An evaluation system and device for eurymerus quantity and size based on side-scan sonar image

Abstract

The present application relates to a kind of based on side scan sonar image's Eriocheir sinensis quantity and size evaluation system and equipment, belong to aquaculture monitoring, underwater acoustic detection and image intelligent identification technical field, the system includes side scan sonar collection module, data preprocessing module, bright spot map patch detection module, Eriocheir sinensis candidate target screening module, bright spot-shading identification module, multiple track deduplication module, resource quantity evaluation and visualization module.The method realizes Eriocheir sinensis individual level automatic identification, the assessment of Eriocheir sinensis quantity, size distribution and total resource quantity in pond.Make Eriocheir sinensis breeding pond can not interfere with production, quickly and accurately obtain Eriocheir sinensis quantity and specification structure information in pond, provide data support for feeding management, disease early warning, fishing decision etc..

Description

Technical Field

[0001] This invention belongs to the field of image intelligent recognition technology, specifically relating to a system and device for evaluating the number and size of Chinese mitten crabs based on side-scan sonar images. Background Technology

[0002] The Chinese mitten crab is one of my country's important freshwater aquaculture species. Crab farming ponds cover a vast area and are numerous, with increasing stocking density and sophisticated management practices. In actual production, farmers and management departments often need to obtain the following information: 1. The number of Chinese mitten crabs in the aquaculture ponds; 2. The individual size structure of the Chinese mitten crabs (proportion of small, medium, and large sizes); 3. The spatial distribution of Chinese mitten crabs in the ponds; 4. The dynamic trend of Chinese mitten crab resources at different times.

[0003] Currently, the main methods commonly used in actual production to assess the status of mitten crab farming include: 1. Manual sampling and empirical estimation: Common methods include setting up traps, dip nets, or conducting trial catches in localized areas of the pond, estimating the overall number and individual structure of mitten crabs in the pond based on the sample size and size. This method is labor-intensive, causes significant disturbance to the pond's ecological environment, has limited sampling points and insufficient representativeness, and results in large estimation errors, often only providing a rough estimate. 2. Estimation based on feeding amount and growth model: Some farms estimate the stock based on the amount of feed given to mitten crabs, their feeding rate, and empirical growth curves. This method is greatly affected by factors such as water quality, weather, disease, escape, and mortality, resulting in unstable estimations. It can usually only be used as a reference indicator and is difficult to provide precise quantities. 3. Traditional acoustic and fish-finding techniques: Single-beam or multi-beam echo sounding devices are mainly used for detecting fish in deeper waters and measuring water depth. The working water depth is usually large, and the beam direction is mostly vertical. However, mitten crabs are benthic organisms with low activity heights and individuals close to the bottom silt and aquatic plants. Traditional fish-finding devices have difficulty distinguishing them from bottom echoes in shallow crab pond environments, and the resolution is insufficient to achieve individual-level identification.

[0004] Existing methods cause significant disturbance to the ponds, making high-frequency monitoring impossible without disrupting aquaculture production. They are also difficult to repeat frequently throughout the aquaculture cycle, failing to meet the timeliness requirements of smart aquaculture and refined management. Therefore, there is an urgent need for a technical solution that can operate stably in shallow-water crab farming ponds, automatically identify individual crabs, and simultaneously assess their quantity and size. This solution aims to address the aforementioned technical challenges of accurately obtaining pond stock levels and achieving non-destructive high-frequency monitoring. Summary of the Invention

[0005] This invention addresses the aforementioned technical problems by providing a side-scan sonar method and system suitable for Chinese mitten crab farming ponds. This method can clearly visualize individual Chinese mitten crabs in shallow water, soft-bottom, and weed-infested environments. Based on a bright spot detection and bright spot-shadow combination recognition algorithm for Chinese mitten crabs in side-scan sonar images, the method achieves automatic identification at the individual crab level, enabling the assessment of the quantity, size distribution, and total resource quantity of Chinese mitten crabs in the pond. This allows Chinese mitten crab farming ponds to quickly and accurately obtain information on the quantity and size structure of Chinese mitten crabs without disrupting production, providing data support for feeding management, disease early warning, and harvesting decisions.

