Method for fish school size estimation based on forward-looking sonar data and corresponding product

Abstract

The application relates to the technical field of aquaculture monitoring, and provides a fish school quantity estimation method based on forward-looking sonar data and a corresponding product. The method comprises the following steps: acquiring sonar monitoring data obtained by monitoring a target water area by a forward-looking sonar device, and constructing an effective monitoring area in the sonar monitoring data according to a field of view parameter of the forward-looking sonar device; pre-processing the sonar monitoring data located in the effective monitoring area to enhance the distinguishability of fish school targets; identifying the fish school targets located in the effective monitoring area based on the pre-processed sonar monitoring data, and generating an initial quantity estimation value according to the quantity of the identified fish school targets; acquiring a gathering feature reflecting the gathering state of the fish schools in the effective monitoring area; and correcting the initial quantity estimation value based on the gathering feature of the gathering state of the fish schools in the effective monitoring area to obtain a corrected fish school quantity estimation result.

Description

Technical Field

[0001] This application relates to the field of aquaculture monitoring technology, and in particular to a method for estimating fish population size based on forward-looking sonar data and a corresponding product. Background Technology

[0002] Forward-looking sonar, as an active acoustic detection device, can overcome the limitation of poor optical penetration in water and intuitively obtain the distance, orientation and outline information of underwater targets. It has been widely used in intensive aquaculture scenarios such as cages and ponds to achieve real-time, non-invasive monitoring of fish distribution, behavior and scale.

[0003] In existing techniques for estimating fish populations using forward-looking sonar, a common approach is to process the echo images acquired by the sonar, identify discrete target regions representing single or small numbers of fish using image segmentation or target detection algorithms, and then estimate the fish population size by counting the number of these independent target regions. This method can achieve relatively reliable estimation results when the fish population is dispersed and the individual echoes have high separation.

[0004] However, in actual aquaculture, especially under specific conditions such as feeding, fish schools tend to congregate highly in the feeding area due to their feeding behavior. In this state, a large number of fish overlap closely in space, resulting in large, bright, continuous areas in the sonar echo image, blurring or even eliminating the boundaries between individual fish targets. At this point, existing technologies that rely on identifying and statistically analyzing independent target regions face serious technical flaws: because targets are difficult to effectively segment and distinguish, the number of targets identified by the system will be significantly lower than the actual size of the fish school, and may even decrease when the fish school is at its densest due to target fusion. This problem of undercounting and misjudgment significantly reduces the accuracy, rationality, and engineering applicability of existing technologies in estimating fish numbers under this critical and common condition of fish aggregation. Summary of the Invention

[0005] Therefore, it is necessary to address the problems of existing technologies, such as the difficulty in accurately counting fish when they are highly concentrated, resulting in seriously low and unreasonable fish count estimates, and to propose a fish population estimation method and corresponding products based on forward-looking sonar data.

[0006] Firstly, a method for estimating fish population size based on forward-looking sonar data is provided, the method comprising: Acquire sonar monitoring data obtained by forward-looking sonar equipment monitoring the target water area, and construct an effective monitoring area in the sonar monitoring data based on the field of view parameters of the forward-looking sonar equipment; Preprocessing is performed on sonar monitoring data located within the effective monitoring area to enhance the distinguishability of fish targets; Based on the preprocessed sonar monitoring data, fish targets located within the effective monitoring area are identified, and an initial quantity estimate is generated based on the number of identified fish targets. Acquire aggregation characteristics that reflect the fish aggregation status within the effective monitoring area; Based on the clustering characteristics, the initial fish population estimate is corrected to obtain a corrected fish population estimate.

[0007] Secondly, a fish population estimation device based on forward-looking sonar data is provided, the device comprising: The construction module is used to acquire sonar monitoring data obtained by forward-looking sonar equipment from monitoring the target water area, and to construct an effective monitoring area in the sonar monitoring data according to the field of view parameters of the forward-looking sonar equipment. The preprocessing module is used to preprocess the sonar monitoring data located within the effective monitoring area to enhance the distinguishability of fish targets. The generation module is used to identify fish targets located within the effective monitoring area based on preprocessed sonar monitoring data, and generate an initial quantity estimate based on the number of identified fish targets. The acquisition module is used to acquire aggregation characteristics that reflect the aggregation status of fish in the effective monitoring area; The correction module is used to correct the initial quantity estimate based on the clustering features to obtain a corrected fish population estimate.

[0008] Thirdly, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described method for estimating the number of fish swarms based on forward-looking sonar data.

[0009] Fourthly, a computer-readable storage medium is provided, the computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described method for estimating the number of fish swarms based on forward-looking sonar data.

