A continuous small target real-time detection method based on sonar image

Abstract

The application belongs to the technical field of target detection, and relates to a small target real-time detection method based on a sonar image, which comprises the following steps: 1) performing enhancement and segmentation on the sonar image to obtain target intensity, target position and target diffusion range; 2) restoring the target shape of the target in the sonar image based on the target position; and 3) based on the target shape and the target intensity, target position and target diffusion range, adopting target multi-frame detection to realize real-time detection of small targets. The method does not depend on large parallel computing devices, is easy to deploy, has small computing amount and fast running speed, can meet the run-time detection of the device on a CPU, has few preset parameters and clear meanings, can be adjusted in time according to the use effect, and is convenient to use. The method adopts multi-frame detection, retains historical target records, can retain target positions when the image is subjected to temporary interference, and has strong anti-interference capability.

Description

Technical Field

[0001] This invention belongs to the field of target detection technology, and relates to a real-time detection method for small targets, and more particularly to a real-time detection method for continuous small targets based on sonar images. Background Technology

[0002] For small targets in sonar images (targets smaller than 100 pixels), current technologies typically rely on deep learning models for detection. However, deep learning models require large labeled datasets for training, making effective training difficult for tasks targeting specific targets. Furthermore, deep learning models need to be deployed on parallel computing devices with sufficient computational power, placing high demands on the sonar platform. Finally, current mainstream methods primarily rely on single-frame detection, lacking active reference to historical target information, resulting in ineffective detection in interfering frames and temporary target loss during runtime.

[0003] Therefore, in order to address the shortcomings of the existing technologies, it is necessary to develop a novel real-time detection method for continuous small targets based on sonar images. Summary of the Invention

[0004] To overcome the shortcomings of existing technologies, this invention proposes a method for real-time detection of continuous small targets based on sonar images, which can achieve continuous real-time detection of small targets.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A method for real-time detection of small targets based on sonar images, characterized by the following steps:

[0007] 1) Enhance and segment the sonar image to obtain target intensity, target location and target diffusion range: enhance the contrast of the sonar image, enhance the pixel intensity value of the local extreme point region to obtain an enhanced image, set a dynamic threshold according to the pixel intensity value of the enhanced image and perform binary segmentation with the dynamic threshold, perform connected component analysis on the segmentation result, take the point with the largest pixel intensity value in the connected component as the target location and its pixel intensity value as the target intensity, and take the standard deviation of the pixel intensity value of the segmented region as the target diffusion range;

[0008] 2) Recover the target shape in the sonar image based on the target location: Calculate the boundary threshold for each target and set a list to record the points belonging to the target. Find the points with pixel intensity values ​​greater than the boundary threshold by breadth-first traversal and add them to the list. For points with pixel intensity values ​​less than the boundary threshold, perform linear interpolation in the horizontal and vertical directions to calculate the superpixel coordinates of the boundary threshold. If the position of the boundary threshold is closer to the point with a value less than the boundary threshold, add the point to the list. Recover the target shape in the sonar image based on the points in the list.

[0009] 3) Based on the target shape, target intensity, target location, and target diffusion range, multi-frame target detection is used to achieve real-time detection of small targets: First, the target in the sonar image of the current frame is compared with the targets in the trusted target list and the target to be observed list. When there is a feature match between the two, the targets in the trusted target list and the target to be observed list are updated. For the current target that does not have a feature match in the trusted target list and the target to be observed list, it is regarded as a new target. If the new target appears at the far end of the sonar image of the current frame, it is added to the target to be observed list; otherwise, it is regarded as a false alarm and discarded. Then, the target to be observed list is checked, and the targets that are updated stably are moved to the trusted target list. Targets that cannot be updated for multiple consecutive frames are regarded as false alarms and discarded. Finally, the trusted target list is checked. For targets that have not been updated, their boundaries are searched according to their target locations. If no boundary is found, the current target location and boundary are retained.

[0010] Preferably, step 1) of enhancing the contrast of the sonar image and increasing the pixel intensity values ​​in local extreme point regions to obtain an enhanced image specifically includes:

[0011] 11) Calculate the global contrast of the pixels in the sonar image, and denot it as the global saliency map;

[0012] 12) Calculate the local contrast of the pixels in the sonar image and record it as a local saliency map;

[0013] 13) The global saliency map and the local saliency map are weighted and superimposed to form an enhanced image.