[0006] This invention is achieved through the following technical solution: A system for assessing the number and size of Chinese mitten crabs based on side-scan sonar images, the system comprising: Side-scan sonar acquisition module: Collects raw side-scan sonar data from the bottom of the crab farming pond along a preset flight path; Data preprocessing module: Performs geometric correction, noise suppression, and contrast enhancement on the raw sonar data obtained by the side-scan sonar acquisition module to generate grayscale sonar images suitable for recognition; Bright spot detection module: Extracts bright spots from the amplitude sonar images generated by the data preprocessing module; Crab candidate target screening module: Based on the area, shape and intensity features of bright spots obtained by the bright spot detection module, crab candidate targets are screened out; Highlight-Shadow Recognition Module: Constructs the combined features of bright spots and their shadows, and uses discrimination rules or intelligent models to identify crabs from the candidate targets output by the crab candidate target screening module; Multi-track deduplication module: Performs spatial matching and feature matching on the identification results from multiple survey tracks to remove duplicate detections of the same mitten crab target; Resource assessment and visualization module: Divide the pond into grids, count the number and size distribution of mitten crabs in each grid, construct a gridded resource assessment model, and output the quantity estimation results and spatial distribution heat map.

[0007] Furthermore, the data preprocessing module includes the following steps: (1) Slant distance and geometric correction: Based on the measured water depth and sound speed, the slant distance is converted and corrected to the horizontal plane projection distance; combined with the carrier attitude (roll, pitch) and attitude sensor data, the image is geometrically corrected. (2) Noise suppression and enhancement: Time-domain and spatial-domain filtering is performed on the original amplitude data to suppress random noise and stripe noise; Methods including logarithmic compression, histogram equalization, or local contrast enhancement are used to improve the contrast between bright spots and the background. (3) Image generation: Convert the preprocessed data into a two-dimensional grayscale image to visualize the bottom structure and target.

[0008] Furthermore, the bright spot detection module operates as follows: (1) Local adaptive thresholding: The mean and standard deviation of grayscale values ​​in the local region are calculated using a sliding window method; the improved Sauvola thresholding method is adopted, and the local threshold calculation formula is as follows: ,in This is the local threshold at pixel (x, y). Centered on this point, with a size of The average gray level within the window. The grayscale standard deviation within the window is given by k, which is an adjustment coefficient ranging from 0.2 to 0.5, preferably k=0.34. R is the dynamic range of the grayscale standard deviation, i.e. the maximum value that the standard deviation can take. For an 8-bit grayscale image, R=128. The window size w is set according to the sonar resolution, preferably 15 to 31 pixels. Based on local statistical characteristics, local thresholds are automatically calculated to adaptively segment the image and obtain potential bright spot regions. (2) Connectivity Analysis: Perform connectivity analysis on each bright spot region to determine the area, perimeter, and boundary shape of each continuous bright spot region; mark each connected region as a bright spot patch; the formula for calculating the shape factor (circularity) is as follows: A represents the area of ​​the bright spot, P represents the perimeter of the bright spot, and the typical value of the shape factor for the Chinese mitten crab ranges from 0.65 to 0.95. The lower limit takes into account the edge jaggedness effect caused by the resolution of the sonar image.

[0009] Furthermore, the aforementioned crab candidate target screening module: Based on actual measurements, bright spots of mitten crabs in side-scan sonar images have a certain range of area, shape, and intensity. This module filters bright spots based on the following conditions to obtain a set of candidate targets for mitten crabs: (1) The area of ​​the highlight patch falls within the preset area threshold range; (2) The aspect ratio of the highlight spots falls within the set range to exclude targets such as slender aquatic plants; (3) The average echo intensity of the target that produces bright spots is higher than the background average by a certain threshold; (4) The edge gradient of the bright spots is large, reflecting the characteristics of hard boundaries.

[0010] Furthermore, the aforementioned bright spot-shadow recognition module (1) Shadow area search: Based on the direction of sound wave propagation, select a search area of ​​a certain length on the side of the bright spot away from the sound source (i.e., the far end along the direction of sound wave propagation); find shadow segments in this area that are significantly lower than the background brightness and extract the shadow area; (2) Feature vector construction: Calculate the features of the bright spot: area, major axis length, minor axis length, average intensity and shape factor; at the same time, calculate the shadow length, width, average brightness and brightness variation of the shadow area; calculate the relative positional relationship between the bright spot and the shadow area, that is, the distance and direction between the shadow area and the center of the bright spot; (3) Classification and recognition: Input the bright spot-shadow combination features into a pre-trained machine learning model or deep learning model; output the recognition result of whether each bright spot is a hairy crab.

