An adaptive target recognition method for a single-beam mechanically scanning sonar system

Abstract

The application discloses a kind of single-beam mechanical scanning sonar system's adaptive target identification method, its advantage is to use the identification strategy of coarse scanning to fine scanning, wide area wide beam scanning is carried out in initial stage, subsequently focus on potential target area, so as to improve the accuracy of near distance target while maintaining long distance detection capability;By dynamically adjusting scanning parameters, such as automatically switching detection range and adjusting scanning angle, the balance between detection accuracy and imaging distance can be significantly improved, while the scanning time is significantly reduced, the response speed of the system is improved, making it more suitable for real-time detection tasks in dynamic environments;The introduction of learning model adaptive target identification, the system can optimize data processing, can more efficiently process scanning data and reduce redundant information, so as to reduce the burden of data processing, but also can significantly improve the imaging clarity, especially in the details of key areas.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of acoustic imaging detection, and particularly to an adaptive target recognition method for a single-beam mechanically scanned sonar system. BACKGROUND

[0002] In current sonar systems, there is often a contradiction between detection accuracy and imaging distance that is difficult to reconcile. Generally, improving detection accuracy will shorten the imaging distance, and pursuing a farther imaging distance will sacrifice accuracy. This limitation greatly limits the applicability of sonar systems, especially in application scenarios that require both long-range detection and high-precision recognition.

[0003] Another common problem with existing sonar systems is the relatively insufficient data processing capability, which directly leads to low imaging clarity. In applications that require high-resolution images (such as underwater target recognition or environmental surveying), this problem is particularly prominent, limiting the actual effectiveness of sonar systems. Since mechanically scanned sonar relies on a point-by-point scanning working method, its speed is slow, making it difficult to meet real-time requirements. This becomes a major bottleneck in dynamic environments or scenarios that require rapid response. Existing single-beam devices lack dynamic adjustment capabilities, and there is currently no single-beam device on the market that has dynamic adjustment capabilities, which needs to be developed urgently. SUMMARY

[0004] One technical problem to be solved by the present application is to overcome the defects of the above related technologies and provide an adaptive target recognition method for a single-beam mechanically scanned sonar system. Through an adaptive control module, based on real-time analysis of echo data, scanning parameters are dynamically adjusted to achieve "from coarse to fine" target recognition.

[0005] The technical solution adopted by the present application to solve the technical problem is as follows: an adaptive target recognition method for a single-beam mechanically scanned sonar system, specifically comprising the following steps:

[0006] S1 Coarse scanning: the single-beam mechanically scanned sonar performs 360° wide-beam scanning at a first step angle to quickly obtain echo data; and performs preliminary feature extraction on the echo data to identify potential target regions;

[0007] S2 Target detection and labeling: analyze the extracted preliminary features, judge and label potential target regions; through Doppler effect analysis, detect the frequency shift of the echo data, and if the target is dynamic, proceed to step S3A, and if the target is static, proceed to step S3B;

[0008] S3A Range adaptive adjustment: use a decision tree algorithm to determine whether the potential target region is located at the range boundary, adjust to an appropriate range according to the distance characteristics of the potential target region and a preset boundary threshold, and proceed to step S4A;

[0009] S4A fine scanning: locally narrow-beam scanning the marked potential target area with an appropriate range at a second step angle to obtain high-resolution target detail data, entering step S5; wherein the second step angle is half of the first step angle;

[0010] S3B medium-range scanning and range self-adaptive adjustment: medium-range confirmation scanning the potential target area to demarcate the angle range and distance characteristics of the target area, using a decision tree algorithm to determine whether the demarcated target area is located at the range boundary, adjusting to an appropriate range, entering step S4B;

[0011] S4B fine scanning: narrow-beam scanning the demarcated target area with an appropriate range at a third step angle to obtain high-resolution target detail data, entering step S5; wherein the third step angle is one percent of the angle range of the demarcated target area;

[0012] S5 target recognition: using a learning model to perform target recognition on the target detail data obtained by step S4A or step S4B fine scanning, outputting the category and position information of the target.

[0013] As preferred, it also includes S6 parameter optimization and adaptive learning: using a learning algorithm to automatically adjust the scanning parameters and the hyperparameters of the learning model through historical recognition results and system performance feedback.