[0014] Alternatively, step 1) may further include the following steps:

[0015] 11) Calculate the global contrast of the pixels in the sonar image, and denot it as the global saliency map;

[0016] 12) Calculate the local contrast of the pixels in the sonar image and record it as a local saliency map;

[0017] 13) Perform convolution operation on the sonar image data using object detection convolution kernel group, and take the maximum value of the component at each pixel position in the obtained convolution image group to obtain the object saliency map.

[0018] 14) The global saliency map, local saliency map and object saliency map are weighted and superimposed to form an enhanced image.

[0019] Preferably, in step 11), the global pixel intensity of the sonar image is represented by a Gaussian probability model, wherein the Gaussian probability model is:

[0020]

[0021] In the formula, x represents the pixel value, σ represents the standard deviation of the pixel value, and μ represents the mean of the pixel value.

[0022] Preferably, in step 12), the global contrast of the pixels in the sonar image is:

[0023]

[0024] In the formula, xy represents the position of the pixel in the sonar image, and P xy This represents the pixel value at that location. This represents the global contrast at that location, and model(·) represents the Gaussian probability model obtained in step 11).

[0025] Preferably, in step 13), the local contrast of the pixels in the sonar image is calculated by: processing the sonar image using Gaussian filtering and mean filtering, and taking the absolute value of the difference between the two filtered images as the local contrast of the pixels in the sonar image.

[0026] Preferably, in step 1), the dynamic threshold is:

[0027]

[0028] In the formula, threshold(α) is the dynamic threshold, and α is the threshold coefficient in the range [0,1], representing the linear position of the dynamic threshold from the mean to the maximum value. To enhance the average pixel intensity value of an image, P max To enhance the maximum pixel intensity value of the image.

[0029] Preferably, in step 2), the boundary threshold is:

[0030] border = 0.707p obj

[0031] In the formula, border is the boundary threshold of the target, pobj The target strength is the target intensity.

[0032] Preferably, in step 3), the features include target intensity and target location, and if the difference in target intensity between the target in the sonar image of the current frame and the targets in the list of trusted targets and the list of targets to be observed is within the target diffusion range and the distance between the target locations does not exceed a distance threshold, then the features are considered to match.

[0033] Preferably, in step 3), if a new target appears at the far end of the sonar image of the current frame, it is considered to be a new target entering the sonar detection range and is added to the list of targets to be observed; otherwise, it is considered a false alarm and is discarded.

[0034] Compared with the prior art, the real-time detection method for continuous small targets based on sonar images of the present invention has one or more of the following beneficial technical effects:

[0035] 1. This invention is based on traditional algorithms, does not rely on large-scale parallel computing devices, is easy to deploy, has a small computational load, runs fast, and can meet the device's runtime detection requirements on the CPU.

[0036] 2. The preset parameters of this invention are few and clearly defined, allowing users to adjust them promptly based on the usage effect, making it easy to use.

[0037] 3. This invention uses multi-frame detection, which preserves historical target records and can retain the target position when the image is briefly disturbed, thus having strong anti-interference ability. Attached Figure Description

[0038] Figure 1 This is a flowchart of the real-time detection method for continuous small targets based on sonar images according to the present invention.

[0039] Figure 2 This is a schematic diagram of an object detection convolution kernel group according to a preferred embodiment of the present invention. Detailed Implementation

[0040] Before detailing any embodiment of the invention, it should be understood that the invention, in its application, is not limited to the details of the construction and arrangement of the components set forth in the following description or illustrated in the following figures. The invention can have other embodiments and can be practiced or carried out in various ways. Furthermore, it should be understood that the wording and terminology used herein are for descriptive purposes and should not be considered limiting. The use of “comprising” or “having” and variations thereof is intended to cover the items set forth below and their equivalents, as well as any additional items. Unless otherwise specified or limited, the terms “installation,” “connection,” “support,” and “linkage,” and variations thereof are used broadly and cover both direct and indirect installation, connection, support, and linking. Moreover, “connection” and “linkage” are not limited to physical or mechanical connections or links.