[0011] Furthermore, regarding the multi-track deduplication module, since there is some overlap in the flight paths during pond surveying, the same crab may be scanned multiple times. The specific method is as follows: (1) Convert the mitten crab targets identified in each flight path to the pond coordinate system; (2) Cluster the target crabs in the neighborhood using a certain spatial distance threshold; (3) Calculate the bright spot-shadow combination feature similarity for targets within each cluster; use weighted Euclidean distance to calculate feature similarity, the distance formula is as follows: Similarity The feature weight vector includes area weights. =0.25, Major axis length weight =0.20, average intensity weight =0.20, shadow length weight =0.20, shape factor weight =0.15, , These are the normalized feature values ​​of the two crab targets to be compared within the cluster on the i-th feature dimension, i=1,2,...,5, which correspond to the five feature dimensions of bright spot area, major axis length, average intensity, shadow length and shape factor, respectively. Each feature is normalized to the [0,1] interval by minimum-maximum normalization. (4) If the similarity exceeds the set threshold, preferably S>0.85, then it is determined to be the same target crab, and only one of them is retained.

[0012] Furthermore, the resource quantity assessment and visualization module is operated as follows: (1) Grid division: Divide the bottom area of ​​the pond into regular grids, or adjust the grid size adaptively according to the density of mitten crabs; (2) Quantity and size statistics: Count the number of deduplicated mitten crabs in each grid; calculate the average size and distribution range of individual mitten crabs in each grid.

[0013] (3) Resource quantity estimation: Estimate the shell width based on bright spot patches, using the following formula: ,in For the estimated carapace width (mm), A spot The area (in pixels) of the bright spot needs to be converted into the actual physical area based on the sonar's lateral resolution and the sampling interval along the flight direction before being used for calculation. , The correction coefficients were obtained through actual measurement and calibration; weight estimation was performed using an allometric growth model. Where M is weight (g) and W is carapace width (mm). , Typical values ​​for the species-specific parameter, Chinese mitten crab. Approximately 0.0003 Approximately 3.1, this parameter varies depending on sex and growth stage, and the specific parameter is determined through field measurement and calibration; combined with the model of the relationship between the size and weight of the Chinese mitten crab, the total weight of the Chinese mitten crabs in each grid and the entire pond is estimated; correction coefficients are introduced for different bottom sediment types to improve the estimation accuracy.

[0014] (4) Visualization: Generate heat maps of the density and size distribution of mitten crabs to visualize the spatial distribution of resources.

[0015] Furthermore, the side-scan sonar frequency can be selected within the range of 400–1200 kHz according to the actual pond area and water depth; recommended sonar parameter settings include: operating frequency of 900–1200 kHz (high frequency) and 400–600 kHz (low frequency), pulse length of 20–50 microseconds, maximum range of 15–25 m (single side), sampling rate of not less than 50 kHz, speed of 1.0–2.0 m / s, and installation depth of 0.3–0.5 m below the water surface.

[0016] Furthermore, dual-frequency side-scan sonar is employed, combined with high-frequency and mid-frequency images to enhance recognition robustness.

[0017] Furthermore, the highlight is the use of rule-based discrimination methods, including thresholding and morphological discrimination, or deep learning models with different structures, such as U-Net and YOLO, for shadow recognition.

[0018] Furthermore, when shadow features are insufficient, identification relies solely on bright spot morphology and intensity features.

[0019] The present invention provides a device equipped with the system.

[0020] Furthermore, the system is carried out on manually piloted small boats, remote-controlled boats, or fully autonomous unmanned boats.

[0021] The principle of this invention is based on the inventors' findings from experiments conducted by deploying side-scan sonar in actual crab farming ponds: (1) Because the carapace of the Chinese mitten crab is a relatively hard structure, it has strong acoustic reflection; (2) On the side-scan sonar images, individual Chinese mitten crabs appear as strong highlights, which contrast sharply with the surrounding silt background and aquatic plant texture; (3) A short shadow area will also be formed behind the bright spots, and the length and shape of the shadow correspond to the posture and body shape of the hairy crab.