[0014] Compared with related art, the present application has the advantages that the identification strategy from coarse scanning to fine scanning is adopted, the single-beam mechanical scanning sonar system can perform wide-area scanning in the initial stage and then focus on the potential target area, thereby improving the accuracy of the near-distance target while maintaining the long-distance detection capability, and according to the state of the target, directly from coarse scanning to fine scanning for high-speed target tracking, or adding a medium-range confirmation scan between coarse scanning and fine scanning, mainly for underwater archaeology and other tasks that require wide coverage and detailed identification; and by dynamically adjusting the scanning parameters, such as automatically switching the detection range and adjusting the scanning angle, the balance between detection accuracy and imaging distance can be significantly improved, while the scanning time is significantly shortened, the response speed of the single-beam mechanical scanning sonar system is improved, and it is more suitable for real-time detection tasks in dynamic environments; and by introducing a learning model for adaptive target recognition, the single-beam mechanical scanning sonar system can optimize the data processing flow, more efficiently process scanning data and reduce redundant information, thereby reducing the burden of data processing while significantly improving the imaging clarity, especially in the detailed presentation of key areas.

[0015] As preferred, the learning algorithm in step S6 is reinforcement learning algorithm, and the scanning parameters include beam width, step angle, fine scanning range. Reinforcement learning (RL) algorithm enables the single-beam mechanically scanned sonar system to continuously optimize the identification strategy and adapt to different environments and targets.

[0016] As preferred, the preliminary feature extraction in step S1 uses convolutional neural network on the echo data, and the preliminary features include intensity and texture.

[0017] As preferred, the analysis of the preliminary features extracted by the convolutional neural network in step S2 uses support vector machine classifier. The combination of convolutional neural network (CNN) and support vector machine (SVM) significantly improves the accuracy and efficiency of data processing, and reduces misjudgment and omission.

[0018] As preferred, the learning model in step S5 is recurrent neural network, and the target detail data is time-series echo data. Recurrent neural network can handle time-series dependency and accurately identify the dynamic characteristics of the target, thereby significantly improving the accuracy and efficiency of target identification by analyzing time-series dependency. The single-beam mechanically scanned sonar system fuses sonar and optical sensor data to improve the robustness of the system through collaborative analysis of multi-element information. The data fusion method uses feature-level fusion to extract the features of sonar and optical data and input them into the joint model (which can be a deep learning model or a convolutional neural network) for analysis. Multi-source fusion significantly reduces the false alarm rate and the missed alarm rate, and finally integrates the recognition result and the clear target image.

[0019] As an improvement, in the range self-adaptive adjustment of step S3A or step S3B, if the target is located at the boundary, the range is expanded; if the target is not at the boundary and the details are sufficient, the range is shortened. Expanding the range to cover a larger area and shortening the range to improve accuracy can improve the balance between detection accuracy and imaging distance.

[0020] As preferred, the scanning angle of the local narrow beam in step S4A is 180° or 120°, which contains the marked potential target area.

[0021] As preferred, the first step angle is 3.6°, and the second step angle is 1.8°. The 360° can be divided into an integer number of parts, which is convenient for subsequent model calculation and algorithm calculation. BRIEF DESCRIPTION OF DRAWINGS

[0022] Figure 1 is a flowchart of the adaptive target identification method of the present application. DETAILED DESCRIPTION

[0023] First, those skilled in the art should understand that these embodiments are only used to explain the technical principles of the embodiments of the present application, and are not intended to limit the protection scope of the embodiments of the present application. Those skilled in the art can adjust them as needed to adapt to specific application occasions.

[0024] The present application will be further described in detail below in combination with the drawings and specific embodiments.

[0025] The present embodiment is an adaptive target recognition method of a single-beam mechanical scanning sonar system, which specifically comprises the following steps:

[0026] S1. Coarse scanning:

[0027] The single-beam mechanical scanning sonar performs 360° wide-beam scanning with a 3.6° step angle to quickly obtain echo data. The single-beam mechanical scanning sonar is also called single-beam mechanical scanning sonar or single-beam mechanical scanning sonar.

[0028] A convolutional neural network (CNN) is used to preliminarily extract features of the echo data and identify potential target regions. The CNN model learns the preliminary features of the targets in the echo signal through training, automatically extracts high-level features such as intensity and texture, and replaces the traditional threshold judgment.

[0029] S2. Target detection and labeling:

[0030] A support vector machine (SVM) classifier is used to analyze the features extracted by the CNN, judge and label the potential target regions. The SVM classifier can effectively distinguish targets and noise through supervised learning, and improve the accuracy of target detection. Moreover, the position information of the labeled potential target regions is stored, which only occupies very little memory (less than half a byte). Through Doppler effect analysis, the frequency shift of the echo data is detected to determine whether the target is dynamic, i.e., when used for high-speed target tracking task, the moving target causes frequency change with obvious shift, and step S3A is entered.