[0041] Furthermore, firstly, in the disclosure of this invention, the terms "longitudinal," "lateral," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the above terms should not be construed as limiting this invention. Secondly, the term "a" should be understood as "at least one" or "one or more," that is, in one embodiment, the number of an element can be one, while in another embodiment, the number of the element can be multiple. The term "a" should not be construed as a limitation on the quantity.

[0042] Figure 1 A flowchart of the real-time detection method for continuous small targets based on sonar images according to the present invention is shown. Figure 1 As shown, the real-time small target detection method based on sonar images of the present invention includes the following steps:

[0043] 1. Acquire a single frame of detected sonar image, and enhance and segment the sonar image to obtain target intensity, target location, and target diffusion range.

[0044] In this invention, a single-frame detected sonar image is acquired, and the contrast of the sonar image is enhanced to increase the pixel intensity value of local extreme point regions to obtain an enhanced image. Subsequently, a dynamic threshold is set based on the pixel intensity value of the enhanced image, and the enhanced image is binary segmented using the dynamic threshold. Pixels with intensity below the threshold are discarded, while those above the threshold are retained. Connectivity analysis is performed on the segmentation results, and the point with the largest pixel intensity value in the connected region is taken as the target location, and its pixel intensity value is taken as the target intensity. The standard deviation of the pixel intensity values ​​in the segmented region is taken as the target diffusion range.

[0045] Specifically, enhancing the contrast of the sonar image and increasing the pixel intensity values ​​in local extreme point regions to obtain an enhanced image includes:

[0046] 1. Calculate the global contrast of the pixels in the sonar image, denoted as the global saliency map. Specifically, first calculate the global pixel intensity of the sonar image to represent the overall pixel intensity level of the sonar image. Then, calculate the global contrast of the pixels in the sonar image based on the overall pixel intensity level of the sonar image to obtain the global saliency map.

[0047] In a specific example of the present invention, the global pixel intensity of the sonar image is represented by a Gaussian probability model, that is, the global pixel intensity of the sonar image is estimated by calculating the mean and standard deviation of the pixel values ​​of all pixels in the sonar image.

[0048] The Gaussian probability model is as follows:

[0049]

[0050] In the formula, x represents the pixel value, σ represents the standard deviation of the pixel value, and μ represents the mean of the pixel value.

[0051] In this invention, the pixel value in the Gaussian probability model is calculated and its negation is used to represent the global contrast of that pixel, as shown in the following formula:

[0052]

[0053] In the formula, xy represents the position of the pixel in the sonar image, and P xy This represents the pixel value at that location. This represents the global contrast at that location, and model(·) represents the Gaussian probability model for estimating the global pixel intensity of the sonar image in step 1.

[0054] 2. Calculate the local contrast of the pixels in the sonar image and record it as a local saliency map.

[0055] In this invention, Gaussian filtering and mean filtering are used to process the sonar image to obtain a Gaussian-filtered image and a mean-filtered image. The absolute value of the difference between the two filtered images is taken as the local contrast of the pixels in the sonar image.

[0056] Among them, Gaussian filtering represents the Gaussian estimation of pixels within a local range (the pixel value is not used directly in order to eliminate abnormal changes in pixel value caused by noise signals), that is, the estimated value calculated by the Gaussian probability model in step 1; mean filtering represents the mean within a local range.

[0057] 3. The global saliency map and the local saliency map are weighted and superimposed to form an enhanced image.

[0058] Specifically, the global saliency map and the local saliency map are normalized to unify the numerical domain, and then the global saliency map and the local saliency map are weighted and superimposed to obtain the enhanced image.

[0059] Once the enhanced image is obtained, a dynamic threshold can be set based on the pixel intensity values ​​of the enhanced image. The dynamic threshold is set as follows:

[0060]

[0061] In the formula, threshold(α) is the dynamic threshold, and α is the threshold coefficient in the range [0,1], representing the linear position of the dynamic threshold from the mean to the maximum value. To enhance the average pixel intensity value of an image, P max To enhance the maximum pixel intensity value of the image.

[0062] In another preferred embodiment of the invention, after step 2, step 4 is further included: utilizing the appended... Figure 2 The object detection convolution kernel group shown performs a convolution operation on the sonar image data, and takes the maximum value of the component at each pixel position in the resulting convolution image group to obtain an object saliency map.