[0022] The above findings indicate that the Chinese mitten crab has relatively stable acoustic texture features in side-scan sonar images, providing a technical basis for automatic identification and resource assessment through image processing and intelligent recognition algorithms.

[0023] The beneficial effects of this invention compared to the prior art are as follows: (1) This invention is based on the discovery and utilization of the acoustic feature that "Chinese mitten crabs present stable bright spots accompanied by short shadows in side-scan sonar images," and proposes a Chinese mitten crab identification algorithm based on the combination features of bright spots and shadows. For the first time, the side-scan sonar system is specifically applied to Chinese mitten crab farming ponds, and the scanning tracks and acoustic parameters are optimized for shallow water, soft bottom, and aquatic plant environments to achieve individual Chinese mitten crab imaging. At the same time, a deduplication method for Chinese mitten crab targets under multiple tracks is proposed, which uses spatial neighborhood and bright spot-shadow feature similarity for matching to avoid duplicate counting and improve the accuracy of quantity estimation. The gridded resource quantity assessment model for Chinese mitten crab farming ponds constructed by the method of this invention realizes the automated estimation and spatial visualization of the number, size distribution, and total resource quantity of Chinese mitten crabs. This invention organically combines side-scan sonar scanning, intelligent image recognition, multi-track deduplication, and resource quantity model, providing a complete digital monitoring technology chain for Chinese mitten crab farming.

[0024] (2) This invention utilizes acoustic scanning, eliminating the need for numerous traps or nets, thus avoiding damage to the pond bottom and disturbance to the crabs. It can be repeated multiple times during the aquaculture process. Through image processing and intelligent recognition algorithms, individual crabs are automatically identified from side-scan sonar images, saving a significant amount of manual interpretation and experience-based judgment.

[0025] (3) Simultaneous Assessment of Quantity and Size: This invention identifies the number of mitten crabs while estimating individual size based on bright spot dimensions and shadow lengths, enabling specification and structural analysis. The total resource volume can be further calculated using a body length-weight model. Furthermore, utilizing high-frequency side-scan sonar and bright spot-shadow combination features, it effectively distinguishes mitten crabs from interfering targets such as rocks and aquatic plants. In actual measurements, the identification accuracy reaches a high level, and the quantity estimation error is significantly smaller than that of traditional sampling methods. It also boasts advantages such as short scanning time, automated processing, and the ability to be implemented multiple times within the breeding cycle, providing dynamic and objective data support for adjusting feeding strategies, disease early warning, and selecting harvesting times. This invention's system can be integrated with unmanned vessel platforms and aquaculture management platforms to form a smart aquaculture integrated monitoring system, possessing good scalability and engineering application prospects. Detailed Implementation

[0026] The technical solution of the present invention will be further explained below through embodiments, but the scope of protection of the present invention is not limited in any way by the embodiments.

[0027] Example 1 This embodiment provides a system for assessing the number and size of mitten crabs based on side-scan sonar images. The system includes: Side-scan sonar acquisition module: Collects raw side-scan sonar data from the bottom of the crab farming pond along a preset flight path; Data preprocessing module: Performs geometric correction, noise suppression, and contrast enhancement on the raw sonar data obtained by the side-scan sonar acquisition module to generate grayscale sonar images suitable for recognition; Bright spot detection module: Extracts candidate bright spots from the amplitude sonar images generated by the data preprocessing module; Crab candidate target screening module: Based on the area, shape and intensity features of bright spots obtained by the bright spot detection module, crab candidate targets are screened out; Highlight-Shadow Recognition Module: Based on the candidate targets output by the crab candidate target screening module, construct the combined features of its highlight patches and the shadow behind them, and use discrimination rules or intelligent models to identify crabs; Multi-track deduplication module: Performs spatial matching and feature matching on the identification results from multiple survey tracks to remove duplicate detections of the same mitten crab target; Resource assessment and visualization module: Divide the pond into grids, count the number and size distribution of mitten crabs in each grid, construct a gridded resource assessment model, and output the quantity estimation results and spatial distribution heat map.

