[0044] Figure 6 This is the final effect diagram of the target positioning method of the lobster eye optical detection system described in the present invention. DETAILED DESCRIPTION

[0045] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0046] In view of the many problems existing in the existing lobster eye optical system target detection technology, the main purpose of the present invention is to provide a lobster eye optical detection system target positioning method which integrates the YOLOv5 target detection model with morphological operations and other processing means and can simultaneously obtain comprehensive information of target position and azimuth.

[0047] Starting from the characteristics of lobster eye images, the present invention proposes a set of target positioning algorithm process design ideas suitable for the lobster eye optical detection system, which can solve various problems such as the interlaced occlusion of cross focal spots in lobster eye images, precise positioning of target source points, and azimuth judgment of targets relative to the detection system, and obtain the precise position information of each target within the detection range.

[0048] like Figure 1 As shown, the present invention provides a target positioning method applied to a lobster eye optical detection system, characterized in that it includes the following steps:

[0049] S1: A deep learning-based target detection model is used to extract the cross-focus key area of ​​each point light source in the target image obtained by the lobster-eye optical detection system;

[0050] The following takes the target detection model as the YOLOv5s target detection model as an example to describe in detail the process of obtaining the cross focus key area of ​​each point light source. The YOLOv5s target detection model includes a CSPDarknet53 structure backbone network, a feature pyramid network, a path aggregation network, a detection head, and a non-maximum suppression network; wherein, the method for obtaining the cross focus key area of ​​any point light source is:

[0051] S11: The CSPDarknet53 structure backbone network of the YOLOv5s target detection model performs initial feature extraction on the target image;

[0052] S12: Use the feature pyramid network and the path aggregation network in turn to fuse the extracted initial features at different levels to obtain fused features;

[0053] S13: The detection head in the YOLOv5s target detection model predicts the target bounding box of the fused features to obtain the target bounding box corresponding to each point light source and the point light source category score of each target bounding box;

[0054] S14: taking the category corresponding to the maximum value of the point light source category score of each target bounding box as the point light source category of each target bounding box;

[0055] S15: A non-maximum suppression network is used to remove overlapping target bounding boxes belonging to the same point light source category in each target bounding box, and only the target bounding box corresponding to the maximum score of the point light source category is retained for each point light source, so as to obtain the cross focus key area corresponding to each point light source in the target image.

[0056] That is to say, the cross focus in the lobster eye image is usually composed of a central bright spot and cross arms, and the source point representing the actual position of the target is located in the central bright spot area. The YOLOv5s target detection algorithm model is used to detect the cross focus key area containing the source point: first, the input image is preprocessed, the image size is adjusted, and the pixel value is normalized; the CSPDarknet53 structure backbone network (Backbone) of YOLOv5s is used to extract the features of the cross focus key area, and FPN (Feature Pyramid Network) and PAN (Path Aggregation Network) are used to fuse features at different levels; then, the detection head (Head) in the YOLOv5s target detection model predicts the target bounding box and its category score; finally, non-maximum suppression (NMS) processing is performed to remove overlapping target bounding boxes, retain the optimal detection result, and obtain the target box information of all cross focus key areas in the image for subsequent processing.

[0057] It should be noted that the YOLOv5s detection model can be replaced with other models in the YOLO series, such as YOLOv5x, YOLOv8s, YOLOv8n, etc., which can all achieve feature extraction, and the present invention will not go into details.

[0058] S2: taking the centroid of each point light source in the cross focus key area as the position coordinate of each point light source on the target image;

[0059] It should be noted that the central source point of the cross focus of the lobster eye is generally located at the center of the highlight spot area. The present invention intercepts each central highlight spot area according to the target frame information provided by the cross focus key area detection module, and performs the following processing operations on the intercepted image area in sequence: Gaussian filtering denoising and smoothing, corrosion operation to eliminate non-connected areas, specifying a threshold value to binarize and screen the candidate target source point area, expansion operation to merge independent bright spots, contour detection to extract the contour information of the candidate area, and extracting the contour centroid as the target source point;

[0060] Specifically, Figure 2 As shown, the method for obtaining the position coordinates of each point light source on the target image is:

[0061] S21: Gaussian filtering is performed on the key area of ​​the cross focal spot of each point light source to obtain a filtered image with scattered noise removed and the central bright spot highlighted; preferably, the captured image is convolved with a 3×3 pixel two-dimensional Gaussian kernel to remove scattered noise and highlight the central bright spot.