[0010] As can be seen from the technical solution provided in this application, on the one hand, by first preprocessing the sonar monitoring data to enhance the distinguishability of fish targets and identifying targets to generate initial quantity estimates, and then specifically acquiring aggregation features reflecting the aggregation state of the fish school, and correcting the initial estimate based on these features, a closed-loop process of "identification-evaluation-correction" is constructed. This can effectively address the situation where the target boundary is blurred and difficult to distinguish due to the high aggregation of fish schools. In the aggregation state, the system can automatically trigger correction logic based on the aggregation features, thereby improving the shortcomings of traditional methods in such scenarios where the count is seriously understated or the results are unreasonable due to target omission or false detection, and improving the overall accuracy and reliability of quantity estimation in complex real-world scenarios. On the other hand, the basis for correcting the initial quantity estimate is the dynamically acquired aggregation features directly from the sonar monitoring data, rather than fixed empirical parameters. These aggregation features can reflect the spatial distribution density of the fish school in the current monitoring image in real time. The method of obtaining aggregation features and making corrections based on these dynamic characteristics ensures that the final quantity estimate is no longer a simple summation of discrete target quantities, but rather correlates with the actual density and scale change trend of the fish school. This allows the estimate to present a more continuous and reasonable quantity change curve when facing the dynamic change process of the fish school from dispersion to aggregation, avoiding abrupt changes or counterintuitive declines. This improves the continuity and trend rationality of the estimate over time, which is more conducive to subsequent data analysis and aquaculture decisions. Thirdly, the method of obtaining aggregation features and the specific correction strategy can be adapted to different sonar equipment parameters and aquaculture environments. Therefore, the technical solution of this application is not only applicable to solving the aggregation problem during feeding, but also capable of handling other scenarios that may lead to dense fish distribution. This adaptive design, which does not rely on specific prior knowledge, enhances the universality and transferability of the technical solution of this application in different application scenarios and under different equipment conditions. In summary, the technical solution of this application effectively solves the problem of inaccurate counting by traditional methods when fish schools are highly aggregated by introducing aggregation features to dynamically correct the initial identification results, thus improving the accuracy and rationality of the estimate. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein: Figure 1 A flowchart illustrating a fish population estimation method based on forward-looking sonar data provided in an embodiment of this application; Figure 2Sonar images and counting diagrams of fish in a dispersed state provided in an embodiment of this application; Figure 3 Sonar images and counting diagrams of fish in a school-like state provided in this application embodiment; Figure 4a and Figure 4b A comparative diagram of different quantity estimation methods provided in the embodiments of this application under the condition of fish gathering; Figure 5a and Figure 5b Typical sonar phenomena under different fish school conditions are provided in the embodiments of this application; Figure 6 Comparison of counting results of the initial counting method provided in this application under different fish population distribution states; Figure 7a and 7b A comparison diagram of the distribution of the initial counting frames under different fish swarm conditions provided in the embodiments of this application; Figure 8 A structural block diagram of a fish population estimation device based on forward-looking sonar data provided in this application embodiment; Figure 9 This is a structural block diagram of a computer device provided in an embodiment of this application. Detailed Implementation

[0012] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0013] In existing technologies for estimating fish populations using forward-looking sonar, a common approach is to process the sonar echo images, identify discrete target regions representing single or small numbers of fish using image segmentation or target detection algorithms, and then estimate the fish population size by counting the number of these independent target regions. This method can achieve relatively reliable estimation results when the fish population is dispersed and the individual echoes are highly separated. However, in actual aquaculture, especially under specific conditions such as feeding, fish populations tend to cluster heavily in the feeding area due to feeding behavior. In this state, a large number of fish overlap closely in space, resulting in large, bright, continuous areas in the sonar echo images, blurring or even eliminating the boundaries between individual fish targets. At this point, the aforementioned existing technologies that rely on identifying and counting independent target regions face serious technical drawbacks: because targets are difficult to effectively segment and distinguish, the number of targets identified by the system will be significantly lower than the actual size of the fish population, and may even decrease when the fish population is at its densest due to target fusion. These omissions and misjudgments significantly reduce the accuracy, rationality, and engineering applicability of existing quantity estimation methods in the critical and common working condition of fish gathering.

[0014] To address the aforementioned problems in existing technologies, this application proposes a method for estimating fish population size based on forward-looking sonar data. The main process is as follows: Figure 1 As shown, it mainly includes steps S1 to S5, which are detailed below: Step S1: Obtain sonar monitoring data from the forward-looking sonar device monitoring the target water area, and construct an effective monitoring area in the sonar monitoring data based on the field of view parameters of the forward-looking sonar device.

[0015] Step S1 aims to extract a high-quality region to be analyzed from the raw sonar data. Forward-looking sonar devices (e.g., multibeam sonar) emit sound waves into the water and receive echoes, forming a two-dimensional acoustic image containing distance, azimuth, and echo intensity. However, the raw image often contains strong noise interference such as near-field blind zones, invalid edge regions, water surface reflections, and fixed echoes from the bottom of the pool. Processing directly across the entire image would significantly affect subsequent identification accuracy. Therefore, this application first constructs an effective monitoring region in the sonar monitoring data based on the field-of-view parameters of the forward-looking sonar device. Specifically, this can involve: reading the inherent emission opening angle θ, maximum detection distance R_max, and single-frame sonar monitoring data (represented as a two-dimensional digital image in the system) of the forward-looking sonar device, and determining the mapping relationship between its pixel coordinate system and the physical detection space, i.e., the actual distance value and azimuth angle corresponding to each pixel in the raw sonar monitoring data image; using the corresponding position of the forward-looking sonar transducer in the raw sonar monitoring data image as the origin, and based on the mapping relationship between the pixel coordinate system and the physical detection space, determining the physical emission opening angle... and detection range Convert to image pixel coordinates, that is, calculate the two sides of the sector (corresponding to azimuth angles ±). / 2) The column coordinates in the image and the maximum detection range. Find the row coordinates of the pixel in the image; create a binary matrix (i.e., a mask) with the same size as the original sonar image, traverse each pixel position of the binary matrix, and determine whether it simultaneously satisfies the following two conditions based on its row and column coordinates: 1) The distance value corresponding to the pixel is less than or equal to the distance value of the original sonar image. , and 2) the azimuth angle corresponding to this pixel is between - / 2 and + If the value is between 2 and 1, it is marked as the first value (e.g., "1", representing valid) in the mask at that pixel location; otherwise, it is marked as the second value (e.g., "0", representing invalid). The generated binary mask is multiplied pixel by pixel with the original sonar monitoring data image to output a clean sonar image that contains only the data within the effective monitoring area of ​​the sector and filters out the monitoring data outside the area.

[0016] In the above embodiments, the emission angle of the forward-looking sonar device determines the horizontal coverage width of the sound beam, while the detection range of the forward-looking sonar device determines the effective range of the sonar. Together, they define a fan-shaped detection area with the sonar transducer as the origin, the detection range as the radius, and the emission angle as the subtended angle. On the digitized sonar image, based on the emission angle and detection range of the forward-looking sonar device, a fan-shaped area is constructed in the sonar monitoring data as the effective monitoring area; that is, the pixel area corresponding to this fan-shaped physical space is identified through coordinate mapping. For example... Figure 2 The figure shows a sonar image and counting diagram of a dispersed fish school according to an embodiment of this application. The shape of the effective fan-shaped monitoring area is clearly shown in the figure, and monitoring data outside the fan-shaped area is filtered out. This means that all subsequent processing is only performed on pixels within the fan-shaped area, and data outside the area is directly ignored or set to zero. Step S1 actively removes a large amount of background interference and invalid information from the data source, providing a clean and physically meaningful analysis range for subsequent processing, significantly improving the signal-to-noise ratio.