[0028] In a preferred embodiment, the data preprocessing module includes the following steps: (1) Slant distance and geometric correction: Based on the measured water depth and sound speed, the slant distance is converted and corrected to the horizontal plane projection distance; combined with the carrier attitude and attitude sensor data, such as the attitude of roll or pitch, the image is geometrically corrected. (2) Noise suppression and enhancement: The original amplitude data is filtered in the time domain and spatial domain to suppress random noise and stripe noise; methods including logarithmic compression, histogram equalization or local contrast enhancement are used to improve the contrast between the bright spots of the hairy crab and the background. (3) Image generation: Convert the preprocessed data into a two-dimensional grayscale image to visualize the bottom structure and target.

[0029] As a preferred embodiment, the bright spot detection module operates as follows: (1) Local adaptive threshold segmentation: The gray mean and standard deviation of the local region are calculated by sliding window; the local threshold is automatically calculated based on the local statistical characteristics to perform adaptive segmentation of the image and obtain potential bright spot regions; (2) Connectivity analysis: Perform connectivity analysis on each potential bright spot region to determine the area, perimeter, and boundary shape of each continuous potential bright spot region; mark each connected region as a bright spot patch.

[0030] As a preferred embodiment, the mitten crab candidate target screening module: Based on actual measurements, bright spots of mitten crabs in side-scan sonar images have a certain range of area, shape, and intensity. This module filters bright spots based on the following conditions to obtain a set of candidate targets for mitten crabs: (1) The area of ​​the bright spot falls within the preset area threshold range. The area threshold of the Chinese mitten crab is 50 to 500 pixels, which corresponds to the individual projection area range of about 30 to 80 mm of shell width under sonar resolution conditions. (2) The aspect ratio of the highlight spots falls within the set range (e.g., 0.5 to 2) to exclude targets such as slender aquatic plants; (3) The average echo intensity of the target that produces bright spots is higher than the background average by a certain threshold, for example, 8 to 15 dB; (4) The edge gradient of the bright spots is large, reflecting the characteristics of hard boundaries.

[0031] As a preferred embodiment, the operation method of the bright spot-shadow recognition module is as follows: (1) Shadow area search: Based on the direction of sound wave propagation, select a search area of ​​a certain length on the side away from the sound source (i.e., the far end along the direction of sound wave propagation) of the bright spot candidate target, i.e. the bright spot patch; find the shadow segment that is significantly lower than the background brightness in this area and extract the shadow area; (2) Feature vector construction: Calculate the features of the bright spot: area, major axis length, minor axis length, average intensity and shape factor; at the same time, calculate the shadow length, width, average brightness and brightness variation of the shadow area; calculate the relative positional relationship between the bright spot and the shadow area, that is, the distance and direction between the shadow area and the center of the bright spot; (3) Classification and recognition: Input the bright spot-shadow combination features into a pre-trained machine learning model or deep learning model, such as support vector machine, random forest, convolutional neural network, etc.; if a convolutional neural network is used, the preferred network structure is: the input layer receives a 64×64 pixel bright spot-shadow region image, the convolutional layer adopts a 3-layer structure with 32, 64 and 128 filters respectively, the pooling layer adopts 2×2 max pooling, the fully connected layer is 256 to 64 to 2 (binary classification), the activation function adopts ReLU, the output layer adopts Softmax, and the loss function adopts cross-entropy loss; output the recognition result of whether each candidate target (bright spot patch) is a hairy crab.

[0032] As a preferred embodiment, the multi-track deduplication module addresses the issue that the same crab may be scanned multiple times during pond surveying due to some overlap in the flight paths. The specific method is as follows: (1) Convert the mitten crab targets identified in each flight path to the pond coordinate system; (2) Cluster the target mitten crabs in a neighborhood using a certain spatial distance threshold (e.g., 0.2–0.5 m); (3) Calculate the bright spot-shadow combination feature similarity for targets within each cluster; (4) If the similarity exceeds the set threshold, they are determined to be the same mitten crab target, and only one of them is retained.

[0033] As a preferred embodiment, the resource quantity assessment and visualization module operates as follows: (1) Grid division: Divide the bottom area of ​​the pond into regular grids, such as 1 m × 1 m, or adjust the grid size adaptively according to the density of mitten crabs; (2) Quantity and size statistics: Count the number of deduplicated mitten crabs in each grid; calculate the average size and distribution range of individual mitten crabs in each grid.