[0031] S3A. Range adaptive adjustment:

[0032] A decision tree algorithm is used to determine whether the target is located at the range boundary. The decision tree automatically determines whether to expand the range or shorten the range according to the distance characteristics of the target and the preset boundary threshold.

[0033] If the target is located at the boundary, the single-beam mechanical scanning sonar system automatically expands the range to cover a farther area; if the target is not at the boundary and the details are sufficient, the single-beam mechanical scanning sonar system shortens the range to improve the precision. Step S4A is entered. The distance characteristics of the potential target region are within 1 / 3~2 / 3 of the appropriate range.

[0034] S4A. Fine scanning:

[0035] The marked target area is scanned with a 1.8° step angle for 180° or 120° local narrow beam, which is 4 times or 6 times faster than the traditional 360° full-range fine scanning, and high-resolution target detail data is obtained.

[0036] S5. Target recognition:

[0037] A recurrent neural network (RNN) is used to recognize the target from the time-series echo data of fine scanning. The RNN model can handle time-series dependencies and accurately identify the dynamic characteristics of the target. Through the model output, the category and location information of the target are obtained, realizing high-precision target recognition.

[0038] S6. Parameter optimization and adaptive learning:

[0039] A reinforcement learning (RL) algorithm is introduced to automatically adjust the scanning parameters (such as step angle, scanning range) and model hyperparameters based on historical recognition results and system performance feedback. The RL algorithm enables the single-beam mechanical scanning sonar system to continuously optimize the recognition strategy and adapt to different environments and targets.

[0040] Here, the wide beam refers to a narrow beam with a larger fan angle, and the narrow beam refers to a narrow beam with a smaller fan angle, i.e. the fan angle of the wide beam is 1.5-3 times larger than that of the narrow beam.

[0041] By dynamically adjusting the scanning parameters (such as automatically switching the detection range and adjusting the scanning angle), the balance between detection accuracy and imaging distance can be significantly improved. Using a "from coarse to fine" recognition strategy, the single-beam mechanical scanning sonar system can perform wide-area scanning in the initial stage and then focus on the region of interest (potential target area), thereby maintaining long-range detection capability while improving the accuracy of close-range targets. This method is a significant breakthrough over existing technologies. By introducing an adaptive target recognition method, the proposed system can optimize the data processing flow, enabling more efficient processing of scanning data and reducing redundant information. This method not only reduces the burden of data processing, but also significantly improves the clarity of the image, especially in the presentation of details in key areas. Although the real-time performance of mechanical scanning sonar is limited by its physical characteristics, the proposed adaptive algorithm effectively improves the response speed and operating efficiency of the single-beam mechanical scanning sonar system by automatically adjusting and optimizing the scanning process (e.g. prioritizing scanning of target areas). This improvement is more advantageous in terms of real-time performance than traditional systems. By combining mechanical scanning sonar with adaptive recognition methods, the proposed single-beam mechanical scanning sonar system can adjust scanning parameters in real time according to the target environment, which is a unique feature in current technology and opens up new possibilities for the application of single-beam sonar.

[0042] The "from coarse to fine" step-by-step scanning strategy effectively balances the relationship between precision and distance. The coarse scanning quickly covers a large range, and the fine scanning enhances the ability to identify details, making the single-beam mechanical scanning sonar system not only able to detect long-distance targets, but also to ensure high-resolution imaging quality; the combination of CNN and SVM significantly improves the accuracy and efficiency of data processing, reducing misjudgment and missed judgment phenomena. The final imaging clarity is greatly improved, and the accuracy of target recognition is also enhanced, providing users with more reliable detection results. Compared with existing technologies, through the synergistic effect of CNN and SVM, the imaging clarity is improved from 80% to 95%. Compared with traditional 360° full-range fine scanning, the scanning speed is increased by 4 times (180° area) or 6 times (120° area), which significantly shortens the scanning time and improves the response speed of the single-beam mechanical scanning sonar system, making it more suitable for real-time detection tasks in dynamic environments. The single-beam mechanical scanning sonar system can dynamically adjust the running scanning parameters according to real-time data, maintaining optimal performance in complex or changing environments. This adaptive ability enhances the stability and practicality of the single-beam mechanical scanning sonar system, especially suitable for variable scenarios such as underwater exploration. RNN significantly improves the accuracy and efficiency of target recognition by analyzing time-dependent relationships, especially in the identification of dynamic targets (such as moving objects), the single-beam mechanical scanning sonar system performs well, providing users with more accurate detection results.

[0043] In step S2, the frequency shift of the echo data is detected through Doppler effect analysis, and when the target is static, it enters step S3B, which is mainly used for tasks such as underwater archaeology that require large-scale coverage and detail recognition. Static targets have no significant shift, and enter step S3B.