[0017] Step S2: Preprocess the sonar monitoring data located within the effective monitoring area to enhance the distinguishability of fish targets.

[0018] After delineating the effective monitoring area, the data within that area still needs enhancement to improve target recognition rates. This is because sound waves attenuate as they propagate in water, resulting in weak echoes from distant targets; the presence of suspended matter in the water or noise generated by the equipment itself; and insufficient contrast between the target and background. Figure 3The image shown is a sonar image and counting diagram of fish in a clustered state, as provided in an embodiment of this application. Whether the fish are dispersed or clustered, proper preprocessing is fundamental for accurate identification.

[0019] Therefore, this application performs a series of preprocessing operations on the data within the effective monitoring area. Preprocessing the sonar monitoring data located within the effective monitoring area includes at least one of the following steps S2.1 to S2.3: Step S2.1: Normalize the sonar echo intensity within the effective monitoring area.

[0020] The purpose of this operation is to compensate for the geometric diffusion loss and absorption attenuation that occur as sound waves travel further, ensuring that targets of the same size exhibit similar brightness at different distances. A common implementation method is time-varying gain control, which multiplies the echo signals at different distances (corresponding to different rows in the image) by a distance-dependent gain coefficient. This coefficient increases with distance, thus counteracting the attenuation effect. This prevents fish targets located at the far end of the fan-shaped area from "disappearing" from the image due to weak signals, ensuring consistent target detection sensitivity throughout the effective monitoring area.

[0021] Step S2.2: Perform background noise suppression processing on the sonar monitoring data within the effective monitoring area.

[0022] Step S2.2 aims to eliminate relatively stable environmental background interference. An effective method is background modeling and subtraction. The system can continuously learn or pre-acquire a sequence of sonar images within a period without fish (or with sparse fish), and establish a background intensity model (such as mean, median, or a more complex Gaussian model) for each pixel. When processing the current frame, each pixel value of the current image is subtracted from its corresponding background model value. Areas with significantly non-zero differences in the image may be foreground targets (fish). This effectively suppresses echoes from fixed structures such as the pool bottom and walls, as well as slowly changing uniform noise in the water, thereby "highlighting" moving fish targets.

[0023] Step S2.3: Perform contrast enhancement processing on the sonar monitoring data within the effective monitoring area.

[0024] Step S2.3 can increase the grayscale difference between the target and the background, making the boundary clearer. Histogram equalization can be used to transform the grayscale histogram of the original image from a relatively concentrated grayscale range to a uniform distribution across the entire grayscale range, thereby increasing local contrast. Alternatively, gamma correction can be used to non-linearly transform the overall brightness of the image, enhancing details in darker or brighter areas. After contrast enhancement, the boundary between the bright areas of the fish target and the surrounding dark water is more distinct, greatly facilitating the extraction of edge and contour features by subsequent target detection algorithms.

[0025] By combining one or more of the preprocessing steps exemplified in steps S2.1 to S2.3 above, the distinguishability of fish targets can be significantly enhanced, creating high-quality input conditions for robust target recognition.

[0026] Step S3: Based on the preprocessed sonar monitoring data, identify fish targets located within the effective monitoring area, and generate an initial quantity estimate based on the number of identified fish targets.

[0027] In existing technologies, fish population estimation typically concludes at this step, directly outputting the number of identified targets. However, as... Figure 5a and Figure 5b The image shows typical sonar phenomena under different fish school conditions as provided in the embodiments of this application. Figure 5a As shown, when the fish are dispersed, the sonar echoes appear as discrete bright spots, making single-target identification and counting relatively reliable. However, under the influence of feeding or other inducements, the fish will highly aggregate, and the sonar echoes will merge into continuous, large-area bright clumps, causing the boundaries of individual fish to become completely blurred, such as... Figure 5b As shown. At this point, any recognition algorithm that relies on distinguishing individual fish will reach a theoretical performance ceiling, and the number of identified targets will be far lower than the actual size of the fish school, even exhibiting the anomaly that the denser the fish school, the lower the count. For example... Figure 6 The figure shown is a comparison of the counting results of the initial counting method provided in this application under different fish population distribution states. The figure intuitively illustrates the failure of the traditional method (i.e., the initial output of this step) in the aggregated state: the counting curve fluctuates drastically and shows a downward trend.

[0028] In view of this, this application repositions this step. This application performs the process of identifying fish targets located within the effective monitoring area based on preprocessed sonar monitoring data, and generates an initial quantity estimate based on the number of identified fish targets. However, its purpose is not to use this as the final result, but rather as a benchmark input and problem indicator for subsequent intelligent correction processes. Specifically, as an embodiment of this application, identifying fish targets located within the effective monitoring area based on preprocessed sonar monitoring data can be achieved through the following steps S3.1 and S3.2: Step S3.1: Process the preprocessed sonar monitoring data using a machine learning-based target detection model.

[0029] Specifically, well-trained deep learning object detection models can be used, such as the YOLO (You Only Look Once) series, SSD (Single Shot MultiBox Detector), or Faster R-CNN (Region-Based Convolutional Neural Network). These models can automatically extract features from preprocessed sonar images and output detection results. The object detection model outputs bounding box information for each identified fish target, including at least the location and size of the bounding box. A "bounding box" is typically represented by a rectangle, its location defined by the image coordinates (x, y) of its center point, and its size defined by the width (w) and height (h) of the box. Each bounding box represents a region where the model believes a single fish or a small group of closely packed fish exists.