[0034] (3) Resource quantity estimation: Combine the model of the relationship between the size and weight of the mitten crabs to estimate the total weight of the mitten crabs in each grid and the whole pond; introduce correction coefficients for different bottom sediment types to improve the estimation accuracy.

[0035] (4) Visualization: Generate heat maps of the density and size distribution of mitten crabs to visualize the spatial distribution of resources.

[0036] As a preferred embodiment, the side-scan sonar frequency can be selected in the range of 400 to 1200 kHz according to the actual pond area and water depth; as a more preferred embodiment, dual-frequency side-scan sonar is used, which combines high-frequency and mid-frequency images to improve recognition robustness.

[0037] Example 2 This embodiment provides a device equipped with the system described above.

[0038] The system can be used for manually piloted small boats, remote-controlled boats, or fully autonomous unmanned boats. Bright spot-shadow recognition employs rule-based discrimination methods, including thresholding and morphological discrimination, or deep learning models with different structures, including U-Net and YOLO. When shadow features are insufficient, recognition is performed solely based on bright spot morphology and intensity features.

[0039] Example 3 To verify the effectiveness of the method of the present invention, field tests were conducted in crab farming ponds in the Yellow River Estuary area of ​​Dongying.

[0040] 1. Test Environment Pond area: approximately 20 mu (approximately 13,333 m²); average water depth: 1.2~1.8 m; bottom type: silty bottom with localized aquatic plant cover; testing time: September 2024 (when the mitten crabs are at maturity), with the scanning period selected from 6:00 to 8:00 in the early morning, when the mitten crabs are less active and mostly rest at the bottom of the pond, which is beneficial to the stability of sonar imaging and the target detection rate; sonar frequency: high frequency 900 kHz, low frequency 450 kHz.

[0041] 2. Testing Methods An unmanned surface vessel (USV) equipped with a side-scan sonar system was used to conduct a full-coverage survey along a parallel route, with the route spacing set to 80% of the effective coverage width of the sonar to ensure adequate overlap. After the survey was completed, the identification algorithm of this invention was used to detect and count the mitten crabs. Simultaneously, three 10m × 10m sample plots were randomly selected within the pond, and a concentrated trapping method using ground cages was employed to obtain the actual number of crabs as a verification benchmark.

[0042] 3. Verification Results (1) Recognition performance indicators: precision was 89.3%, recall was 86.7%, F1 score was 87.9%, false positive rate was 10.7%, and false negative rate was 13.3%.

[0043] (2) Comparison of quantity estimation accuracy: The actual number of sample 1 was 47, while the present invention estimated 43, with a relative error of -8.5%; the actual number of sample 2 was 52, while the present invention estimated 48, with a relative error of -7.7%; the actual number of sample 3 was 38, while the present invention estimated 41, with a relative error of +7.9%; the average absolute relative error was 8.0%.

[0044] (3) Individual size estimation accuracy (shell width): The average absolute error was 4.2 mm, the relative error was 6.8%, and the correlation coefficient R² was 0.91.

[0045] 4. Comparison with traditional methods Regarding single-pond testing time, traditional trap sampling takes 4-6 hours, while the method of this invention only takes 30-45 minutes; regarding quantity estimation error, traditional methods have an error rate of 25-40%, while the method of this invention is less than 10%; regarding interference with aquaculture, traditional methods cause significant interference, while the method of this invention causes no interference; regarding repeatability, traditional methods are poor, while the method of this invention is good; regarding labor costs, traditional methods require 3-5 people, while the method of this invention only requires 1-2 people; regarding spatial distribution information acquisition, traditional methods cannot obtain it, while the method of this invention can output a heat map.

[0046] 5. Conclusion The crab quantity and size assessment system based on side-scan sonar imagery provided by this invention has demonstrated good identification accuracy and practicality in actual aquaculture pond tests. Compared with traditional sampling methods, the detection efficiency is improved by 5 to 10 times, and the quantity estimation error is reduced by more than 60%, while achieving non-destructive, rapid, and repeatable monitoring. It should be noted that in areas with dense aquatic plant coverage, sound wave attenuation and obstruction may prevent some crab targets from forming effective bright spots, resulting in a reduced detection rate. This system compensates for this by introducing correction coefficients for different bottom substrates and aquatic plant coverage types in resource estimation. Further integration with multi-band sonar data can improve the detection capability in aquatic plant areas.