[0062] S22: performing an erosion operation on the filtered image to obtain a connected image with disconnected areas removed;

[0063] S23: performing a binarization operation on the connected image according to a set threshold value to obtain a binary image highlighting each point light source candidate area;

[0064] S24: performing a dilation operation on the binary image to obtain a dilated image that merges isolated bright spots;

[0065] S25: performing contour detection operation on the dilated image to obtain the contour of each point light source;

[0066] S26: Calculate the centroid of each contour respectively by means of zero-order moment, and use the centroid of each contour as the position coordinate of the point light source corresponding to each contour on the target image;

[0067] S3: Obtaining the image height of each point light source on the target image according to the position coordinates of each point light source on the target image;

[0068] S4: Obtain the azimuth angle of each point light source relative to the lobster eye optical detection system according to the image height of each point light source on the target image, wherein the calculation method of the azimuth angle ω' corresponding to any point light source is: Where h is the image height and R is the radius of curvature of the lobster eye optical detection system.

[0069] It should be noted that the conventional focusing imaging principle of the lobster eye optical detection system is as follows Figure 3As shown. Where y represents the image distance of the lobster eye optical system, x represents the object distance, and R is the radius of curvature of the lobster eye optical system. The focal length f of the lobster eye optical system is R / 2, and the object-image relationship is as follows:

[0070]

[0071] When the application scenario is long-distance and large-field optical detection with a working distance of 2km and a working field angle of 42°, it can be assumed that the light source representing the target object enters the system in the form of parallel light, the object distance x of the lobster eye optical detection system tends to positive infinity, and the image distance y is R / 2. At this time, the parallel light incident situation of the lobster eye optical detection system can be simplified as follows Figure 4 As shown. Figure 4 In the equation, y is the image distance and h is the image height. Then the azimuth angle of the target object relative to the detector and the image height h formed by it will have a geometric relationship as shown in the following formula:

[0072]

[0073] By obtaining the image height information h, the azimuth angle ω′ of the target relative to the detector can be calculated according to the above formula to determine the position of the target object. The image height h formed by the target object can be obtained by processing the image received by the detector. ω′ is the object field angle, and y is the image distance. When the incident light is parallel, the image distance y is R / 2. Substituting R / 2 into the above series of formulas, the azimuth angle calculation formula is as follows:

[0074] As mentioned above, the coordinates (xi, yi) of each target source point in the detector received image have been obtained. In this embodiment, the geometric center of the detector received image corresponds to the center of the lobster eye lens in the detection system. Therefore, given the pixel size K of the detector received image and the distance L between the i-th target source point (xi, yi) and the geometric center point (x0, y0) of the image received by the detector, the actual image height h of the lobster eye optical detection system can be calculated as:

[0075]

[0076] h=KL

[0077] The target azimuth angle ω' can be obtained according to the image height h and the curvature radius R of the lobster eye lens. Finally, data integration and visualization are performed to summarize all target position coordinates and azimuth angle information and display them in the form of (xi, yi, ω'). The upper limit of i is the total number of point light sources in the current target image.

[0078] Next, we will build a lobster eye optical detection system in the optical simulation analysis software and perform imaging simulation analysis. Figure 5The simulation image containing five targets to be located is a sample. In this embodiment, the sample image size is 640×640 pixels, the detector pixel size of the lobster eye optical detection system is 2.4 μm, and the curvature radius of the lobster eye lens is 500 mm. The specific implementation content is as follows:

[0079] 1. Cross focus key area detection based on YOLOv5s algorithm model

[0080] The cross focus key area detection described in this embodiment is used to detect and filter out the key areas containing the source points in each cross focus in the image before locating the target source point, and the YOLOv5s detection algorithm model is used to perform screening detection of the cross focus key areas to obtain the target frame information.

[0081] According to the simulation results of the optical simulation analysis software, a lobster eye image dataset was constructed for training and testing the YOLOv5s detection model. At the same time, combined with the image feature analysis, random scaling, random rotation, Mosaic data enhancement, MixUp data enhancement, and HSV color space transformation data enhancement methods were used in training to improve the model performance. Finally, the trained model was used to detect the sample image and extract the target box information of the detection result for subsequent processing.