[0030] Step S3.2: The target detection model outputs the bounding box information of each identified fish group target, wherein the bounding box information of the identified fish group target includes at least the position and size of the bounding box.

[0031] The total number of bounding boxes output by the object detection model within the effective monitoring area of ​​the current frame is counted. This number is then used directly as the initial quantity estimate. Figure 7a and 7b The image shown is a comparison of the distribution of the initial counting frames under different fish swarm conditions provided in this application embodiment. In the dispersed state (corresponding to...) Figure 7a The initial value itself is already quite accurate, given the large number of bounding boxes, their uniform distribution, and their high degree of agreement with discrete bright spots. However, in the clustered state (corresponding to...), the initial value may still be problematic. Figure 7b The number of bounding boxes decreases drastically, and can only outline the outer contours of large clusters, failing to reflect the number of internal individuals. In this case, the initial value will be significantly underestimated.

[0032] Through the above design, step S3 achieves a dual purpose: first, under normal conditions where the fish are dispersed, it provides directly usable and accurate counting results; second, and more importantly, under challenging conditions where the fish are clustered, it quantitatively exposes the inherent defects of traditional single-target identification methods (i.e., the initial estimate is too low), thus providing clear targets that need to be corrected and a quantified signal (abnormally low initial value) to trigger correction in subsequent steps S4 and S5. Therefore, this initial estimate is a crucial bridge connecting the defects of traditional methods with the innovative solution of this application.

[0033] Step S4: Obtain aggregation characteristics that reflect the aggregation status of fish in the effective monitoring area.

[0034] As previously mentioned, the initial quantity estimate provided in step S3 becomes severely inaccurate when the fish are highly clustered. Traditional improvement approaches often limit themselves to optimizing the target detection model itself, attempting to "separate" more targets in highly overlapping images, which is largely ineffective when the physical signals are already severely aliased. This application takes a different approach, raising a fundamental question: since it is impossible to accurately "count" how many individuals there are, can we instead "perceive" the degree to which the fish have "clustered"? This degree of clustering itself is the root cause of the initial count's failure and should be the key basis for correcting the count.

[0035] Therefore, this application introduces a mechanism to obtain aggregation characteristics reflecting the aggregation state of fish schools within an effective monitoring area. Its purpose is to bypass the limitation of "several individuals" and extract one or more indicators that can quantify the "degree of aggregation" from the macroscopic spatial distribution of fish schools. This application provides multiple parallel technical paths to achieve this goal. These paths characterize the aggregation state from different dimensions, as shown in Implementation Method 1, Implementation Method 2, and Implementation Method 3 below: 1) Implementation Method 1: This implementation method is based on macroscopic coverage area. Specifically, it can be as follows: Calculate the total coverage area of ​​the target regions of all identified fish targets within the effective monitoring area; use the ratio of the total coverage area to the area of ​​the effective monitoring area as the first aggregation feature characterizing the degree of fish aggregation. Here, the "target region" refers to the set of pixels covered by each bounding box obtained in step S3.1. To calculate the total coverage area of ​​the target regions of all identified fish targets within the effective monitoring area, a union operation needs to be performed on the pixels of all bounding boxes to ensure that overlapping areas are calculated only once, thus obtaining the total pixel area occupied by the fish targets. The ratio of the total coverage area to the area of ​​the effective monitoring area is used as the first aggregation feature characterizing the degree of fish aggregation. The area of ​​the effective monitoring area is fixed (the total number of pixels in the fan-shaped area). This ratio directly reflects the "area ratio" of the fish targets in the monitoring image. When the fish are dispersed, the bounding boxes are discrete, and the total area ratio is small; when the fish are aggregated, the bounding boxes are dense and overlap each other, and the total area ratio will increase significantly. Therefore, the first aggregation feature is a macroscopic and intuitive quantitative indicator of the degree of aggregation.

[0036] 2) Implementation Method Two: This implementation method is based on microscopic spatial relationships. Specifically, it involves analyzing the spatial distance or overlap between the target areas of each fish swarm. Based on the statistical analysis results of the spatial distance or overlap, a second clustering feature characterizing the spatial density of the fish swarm is generated. This method focuses on the interrelationships between individual targets. Analyzing the spatial distance or overlap between the target areas of each fish swarm involves calculating the Euclidean distance between all pairwise center points of the bounding boxes and then obtaining the average or median of these distances. The overlap ratio is calculated by taking the intersection-union ratio (IU) of any two bounding boxes, i.e., the ratio of the intersection area to the union area of ​​the two boxes. For example, the reciprocal of the average distance can be normalized (the smaller the distance, the larger the feature value), or the proportion of box pairs with an IU greater than a certain threshold (e.g., 0.1) can be directly calculated. The second clustering feature microscopically characterizes whether the targets are arranged in a "loose" or "dense" manner, and can sensitively reflect the process of targets moving closer to each other in the early stages of clustering.

[0037] 3) Implementation Method Three: This implementation method, based on the original acoustic signal intensity distribution, can be: analyzing the distribution of sonar echo brightness within the effective monitoring area; generating a third clustering feature based on the degree of concentration of sonar echo brightness in a local area. The implementation scheme for generating the third clustering feature based on the degree of concentration of sonar echo brightness in a local area in the above embodiments can be: extracting brightness distribution features characterizing the degree of fish aggregation based on the distribution of sonar echo brightness within the effective monitoring area; characterizing the degree of fish aggregation in a local area by statistically analyzing the number or percentage of pixels with brightness exceeding a preset threshold; characterizing the overall distribution state of the fish group by statistically analyzing the mean brightness, variance brightness, or degree of concentration of brightness distribution. Specifically, this includes: within the effective monitoring area, defining the analysis range, i.e., the local area, for calculating brightness concentration (for example, using a sliding window method to divide the effective monitoring into multiple overlapping or adjacent sub-windows, using these sub-windows as the analysis range for calculating brightness concentration); for the current local area, using image texture statistics or high-brightness pixel density statistics... Calculate one or more mathematical indicators that can quantify the concentration of sonar echo brightness within the local area. For example, calculate the gray-level co-occurrence matrix of the local area image based on image texture statistics, and extract the contrast or entropy value from the gray-level co-occurrence matrix as an indicator. Set a high brightness threshold based on high brightness pixel density statistics, count the number of pixels in the current local area whose gray value is greater than the high brightness threshold, and calculate the ratio of this ratio to the total number of pixels in the local area. Use this high brightness pixel density ratio as an indicator. Output the calculated one or more of the above mathematical indicators directly or after normalization and weighted combination as the third clustering feature. For example, when the local area is the entire effective monitoring area, the feature value is directly output. When using the sliding window method, the maximum value, average value, or spatial distribution entropy of the indicators calculated from all sub-windows can be used as the final third clustering feature, and so on.