Claims

1. A system for assessing the number and size of Chinese mitten crabs based on side-scan sonar images, characterized in that, The system includes: Side-scan sonar acquisition module: Collects raw side-scan sonar data from the bottom of the crab farming pond along a preset flight path; Data preprocessing module: Performs geometric correction, noise suppression, and contrast enhancement on the raw sonar data obtained by the side-scan sonar acquisition module to generate grayscale sonar images suitable for recognition; Bright spot detection module: Extracts bright spots from the amplitude sonar images generated by the data preprocessing module; Crab candidate target screening module: Based on the area, shape and intensity features of bright spots obtained by the bright spot detection module, crab candidate targets are screened out; Highlight-Shadow Recognition Module: Constructs combined features of bright spots and their shadows, and uses discrimination rules or intelligent models to identify candidate targets of hairy crabs; Multi-track deduplication module: Performs spatial matching and feature matching on the identification results from multiple survey tracks to remove duplicate detections of the same mitten crab target; Resource assessment and visualization module: Divide the pond into grids, count the number and size distribution of mitten crabs in each grid, construct a gridded resource assessment model, and output the quantity estimation results and spatial distribution heat map.

2. The system for assessing the quantity and size of mitten crabs based on side-scan sonar images according to claim 1, characterized in that, The data preprocessing module includes the following steps: (1) Slant distance and geometric correction: Based on the measured water depth and sound speed, the slant distance is converted and corrected to the horizontal plane projection distance; combined with the carrier attitude and attitude sensor data, the image is geometrically corrected. (2) Noise suppression and enhancement: Time-domain and spatial-domain filtering is performed on the original amplitude data to suppress random noise and stripe noise; Methods including logarithmic compression, histogram equalization, or local contrast enhancement are used to improve the contrast between bright spots and the background. (3) Image generation: Convert the preprocessed data into a two-dimensional grayscale image to visualize the bottom structure and target.

3. The system for assessing the quantity and size of mitten crabs based on side-scan sonar images according to claim 1, characterized in that, The bright spot detection module described above operates as follows: (1) Local adaptive thresholding: The mean and standard deviation of grayscale values ​​in the local region are calculated using a sliding window method; the improved Sauvola thresholding method is adopted, and the local threshold calculation formula is as follows: ,in This is the local threshold at pixel (x, y). Centered on this point, with a size of The average gray level within the window. R represents the standard deviation of grayscale within this window, k is the adjustment coefficient with a value range of 0.2 to 0.5, and R is the dynamic range of the standard deviation of grayscale, i.e., the maximum value that the standard deviation may take. Based on local statistical characteristics, local thresholds are automatically calculated to adaptively segment the image and obtain potential bright spot regions. (2) Connectivity analysis: Perform connectivity analysis on each bright spot region to determine the area, perimeter, and boundary shape of each continuous bright spot region; mark each connected region as a bright spot patch; The formula for calculating the shape factor is: A represents the area of ​​the bright spot, and P represents the perimeter of the bright spot.

4. The system for assessing the quantity and size of mitten crabs based on side-scan sonar images according to claim 1, characterized in that, The aforementioned crab candidate target screening module filters bright spots based on the following conditions to obtain a set of crab candidate targets: (1) The area of ​​the highlight patch falls within the preset area threshold range; (2) The aspect ratio of the highlight spots falls within the set range to exclude targets such as slender aquatic plants; (3) The average echo intensity of the target that produces bright spots is higher than the background average by a certain threshold; (4) The edge gradient of the bright spots is large, reflecting the characteristics of hard boundaries.