[0044] S3B: Mid-range scanning and range adaptive adjustment: Adopt mid-range confirmation scanning on potential target area, calibrate the angle range and distance characteristics of the target area, and use decision tree algorithm to judge whether the calibrated target area is located at the range boundary, adjust to the appropriate range, and enter step S4B; In the mid-range confirmation scanning, the range used is 2 times the maximum value of the distance characteristics of the potential target area. The distance characteristics of the calibrated target area are within 1 / 3~2 / 3 of the appropriate range.

[0045] S4B: Fine scanning: Adopting appropriate range for narrow-beam scanning of the calibrated target area at the third step angle, obtaining high-resolution target detail data, and entering step S5; Wherein, the third step angle is one percent of the angle range of the calibrated target area.

[0046] Through the medium-distance confirmation scanning, the area detected by the wide-area scanning (i.e. coarse scanning) is focused on to verify whether it is a real target or to exclude false positives. Then through fine scanning, the confirmed target is scanned at high resolution to obtain detailed feature information.

[0047] The above description is merely a specific implementation of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical scope disclosed by the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. An adaptive target recognition method for a single-beam mechanically scanned sonar system, characterized in that, Specifically comprising the following steps: S1 rough scanning: a single-beam mechanically scanned sonar performs 360° wide-beam scanning at a first step angle to quickly obtain echo data; and performs preliminary feature extraction on the echo data to identify potential target regions; S2 target detection and marking: analyzing the extracted preliminary features, judging and marking potential target regions; Through Doppler effect analysis, the frequency shift of the echo data is detected to determine whether the target is dynamic, and if so, step S3A is entered, or whether the target is static, and if so, step S3B is entered; S3A range self-adaptive adjustment: using a decision tree algorithm to determine whether the potential target region is located at the range boundary, adjusting to an appropriate range according to the distance characteristics of the potential target region and the preset boundary threshold, and entering step S4A; S4A fine scanning: performing local narrow-beam scanning on the marked potential target region using an appropriate range at a second step angle to obtain high-resolution target detail data, and entering step S5; wherein the second step angle is half of the first step angle; S3B mid-range scanning and range self-adaptive adjustment: performing mid-range confirmation scanning on the potential target region to calibrate the angle range and distance characteristics of the target region, using a decision tree algorithm to determine whether the calibrated target region is located at the range boundary, adjusting to an appropriate range, and entering step S4B; S4B fine scanning: performing narrow-beam scanning on the calibrated target region using an appropriate range at a third step angle to obtain high-resolution target detail data, and entering step S5; wherein the third step angle is one percent of the angle range of the calibrated target region; S5 target identification: using a learning model to perform target identification on the target detail data obtained by step S4A or step S4B fine scanning, and outputting the category and position information of the target.

2. The adaptive target recognition method of a single-beam mechanically scanned sonar system according to claim 1, characterized in that: It also includes S6 parameter optimization and adaptive learning: using a learning algorithm to automatically adjust the scanning parameters and the hyperparameters of the learning model through historical identification results and system performance feedback.

3. The adaptive target recognition method of a single-beam mechanically scanned sonar system according to claim 2, characterized in that: The learning algorithm in step S6 is a reinforcement learning algorithm, and the scanning parameters include beam width, step angle, and fine scanning range.

4. The adaptive target recognition method of a single-beam mechanically scanned sonar system according to any one of claims 1-3, characterized in that: In step S1, a convolutional neural network is used to extract preliminary features from the echo data, including intensity and texture.

5. The adaptive target recognition method of a single-beam mechanically scanned sonar system according to claim 4, characterized in that: In step S2, a support vector machine classifier is used to analyze the preliminary features extracted by the convolutional neural network.

6. The adaptive target recognition method of a single-beam mechanically scanned sonar system according to claim 5, characterized in that: The learning model in step S5 is a recurrent neural network, and the target detail data is time-series echo data.

7. The adaptive target recognition method of a single-beam mechanically scanned sonar system according to any one of claims 1-3, characterized in that: In the range self-adaptive adjustment of step S3A or step S3B, if the target is located at the boundary, the range is expanded; if the target is not at the boundary and the details are sufficient, the range is shortened.

8. The adaptive target recognition method of a single-beam mechanically scanned sonar system according to any one of claims 1-3, characterized in that: The scanning angle of the local narrow-beam in step S4A is 180° or 120°, which contains the marked potential target region.

9. The adaptive target recognition method of a single-beam mechanically scanned sonar system according to claim 8, characterized in that: The first step angle is 3.6°, and the second step angle is 1.8°.

Images