[0038] The third implementation method described above does not rely entirely on potentially invalid target recognition results, but directly analyzes the preprocessed original grayscale image. It analyzes the distribution of sonar echo brightness within the effective monitoring area. For example, it can calculate image texture features within the entire fan-shaped area or sliding window, such as the contrast or entropy of the gray-level co-occurrence matrix; or simply count the density of pixels with grayscale values ​​exceeding a certain high threshold. Areas where fish are highly concentrated necessarily correspond to areas with strong acoustic reflection, which manifests as a high concentration of bright pixels in the image. By quantifying this brightness concentration (e.g., the degree of clustering of bright pixels), the third clustering feature can directly reflect the clustering intensity of the acoustic signal, and this feature remains effective even when the number of target recognition boxes is small.

[0039] Step S5: Based on the aggregation characteristics of fish populations within the effective monitoring area, the initial fish population estimate is corrected to obtain the corrected fish population estimate.

[0040] The core logic of step S5 is to correct the "incorrect measurement values" (initial quantity estimates) generated in step S3 based on the "severity of the condition" (clustering characteristic values) diagnosed in step S4. The basic idea is that the higher the degree of clustering, the greater the possibility of underestimation of the initial estimate, and the stronger the correction should be. Specifically, As an embodiment of this application, the correction of the initial quantity estimate based on the clustering feature can be achieved as follows: Determine whether the value of the clustering feature exceeds a preset clustering judgment threshold; if not, directly output the initial quantity estimate as the corrected fish population estimate; if it exceeds, then, based on the value of the clustering feature, weight amplify or segmentally adjust the initial quantity estimate to obtain the corrected fish population estimate. In the above embodiment, the clustering judgment threshold is based on historical data or experimental calibration and is used to distinguish between "normal dispersion" and "clustering requiring correction." This threshold judgment mechanism ensures that the method maintains the accuracy of traditional methods when the fish population is dispersed, and only initiates correction when clustering occurs, thus combining efficiency and accuracy. When correction is triggered, weighting the initial quantity estimate based on the value of the clustering feature is a direct and effective method. Specifically, it can be as follows: determine a correction coefficient greater than 1 based on the value of the clustering feature; multiply the initial quantity estimate by the correction coefficient to obtain the corrected fish population estimate. The correction coefficient can be a linear function of the clustering feature value, for example, K = 1 + α * ( - T) ,in, α This is the gain coefficient. This is a clustering characteristic value, specifically the ratio of the total coverage area of ​​the fish swarm target to the effective monitoring area. T A preset threshold is used to distinguish whether a school of fish is in a state of aggregation, for example, T It can be set to 0.3. α This is a preset gain coefficient used to control the strength of the correction. α The value range is [0.5, 2.0], and the specific value can be determined experimentally, for example, by fitting the value based on the rate of change in the number of fish before and after feeding. When ≤ T At that time, it was determined that the fish had not gathered together. K =1, no correction; when > T When it is determined that the fish are gathering, the above formula is used to calculate... K , K >1, which amplifies the initial quantity estimate.

[0041] As another embodiment of this application, the correction of the initial quantity estimate based on the clustering feature can also be achieved by: when using the first clustering feature, a more refined strategy can be implemented: when the ratio of the total coverage area to the effective monitoring area is greater than a first threshold, the fish population is determined to be in a highly clustered state; in the highly clustered state, a nonlinear mapping function is used to map the initial quantity estimate to a higher value, which serves as the corrected fish population estimate. Here, the "nonlinear mapping function" (e.g., an exponential function, a sigmoid function) can better simulate the real-world situation: when the coverage area reaches a high proportion, the overlapping and occlusion effect between fish bodies will intensify nonlinearly, accelerating the counting loss, thus requiring nonlinear compensation. This makes the correction result more consistent with physical laws in extreme clustering situations.

[0042] As another embodiment of this application, the correction of the initial quantity estimate based on the clustering feature can also be achieved by: counting the number of overlapping pixels between target regions or calculating the average distance between the centers of target regions; when the number of overlapping pixels exceeds a second threshold or the average distance is less than a third threshold, the correction weight of the initial quantity estimate is increased. This means that the correction strength is not a fixed value, but is dynamically adjusted according to the real-time changes in the "crowding" degree between targets, achieving more refined adaptive compensation.

[0043] To improve the robustness and accuracy of the system and avoid misjudgment based on a single feature, a fusion strategy can be adopted. Specifically, as another embodiment of this application, correcting the initial fish population estimate based on clustering features can also involve: obtaining at least two different clustering features; assigning a weight to each clustering feature based on its value; calculating a correction sub-result based on each clustering feature and its weight; and combining all correction sub-results to generate the final corrected fish population estimate. For example, both area proportion and average intersection-union ratio can be used simultaneously, employing weighted averaging, voting, or rule-based fusion methods to combine their judgments. This method fully utilizes the complementarity of information from different dimensions, making the final correction decision more reliable. Figure 4a and Figure 4b The diagram shown is a comparison of different fish quantity estimation methods provided in the embodiments of this application under the condition of fish gathering. Figure 4a It is an estimation result that relies solely on the initial count. Figure 4b This is the estimation result after introducing clustering feature correction. The figure visually compares the difference between relying solely on the initial count and the result after introducing clustering feature correction, showing that the method in this application can make the estimation trend more consistent with reality.