5. The system for assessing the quantity and size of mitten crabs based on side-scan sonar images according to claim 1, characterized in that, The operation method of the aforementioned highlight-shadow recognition module is as follows: (1) Shadow area search: Based on the direction of sound wave propagation, select a search area of ​​a certain length downstream of the bright spot; find shadow sections that are significantly lower than the background brightness within this area and extract the shadow area; (2) Feature vector construction: Calculate the features of the bright spot: area, major axis length, minor axis length, average intensity and shape factor; at the same time, calculate the shadow length, width, average brightness and brightness variation of the shadow area; calculate the relative positional relationship between the bright spot and the shadow area, that is, the distance and direction between the shadow area and the center of the bright spot. (3) Classification and recognition: Input the bright spot-shadow combination features into a pre-trained machine learning model or deep learning model; output the recognition result of whether each bright spot is a hairy crab.

6. The system for assessing the quantity and size of mitten crabs based on side-scan sonar images according to claim 1, characterized in that, The aforementioned multi-track deduplication module addresses the issue that overlapping flight paths during pond surveying can lead to the same crab being scanned multiple times. The specific method is as follows: (1) Convert the mitten crab targets identified in each flight path to the pond coordinate system; (2) Cluster the target crabs in the neighborhood using a certain spatial distance threshold; (3) Calculate the bright spot-shadow combination feature similarity for targets within each cluster; use weighted Euclidean distance to calculate feature similarity, the distance formula is as follows: Similarity , where the feature weight vector Including area weight =0.25, Major axis length weight =0.20, average intensity weight =0.20, shadow length weight =0.20, shape factor weight =0.15, , These are the normalized feature values ​​of the two crab targets to be compared within the cluster on the i-th feature dimension, i=1,2,...,5, which correspond to the five feature dimensions of bright spot area, major axis length, average intensity, shadow length and shape factor, respectively. Each feature is normalized to the [0,1] interval by minimum-maximum normalization. (4) If the similarity exceeds the set threshold, i.e. S>0.85, it is determined to be the same mitten crab target, and only one of them is retained.

7. The system for assessing the quantity and size of mitten crabs based on side-scan sonar images according to claim 1, characterized in that, The resource quantity assessment and visualization module is operated as follows: (1) Grid division: Divide the bottom area of ​​the pond into regular grids, or adjust the grid size adaptively according to the density of mitten crabs; (2) Quantity and size statistics: Count the number of deduplicated mitten crabs in each grid; calculate the average size and distribution range of individual mitten crabs in each grid; (3) Resource quantity estimation: Estimate the shell width based on bright spot patches, using the following formula: ,in For the estimated carapace width in mm, A spot To determine the area of ​​the bright spots, it is necessary to convert them into actual physical areas based on the sonar's lateral resolution and the sampling interval along the flight direction, and then input them into the calculation. , The correction coefficients were obtained through actual measurement and calibration; weight estimation was performed using an allometric growth model. Where M is body weight in grams and W is carapace width in millimeters. , Species-specific parameters were used; combined with a model of individual size and allometric growth of Chinese mitten crabs, the total weight of Chinese mitten crabs in each grid and the entire pond was estimated; correction coefficients were introduced for different substrate types to improve the estimation accuracy; (4) Visualization: Generate heat maps of the density and size distribution of mitten crabs to visualize the spatial distribution of resources.

8. The system for assessing the quantity and size of mitten crabs based on side-scan sonar images according to claim 1, characterized in that, The side-scan sonar frequency can be selected in the range of 400–1200 kHz according to the actual pond area and water depth; the operating frequency is 900–1200 kHz for high frequency and 400–600 kHz for low frequency; the pulse length is 20–50 microseconds; the maximum single-side range is 15–25 m; the sampling rate is not less than 50 kHz; the speed is 1.0–2.0 m / s; and the installation depth is 0.3–0.5 m below the water surface.

9. The system for assessing the quantity and size of mitten crabs based on side-scan sonar images according to claim 1, characterized in that, Dual-frequency side-scan sonar is used, combined with high-frequency and mid-frequency images to improve recognition robustness.

10. A system for assessing the quantity and size of mitten crabs based on side-scan sonar images according to claim 1, characterized in that, Bright spot-shadow recognition employs rule-based discrimination methods, including thresholding and morphological discrimination, or deep learning models with different structures, including U-Net and YOLO. When shadow features are insufficient, recognition is based solely on bright spot morphology and intensity features.

11. A device equipped with the system according to any one of claims 1-10, characterized in that, The carrier of the system can be a manually piloted small boat, a remotely controlled boat, or a fully autonomous unmanned boat.