[0063] Thus, in this preferred embodiment, corresponding to step 3, the global saliency map, local saliency map, and object saliency map are weighted and superimposed to form an enhanced image.

[0064] Specifically, the global saliency map, local saliency map, and object saliency map are normalized to unify the numerical domain, and then the global saliency map, local saliency map, and object saliency map are weighted and superimposed to obtain the enhanced image.

[0065] It is worth mentioning that the global saliency map and the local saliency map are used for global contrast enhancement, while the object saliency map is used to enhance the contrast of small connected regions. In a specific example of the present invention, the weights of the global saliency map and the local saliency map can be set to 0.2 respectively, and the weight of the object saliency map can be set to 0.6. The enhanced image is obtained by weighting and superimposing the global saliency map, the local saliency map, and the object saliency map according to the above weights, so as to highlight small targets in the enhanced image.

[0066] The dynamic threshold set in this invention can effectively address the problem of fixed threshold failure caused by fluctuations in the background intensity of sonar images after interference with the sonar equipment. When the background intensity of the sonar image fluctuates, both the average pixel intensity value and the maximum pixel intensity value will shift accordingly. At this time, the dynamic threshold shifts accordingly, which can effectively reduce the interference of sonar image intensity fluctuations on the fixed threshold.

[0067] With the dynamic threshold, the intensity image is binary segmented using the dynamic threshold. Connectivity analysis is performed on the segmentation result (the specific connectivity analysis method belongs to the prior art, and for the sake of simplicity, the present invention will not describe it in detail). The point with the largest pixel intensity value in the connected region is taken as the target position and its pixel intensity value is taken as the target intensity. The standard deviation of the pixel intensity values ​​in the segmented region is taken as the target diffusion range.

[0068] For example, a series of forward-looking sonar images are obtained, which include not only the target but also parts of the riverbank and underwater reflections. Assuming the sonar images are uploaded and recorded at 5Hz, the following uses one frame of the sonar image as an example to illustrate how to process them.

[0069] First, a sonar image frame is processed to calculate its global saliency map, local saliency map, and object saliency map. These are then weighted and superimposed to obtain the enhanced image of the frame. Next, a dynamic threshold of 0.3 is set and used to segment the enhanced image. Then, connected component analysis is performed on the segmentation results. Specifically, based on the ratio of the maximum sonar image range of 100 meters to the pixel radius of 500 pixels, the distance corresponding to a unit pixel in the sonar image is calculated to be 0.2 meters. The area of ​​the connected components is calculated accordingly. A segmentation threshold of three times the estimated target's projected area is used, discarding regions with areas larger than the threshold. For example, if a target's pixel area is 26, the corresponding area is 5.2 square meters. If the estimated target size is 5 square meters, the segmented region is retained. Then, the maximum pixel intensity and its position in each remaining connected region are recorded as the target intensity and target position, respectively. The pixel value at the target position is the pixel value of a small target. The minimum bounding rectangle is calculated and recorded as the target size. The standard deviation of the pixel intensity values ​​in the segmented region is calculated and recorded as the target diffusion range.

[0070] 2. Recover the target shape in the sonar image based on the target location.

[0071] Since dynamic thresholding based on enhanced images can distort the original shape of the target to some extent, it is necessary to restore the target shape in the original sonar image according to the target location. This invention sets a threshold based on the pixel intensity value at the target location, divides the target boundary in superpixel space, and then uses linear interpolation to sample the boundary into pixel space.

[0072] Specifically, for each target, a boundary threshold is calculated, and a list is created to record the points belonging to that target. Points with pixel intensity values ​​greater than the boundary threshold are found using a breadth-first search and added to the list. For points with pixel intensity values ​​less than the boundary threshold, linear interpolation is performed in the horizontal and vertical directions to calculate the superpixel coordinates of the boundary threshold. If the boundary threshold is closer to a point with a value less than the threshold, that point is added to the list. Thus, all boundary points of the target can be obtained, i.e., the target boundary can be obtained. The target shape in the sonar image can then be reconstructed using the points in the list.

[0073] In this invention, the boundary threshold is set in the following manner:

[0074] border = 0.707p obj

[0075] In the formula, border is the pixel value threshold at the target location, and p obj This is the pixel value at the target location, i.e., the maximum pixel intensity.