[0082] 2. Target source location based on morphological operations

[0083] The central source point of the cross focus of the lobster eye is generally located at the center of the highlight spot area. According to the target frame information provided by the cross focus key area detection module, the key area is intercepted and the source point is located. The specific steps are as follows:

[0084] (1) Gaussian filtering: Preferably, the captured image is convolved with a 3×3 pixel two-dimensional Gaussian kernel to remove scattered noise points and highlight the central bright spot.

[0085] (2) Erosion: Preferably, an erosion operation is performed on the Gaussian filtered image, and the kernel size is selected as 2×2 pixels to further remove stray light and other small non-connected areas.

[0086] (3) Specifying a threshold for binarization: The pixel with the maximum gray value of the cross focus spot will appear at the center of the bright spot. Preferably, the threshold is set to 95% of the maximum gray value of the current captured image, and a binarization operation is performed to obtain the target source point candidate area.

[0087] (4) Dilation: Multiple bright spots may appear in the center of the binarized bright spot. Preferably, the binarized images are dilated to merge them, and the kernel size is set to 2×2 pixels, which is close to the cross arm width.

[0088] (5) Contour detection: After the dilation operation, a binary image is obtained. The openCV toolkit is used to perform contour detection on the binary image to obtain the contour information of the candidate area of ​​the target source point.

[0089] (6) Centroid extraction: After obtaining the contour of the central area of ​​the cross-focus bright spot, preferably, the centroid is extracted by calculating the zero-order moment of the contour as the central source point of the target cross-focus.

[0090] 3. Target azimuth calculation of lobster-eye optical detection system

[0091] In this embodiment, the detector pixel size K is 2.4 μm, the curvature radius R of the lobster eye lens is 500 mm, and the calculation formulas of the distance L between the i-th target source point (xi, yi) and the image geometric center point (x0, y0) in the detector received image, the image height h and the target azimuth angle ω' are respectively as follows:

[0092]

[0093] h=KL

[0094]

[0095] According to the coordinates (xi, yi) of each source point acquired by the target source point positioning module and the above formula, the corresponding azimuth angle ω' is calculated, and the comprehensive position coordinate information and azimuth angle information are stored and visualized in the form of (xi, yi, ω'). In this embodiment, the final positioning effect of the lobster eye optical detection system target positioning method is as follows: Figure 6 As shown by Figure 5 , Figure 6 It can be seen that the five targets to be located in the sample image are successfully detected and accurately located, and are returned in the sample image in the form of rectangular frame selection, solid circle marking target source point, and additional (xi, yi, ω') labels. At this time, the sequence number of the target to be located i = 1, 2, 3, 4, 5.

[0096] In summary, the target positioning method applied to the lobster eye optical detection system provided by the present invention can be summarized as follows:

[0097] The target image is first detected by the YOLOv5s target detection algorithm model for cross focus key area detection. After image preprocessing to eliminate background light and noise interference, the CSPDarknet53 structure Backbone network of YOLOv5s and the Neck network combined with the feature pyramid network (FPN) and path aggregation network (PAN) are used to extract the cross focus features in the image. The detection box selects the key areas of each cross focus and obtains the corresponding target box information for more accurate target positioning processing in subsequent steps.

[0098] The key area is intercepted according to the target frame information obtained by the key area detection module, and the center source point of the cross focus representing the actual position of the target is extracted in the intercepted area to obtain the position coordinate information of the target.

[0099] Based on the fact that the position of the target cross focus source point can reflect the actual azimuth of the target relative to the detector, the present invention performs optical path analysis on the lobster eye optical detection system and derives the target azimuth calculation formula. According to the azimuth calculation formula, the corresponding azimuth is calculated in combination with the acquired target source point coordinates to obtain the relative azimuth information of the target.

[0100] After the above processing, data integration and visualization are carried out to integrate the target position coordinates and orientation information, and finally achieve accurate positioning of the detected target.

[0101] It can be seen that compared with the conventional lobster eye image target detection method, the process design idea described in the present invention starts from the lobster eye image characteristics, and converts the target positioning problem of the lobster eye optical detection system into a key point detection problem of the cross focal spot, thereby realizing the acquisition of precise target position information including position coordinates and their relative azimuth, and solving the problems of messy distribution of the target cross in the image and mutual occlusion and interference of the cross arms. The false detection rate is low, the detection and positioning effect is good, and it is more in line with the needs of optical detection.