[0044] By combining steps S4 and S5 above, this application successfully transforms the problem of estimating the "number of individuals" into the problem of perceiving the "aggregation state" and compensating for the "counting error". Thus, even when the traditional method fails due to the high aggregation of fish, it can still output reasonable, stable and trend-correct quantity estimation results.

[0045] After generating the initial quantity estimate in step S3 and before performing clustering correction in step S5, this application introduces a key optimization step aimed at further improving the temporal stability and reliability of the output results. This is because, in actual continuous monitoring, even if the fish population is in a relatively stable dispersed state, a single frame of sonar image may experience random instantaneous fluctuations in target identification results due to instantaneous disturbances in the water, rapid turning of individual fish, or image noise. Such fluctuations do not represent changes in the actual number of fish. If the initial estimate with such fluctuations is directly fed into the correction stage, it may cause unnecessary jitter in the correction results, affecting the smoothness of long-term observations and trend judgment.

[0046] To address the aforementioned issues, this application introduces a temporal smoothing mechanism before correction: After generating an initial quantity estimate based on the identified fish population, and before correcting the initial quantity estimate based on clustering characteristics, the method further includes: setting a sliding time window to statistically analyze the initial quantity estimates over multiple consecutive time periods, generating a statistically processed initial quantity estimate for correction. As an embodiment of this application, statistically analyzing the initial quantity estimates over multiple consecutive time periods to generate a statistically processed initial quantity estimate can be achieved by: acquiring multiple initial quantity estimates within the current time period and a preset time period prior; calculating the average, maximum, or weighted average of the multiple initial quantity estimates, and using the calculation result as the statistically processed initial quantity estimate. Specifically, the system maintains a first-in, first-out queue of length N (e.g., corresponding to the past 10 seconds), continuously storing the latest N "initial quantity estimates" output in step S3. Whenever a new initial estimate arrives, it is added to the queue, and the oldest value is removed. Subsequently, the average, maximum, or weighted average of the multiple initial quantity estimates is calculated. Using a "sliding average" is the most common method, effectively filtering out random high-frequency noise. In other embodiments, taking the "maximum value" can avoid instantaneous underestimation due to missed detections in a single frame, ensuring that the smoothed value is closer to the possible upper limit. The "weighted average" can assign higher weights to more recent data, preserving better real-time responsiveness while smoothing the data. The calculated result is used as the initial quantity estimate after statistical processing. This value replaces the original initial value for a single frame and is sent to steps S4 and S5 for the acquisition and correction calculation of clustered features. This operation significantly improves the smoothness of the data stream.

[0047] To further enhance adaptability, the sliding time window setting in the above embodiments may also include dynamically adjusting the duration of the sliding time window based on the rate of change or fluctuation of the initial quantity estimate over time. This is because a fixed-length sliding window may not be suitable for all operating conditions. For example, at the start of feeding, when the fish population rapidly shifts from dispersed to concentrated, the actual fish population changes drastically. If the window is too long at this time, the smoothing effect will be excessive, leading to a sluggish system response and masking the true upward trend. Conversely, when the fish are calmly swimming, a longer window is needed to adequately suppress random fluctuations. Therefore, the system can monitor the statistical characteristics of the initial estimate sequence in real time and dynamically adjust the duration of the sliding time window based on the rate of change or fluctuation of the initial quantity estimate over time. One implementation method is to calculate the variance or standard deviation of recent estimates. When the variance exceeds a high threshold, it indicates that the fish population state may be changing drastically, and the system automatically shortens the window length (e.g., from 10 seconds to 3 seconds) to improve response speed; when the variance is below a low threshold, it indicates that the state is stable, and the system automatically increases the window length (e.g., from 10 seconds to 20 seconds) to enhance the smoothing effect. This dynamic adjustment mechanism enables the system to achieve an intelligent balance between "suppressing noise" and "tracking real changes".

[0048] Through the technical solutions described above, the system ultimately outputs a corrected estimate of the fish population size. This result maintains high accuracy based on single-target recognition when the fish are dispersed, avoids severe underestimation through aggregation feature perception and correction mechanisms when the fish are highly clustered, and ensures the stability of the curve through smoothing processing in the time dimension. Thus, it outputs a continuous, reasonable, and more realistic estimate curve reflecting the trend of fish population size changes throughout the entire monitoring period.

[0049] Figure 1 The example method can also apply the corrected fish population estimate to feeding management decisions or fish behavior analysis in aquaculture areas. In feeding management decisions: the system can monitor the estimated fish population near feeding points in real time. When the population reaches or remains at a high level, it indicates active feeding, and the feeding strategy can be maintained or adjusted. When the population begins to decline, it may indicate that feeding is approaching saturation, providing direct quantitative evidence for optimizing feeding amounts and reducing feed waste, thus achieving precision feeding. In fish behavior analysis: long-term recording and analysis of the time series, spatial distribution (combined with location information), and aggregation characteristics of the estimated fish population can be used to study the diurnal behavioral rhythms of fish in response to different environmental factors (such as water temperature, dissolved oxygen, and light), and their response patterns to feeding signals, providing data support for optimizing aquaculture processes and improving animal welfare.