[0076] For example, in a specific example of the present invention, the target positions obtained in step one are traversed, and the target boundary threshold is calculated based on the pixel intensity at the target position. The target boundary position in the superpixel space is obtained through linear calculation at the target boundary, and the target boundary position is sampled into the pixel space through linear interpolation to update the boundary of the small target, thereby realizing the target shape restoration of the small target. For example, in a more specific example of the present invention, the calculated target boundary threshold is 1.8, the pixel value at coordinate (1,1) is 1, and the pixel value at coordinate (1,2) is 2. Then the coordinates of the target boundary position in the superpixel space are (1,1.18). Subsequently, linear interpolation is performed on the coordinates (1,1.18) to obtain the target boundary coordinates in the pixel space as (1,2).

[0077] Third, based on the target shape, target intensity, target location, and target diffusion range, multi-frame target detection is used to achieve real-time detection of small targets.

[0078] Targets detected in a single frame inevitably contain false alarms, and sudden acceleration of the vehicle or interference with sonar can affect imaging performance. To address this, this invention uses multi-frame detection to improve target performance. Therefore, this invention maintains two target lists: a list of trusted targets and a list of targets to be observed. The list of trusted targets stores features of trusted targets, such as their shape (i.e., target boundary), location, intensity, diffusion range, and update records. The list of targets to be observed stores features of targets to be observed, such as their shape (i.e., target boundary), location, intensity, diffusion range, and update records.

[0079] When using multi-frame target detection to achieve real-time detection of small targets, this invention first traverses the targets containing target boundaries obtained in step two, and compares them sequentially with the targets in the list of trusted targets and the list of targets to be observed.

[0080] When a feature match occurs between the two, the targets in the trusted target list and the target list to be observed are updated. That is, if the feature of a target containing the target boundary obtained in step two matches the feature of a target in the trusted target list, then that target in the trusted target list is updated; if the feature of a target containing the target boundary obtained in step two matches the feature of a target in the target list to be observed, then that target in the target list to be observed is updated.

[0081] A current target whose features do not match in either the trusted target list or the target to be observed list is considered a new target. That is, if the features of a target with a target boundary obtained in step two do not match the features of any target in the trusted target list, nor do they match the features of any target in the target to be observed list, then the target with a target boundary obtained in step two is considered a new target.

[0082] The features include target intensity and target location. Furthermore, if the difference in target intensity between a target in the current frame's sonar image and targets in the list of trusted targets and the list of targets to be observed is within the target diffusion range, and the distance between the target locations does not exceed a distance threshold, for example, 0.5 meters, then the features are considered to match.

[0083] If a new target appears at the far end of the sonar image in the current frame, i.e. at the edge of the sonar image, for example, within 5% of the range, it is considered a new target entering the sonar detection range and is added to the list of targets to be observed; otherwise, it is considered a false alarm and is discarded.

[0084] The list of targets to be observed is then checked, and targets that update stably are moved to the list of trusted targets. Targets that fail to update for multiple consecutive frames are considered false alarms and discarded. In a specific example of this invention, the trust threshold can be set to 5, and the discard threshold to 8. That is, a target that updates successfully for 5 consecutive frames is considered to be stable and moved to the list of trusted targets, while a target that fails to update for 8 consecutive frames is considered a false alarm due to flickering and is discarded. The trust threshold is merely an example, and other values ​​can be selected according to actual needs.

[0085] Finally, the list of trusted targets is checked. For targets that have not been updated, their boundaries are found based on their locations and then updated. If no boundary is found, the current target location and its boundaries are retained.

[0086] In the process of finding the boundary, the method of finding the target boundary in step two can be used.

[0087] The present invention uses the targets in the trusted list as the target detection results for this frame.

[0088] This invention is based on traditional algorithms, does not rely on large-scale parallel computing devices, is easy to deploy, has low computational load, and runs quickly, meeting the runtime detection requirements of devices on a CPU. Furthermore, the invention has few preset parameters with clear meanings, allowing users to adjust them promptly based on performance, making it easy to use. Moreover, this invention employs multi-frame detection, preserving historical target records and retaining target positions even when the image is briefly disturbed, demonstrating strong anti-interference capabilities.