[0102] Of course, the present invention may have many other embodiments. Without departing from the spirit and essence of the present invention, those skilled in the art may certainly make various corresponding changes and modifications based on the present invention, but these corresponding changes and modifications should all fall within the scope of protection of the claims attached to the present invention.

Claims

1. A target positioning method applied to a lobster eye optical detection system, characterized in that: The following steps are involved: A deep learning-based target detection model is used to extract the cross-focus key area of ​​each point light source in the target image obtained by the lobster-eye optical detection system. The centroid of each point light source in the cross focus key area is used as the position coordinate of each point light source on the target image; Obtaining the image height of each point light source on the target image according to the position coordinates of each point light source on the target image; The azimuth angle of each point light source relative to the lobster eye optical detection system is obtained according to the image height of each point light source on the target image, where the calculation method of the azimuth angle ω' corresponding to any point light source is: Where h is the image height and R is the radius of curvature of the lobster eye optical detection system.

2. A target positioning method applied to a lobster eye optical detection system as claimed in claim 1, characterized in that: The method for obtaining the image height h of any point light source on the target image is: h=KL Among them, K is the pixel size of the detector receiving image in the lobster eye optical detection system, and L is the distance between the position coordinates (xi, yi) of the current point light source on the target image and the geometric center point (x0, y0) of the detector receiving image.

3. A target positioning method applied to a lobster eye optical detection system as claimed in claim 1, characterized in that: The target detection model is a YOLOv5s target detection model, and the YOLOv5s target detection model includes a CSPDarknet53 structure backbone network, a feature pyramid network, a path aggregation network, a detection head, and a non-maximum suppression network; The method for obtaining the cross focus key area of ​​any point light source is as follows: The CSPDarknet53 structure backbone network of the YOLOv5s target detection model performs initial feature extraction on the target image; The feature pyramid network and the path aggregation network are used in turn to fuse the extracted initial features at different levels to obtain fused features; The detection head in the YOLOv5s target detection model predicts the target bounding box of the fused features to obtain the target bounding box corresponding to each point light source and the point light source category score of each target bounding box; The category corresponding to the maximum value of the point light source category score of each target bounding box is taken as the point light source category of each target bounding box; A non-maximum suppression network is used to remove overlapping target bounding boxes belonging to the same point light source category in each target bounding box. For each point light source, only the target bounding box corresponding to the maximum score of the point light source category is retained to obtain the cross focus key area corresponding to each point light source in the target image.

4. A target positioning method for a lobster eye optical detection system as claimed in claim 1, characterized in that: The method for obtaining the position coordinates of each point light source on the target image is: Gaussian filtering is performed on the key area of ​​the cross focus spot of each point light source to obtain a filtered image that removes scattered noise points and highlights the central bright spot; Perform an erosion operation on the filtered image to obtain a connected image with disconnected areas removed; Binarization operation is performed on the connected image according to the set threshold value to obtain a binary image highlighting the candidate areas of each point light source; Perform a dilation operation on the binary image to obtain a dilated image that merges isolated bright spots; Perform contour detection on the dilated image to obtain the contour of each point light source; The centroid of each contour is calculated by means of zero-order moment, and the centroid of each contour is used as the position coordinate of the point light source corresponding to each contour on the target image.

5. A target positioning method applied to a lobster eye optical detection system as claimed in any one of claims 1, characterized in that: The deep learning-based target detection model is the YOLOv5s target detection model, the YOLOv5x target detection model, the YOLOv8s target detection model, or the YOLOv8n target detection model.

6. A target positioning method applied to a lobster eye optical detection system as claimed in any one of claims 1 to 5, characterized in that: The lobster eye optical detection system has a working distance of 2 km and a working field angle of 42°.

7. A target positioning method applied to a lobster eye optical detection system as claimed in any one of claims 1 to 5, characterized in that: The following steps are also included: A rectangular frame is used to select and a solid circle is used to mark each point light source on the target image. At the same time, additional data labels (xi, yi, ω') are displayed around the rectangular frame of each point light source on the target image.

Images