[0050] From the above Figure 1The example of a fish population estimation method based on forward-looking sonar data demonstrates two key aspects. First, by preprocessing the sonar monitoring data to enhance the distinguishability of fish targets and generating an initial population estimate, and then specifically acquiring aggregation features reflecting the fish's aggregation state, the initial estimate is corrected based on these features. This constructs a closed-loop process of "identification-evaluation-correction," effectively addressing situations where fish populations are highly aggregated, leading to blurred target boundaries and difficulty in differentiation. In the aggregated state, the system can automatically trigger correction logic based on aggregation features, thereby improving the shortcomings of traditional methods that suffer from severely understated counts or unreasonable results due to missed or false detections of targets in such scenarios. This enhances the overall accuracy and reliability of population estimation in complex real-world scenarios. Second, the basis for correcting the initial population estimate is the dynamically acquired aggregation features directly derived from the sonar monitoring data, rather than fixed empirical parameters. These aggregation features reflect the spatial distribution of the fish population in the current monitoring image in real time. By modifying the density or coverage based on this dynamic characteristic, the final quantity estimate is no longer a simple summation of discrete target quantities, but rather correlated with the actual density and scale change trend of the fish school. This allows the estimate to present a more continuous and reasonable quantity change curve when facing the dynamic change process of the fish school from dispersion to aggregation, avoiding abrupt changes or counterintuitive declines, thereby improving the continuity and trend rationality of the estimate over time series, which is more conducive to subsequent data analysis and aquaculture decisions. Thirdly, the method of obtaining aggregation characteristics and the specific correction strategy can be adapted to different sonar equipment parameters and aquaculture environments. Therefore, the technical solution of this application is not only applicable to solving the aggregation problem during feeding, but also capable of dealing with other scenarios that may lead to dense fish distribution. This adaptive design, which does not rely on specific prior knowledge, enhances the universality and transferability of the technical solution of this application in different application scenarios and under different equipment conditions. In summary, the technical solution of this application effectively solves the problem of inaccurate counting by traditional methods when fish schools are highly aggregated by introducing aggregation characteristics to dynamically correct the initial identification results, thus improving the accuracy and rationality of the estimate.

[0051] Please see Figure 8 As shown, in one embodiment, a fish population estimation device based on forward-looking sonar data is provided. This device may include a construction module 801, a preprocessing module 802, a generation module 803, an acquisition module 804, and a correction module 805, as detailed below: The construction module 801 is used to acquire sonar monitoring data obtained by forward-looking sonar equipment from monitoring the target water area, and to construct an effective monitoring area in the sonar monitoring data based on the field of view parameters of the forward-looking sonar equipment. The preprocessing module 802 is used to preprocess sonar monitoring data located within the effective monitoring area to enhance the distinguishability of fish targets. The generation module 803 is used to identify fish targets located within the effective monitoring area based on preprocessed sonar monitoring data, and generate an initial quantity estimate based on the number of identified fish targets. The acquisition module 804 is used to acquire aggregation characteristics that reflect the aggregation status of fish in the effective monitoring area; The correction module 805 is used to correct the initial fish population estimate based on the aggregation characteristics of the fish population aggregation state in the effective monitoring area, so as to obtain the corrected fish population estimate.

[0052] From the above appendix Figure 8 As illustrated by the example of a fish population estimation device based on forward-looking sonar data, on the one hand, by preprocessing the sonar monitoring data to enhance the distinguishability of fish targets and generating an initial population estimate, and then specifically acquiring aggregation features reflecting the fish population state, the initial estimate is corrected based on these features. This constructs a closed-loop process of "identification-evaluation-correction," which effectively addresses situations where fish populations are highly aggregated, leading to blurred target boundaries and difficulty in distinguishing them. In the aggregated state, the system can automatically trigger correction logic based on the aggregation features, thereby improving the shortcomings of traditional methods in such scenarios, which result in severely understated counts or unreasonable results due to missed or false detections of targets. This enhances the overall accuracy and reliability of population estimation in complex real-world scenarios. On the other hand, the basis for correcting the initial population estimate is the dynamically acquired aggregation features directly derived from the sonar monitoring data, rather than fixed empirical parameters. These aggregation features can reflect the spatial distribution of the fish population in the current monitoring image in real time. By modifying the density or coverage based on this dynamic characteristic, the final quantity estimate is no longer a simple summation of discrete target quantities, but rather correlated with the actual density and scale change trend of the fish school. This allows the estimate to present a more continuous and reasonable quantity change curve when facing the dynamic change process of the fish school from dispersion to aggregation, avoiding abrupt changes or counterintuitive declines, thereby improving the continuity and trend rationality of the estimate over time series, which is more conducive to subsequent data analysis and aquaculture decisions. Thirdly, the method of obtaining aggregation characteristics and the specific correction strategy can be adapted to different sonar equipment parameters and aquaculture environments. Therefore, the technical solution of this application is not only applicable to solving the aggregation problem during feeding, but also capable of dealing with other scenarios that may lead to dense fish distribution. This adaptive design, which does not rely on specific prior knowledge, enhances the universality and transferability of the technical solution of this application in different application scenarios and under different equipment conditions. In summary, the technical solution of this application effectively solves the problem of inaccurate counting by traditional methods when fish schools are highly aggregated by introducing aggregation characteristics to dynamically correct the initial identification results, thus improving the accuracy and rationality of the estimate.

[0053] Optionally, Figure 8 The example acquisition module 804 may include a calculation unit and a confirmation unit, wherein: The calculation unit is used to calculate the total coverage area of ​​the target area of ​​all identified fish targets within the effective monitoring area; The confirmation unit is used to measure the ratio of the total coverage area to the area of ​​the effective monitoring area as the first aggregation feature characterizing the degree of fish aggregation.

[0054] Optionally, Figure 8 The example acquisition module 804 may include an analysis unit and a generation unit, wherein: The analysis unit is used to analyze the spatial distance or overlap between the target areas of each fish group; The generation unit is used to generate a second clustering feature that characterizes the spatial density of fish groups based on statistical analysis results of spatial distance or overlap.

[0055] Optionally, Figure 8 The example apparatus may also include a statistical analysis module for setting a sliding time window to perform statistical analysis on the initial quantity estimates at multiple consecutive time points, generating statistically processed initial quantity estimates for correction.

[0056] Optionally, Figure 8 The example correction module 805 may include a judgment unit and a mapping unit, wherein: The judgment unit is used to determine that the fish are in a highly concentrated state when the ratio of the total coverage area to the effective monitoring area is greater than a first threshold. The mapping unit is used to map the initial fish population estimate to a higher value using a nonlinear mapping function in highly clustered states, as the corrected fish population estimate.