[0089] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit the scope of protection of the present invention. Those skilled in the art can modify or make equivalent substitutions to the technical solutions of the present invention based on the concept of the present invention, without departing from the essence and scope of the technical solutions of the present invention.

Claims

1. A method for real-time detection of small targets based on sonar images, characterized in that, Includes the following steps: 1) Acquire a single frame of sonar image, enhance and segment the sonar image to obtain target intensity, target location and target diffusion range. Specifically, acquire a single frame of sonar image, enhance the contrast of the sonar image, enhance the pixel intensity value of local extreme point region to obtain enhanced image, then set a dynamic threshold according to the pixel intensity value of enhanced image and perform binary segmentation on enhanced image with the dynamic threshold. The part of pixel intensity below the threshold is discarded, and the part above the threshold is retained. Perform connected component analysis on the segmentation result, take the point with the largest pixel intensity value in the connected region as the target location and its pixel intensity value as the target intensity, and take the standard deviation of pixel intensity value of segmented region as the target diffusion range. 2) Recover the target shape in the sonar image based on the target location; 3) Based on the target shape, target intensity, target location, and target diffusion range, multi-frame target detection is used to achieve real-time detection of small targets: First, the target in the sonar image of the current frame is compared with the targets in the trusted target list and the target list to be observed. When a feature match is found between the two, the targets in the trusted target list and the target list to be observed are updated. For the current target that does not have a feature match in the trusted target list and the target list to be observed, it is regarded as a new target. Then, the target list to be observed is checked, and the targets that are updated stably are moved to the trusted target list. Targets that cannot be updated for multiple consecutive frames are regarded as false alarms and discarded. Finally, the trusted target list is checked. For targets that have not been updated, their boundaries are searched according to their target locations. If no boundary is found, the current target location and boundary are retained, and the targets in the trusted list are used as the target detection results for this frame. Step 1) involves enhancing the contrast of the sonar image and increasing the pixel intensity values ​​in local extreme point regions to obtain an enhanced image. Specifically, this includes the following steps: 11) Calculate the global contrast of the pixels in the sonar image, and denot it as the global saliency map; 12) Calculate the local contrast of the pixels in the sonar image and record it as a local saliency map; 13) Perform convolution operation on the sonar image data using object detection convolution kernel group, and take the maximum value of the component at each pixel position in the obtained convolution image group to obtain the object saliency map. 14) The global saliency map, local saliency map, and object saliency map are weighted and superimposed to form an enhanced image.

2. The real-time small target detection method based on sonar images according to claim 1, characterized in that, In step 11), the global pixel intensity of the sonar image is represented by a Gaussian probability model, which is: In the formula, x represents the pixel value, σ represents the standard deviation of the pixel value, and μ represents the mean of the pixel value.

3. The real-time small target detection method based on sonar images according to claim 2, characterized in that, In step 11), the global contrast of the pixels in the sonar image is: In the formula, xy represents the position of the pixel in the sonar image, and P xy This represents the pixel value at that location. This represents the global contrast at that location, and model(·) represents the Gaussian probability model obtained in step 11).

4. The real-time small target detection method based on sonar images according to claim 3, characterized in that, In step 12), the local contrast of the pixels in the sonar image is calculated by processing the sonar image with Gaussian filtering and mean filtering, and taking the absolute value of the difference between the two filtered images as the local contrast of the pixels in the sonar image.

5. The real-time small target detection method based on sonar images according to claim 4, characterized in that, In step 1), the dynamic threshold is: In the formula, threshold(α) is the dynamic threshold, and α is the threshold coefficient in the range [0,1], representing the linear position of the dynamic threshold from the mean to the maximum value. To enhance the average pixel intensity value of an image, P max To enhance the maximum pixel intensity value of the image.

6. The real-time small target detection method based on sonar images according to claim 5, characterized in that, In step 3), the features include target intensity and target location. If the difference in target intensity between the target in the sonar image of the current frame and the targets in the list of trusted targets and the list of targets to be observed is within the target diffusion range and the distance between the target locations does not exceed the distance threshold, then the features are considered to match.

7. The real-time small target detection method based on sonar images according to claim 6, characterized in that, In step 3), if a new target appears at the far end of the sonar image of the current frame, it is considered a new target that has entered the sonar detection range and is added to the list of targets to be observed; otherwise, it is considered a false alarm and is discarded.

Images