[0057] Optionally, Figure 8 The example correction module 805 may include a statistical unit and a weight correction unit, wherein: The statistical unit is used to count the number of overlapping pixels between target regions or to calculate the average distance between the centers of target regions. The weight correction unit is used to increase the correction weight of the initial quantity estimate when the number of overlapping pixels exceeds the second threshold or the average distance is less than the third threshold.

[0058] Optionally, Figure 8 The example correction module 805 may include an acquisition unit, an allocation unit, a sub-result calculation unit, and an estimation result generation unit, wherein: An acquisition unit is used to acquire at least two different clustering features; The allocation unit is used to assign a weight to each cluster feature based on the value of each cluster feature; The sub-result calculation unit is used to calculate a modified sub-result based on each clustering feature and its weight. The estimation result generation unit is used to integrate the results of each correction sub-result to generate the final corrected fish population estimation result.

[0059] In one embodiment, a computer device is provided, the internal structure of which can be shown as follows: Figure 9 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computational and control capabilities. The memory includes non-volatile and / or volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface is used to communicate with external clients via a network connection. When executed by the processor, the computer program implements the functions or steps of a fish swarm estimation method based on forward-looking sonar data.

[0060] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, performs the following steps: Acquire sonar monitoring data obtained by forward-looking sonar equipment from monitoring the target water area, and construct an effective monitoring area in the sonar monitoring data based on the field of view parameters of the forward-looking sonar equipment; Preprocessing of sonar monitoring data within the effective monitoring area enhances the distinguishability of fish targets; Based on the preprocessed sonar monitoring data, fish targets located within the effective monitoring area are identified, and an initial quantity estimate is generated based on the number of identified fish targets. Acquire aggregation characteristics that reflect the aggregation status of fish populations within the effective monitoring area; Based on the aggregation characteristics of fish populations within the effective monitoring area, the initial fish population estimate is corrected to obtain a corrected fish population estimate.

[0061] The computer program described above effectively solves the problem of inaccurate counting by traditional methods when fish are highly clustered by introducing clustering features to dynamically correct the initial identification results, thereby improving the accuracy and rationality of the estimation.

[0062] In one embodiment, a computer-readable storage medium is provided, the computer-readable storage medium storing a computer program that, when executed by a processor, performs the following steps: Acquire sonar monitoring data obtained by forward-looking sonar equipment from monitoring the target water area, and construct an effective monitoring area in the sonar monitoring data based on the field of view parameters of the forward-looking sonar equipment; Preprocessing of sonar monitoring data within the effective monitoring area enhances the distinguishability of fish targets; Based on the preprocessed sonar monitoring data, fish targets located within the effective monitoring area are identified, and an initial quantity estimate is generated based on the number of identified fish targets. Acquire aggregation characteristics that reflect the aggregation status of fish populations within the effective monitoring area; Based on the aggregation characteristics of fish populations within the effective monitoring area, the initial fish population estimate is corrected to obtain a corrected fish population estimate.

[0063] The computer program described above effectively solves the problem of inaccurate counting by traditional methods when fish are highly clustered by introducing clustering features to dynamically correct the initial identification results, thereby improving the accuracy and rationality of the estimation.

[0064] The steps described above, implemented when the computer program is executed by the processor, significantly improve computational efficiency, stability, and resource utilization through coordinate pre-calculation, atomic-free kernel design, and circular buffer optimization.

[0065] It should be noted that the functions or steps that can be implemented by the computer-readable storage medium or computer device described above can be referred to the relevant descriptions on the server side and client side in the foregoing method embodiments. To avoid repetition, they will not be described one by one here.

[0066] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0067] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0068] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A method for estimating fish population size based on forward-looking sonar data, characterized in that, The method includes: Acquire sonar monitoring data obtained by forward-looking sonar equipment monitoring the target water area, and construct an effective monitoring area in the sonar monitoring data based on the field of view parameters of the forward-looking sonar equipment; Preprocessing is performed on sonar monitoring data located within the effective monitoring area to enhance the distinguishability of fish targets; Based on the preprocessed sonar monitoring data, fish targets located within the effective monitoring area are identified, and an initial quantity estimate is generated based on the number of identified fish targets. Acquire aggregation characteristics that reflect the fish aggregation status within the effective monitoring area; Based on the clustering characteristics, the initial fish population estimate is corrected to obtain a corrected fish population estimate.

2. The fish population estimation method based on forward-looking sonar data according to claim 1, characterized in that, The acquisition of aggregation characteristics reflecting the fish aggregation status within the effective monitoring area includes: Calculate the total coverage area of ​​the target regions of all identified fish targets within the effective monitoring area; The ratio of the total coverage area to the area of ​​the effective monitoring area is used as the first aggregation feature characterizing the degree of fish aggregation.

3. The fish population estimation method based on forward-looking sonar data according to claim 1, characterized in that, The acquisition of aggregation characteristics reflecting the fish aggregation status within the effective monitoring area includes: Analyze the spatial distance or degree of overlap between the target areas of each fish group; Based on the statistical analysis results of the spatial distance or overlap, a second clustering feature is generated to characterize the spatial density of the fish population.

4. The method according to claim 1, characterized in that, The acquisition of aggregation characteristics reflecting the fish aggregation status within the effective monitoring area includes: Based on the distribution of sonar echo brightness within the effective monitoring area, brightness distribution features characterizing the degree of fish aggregation are extracted. The degree of fish aggregation in a local area is characterized by counting the number or percentage of pixels with brightness higher than a preset threshold. The overall distribution of fish populations can be characterized by statistically analyzing the mean brightness, variance of brightness, or the degree of concentration of brightness distribution.

5. The fish population estimation method based on forward-looking sonar data according to claim 1, characterized in that, After generating an initial quantity estimate based on the identified fish population, and before revising the initial quantity estimate based on the clustering features, the method further includes: A sliding time window is set to perform statistical analysis on the initial quantity estimates at multiple consecutive time points, generating statistically processed initial quantity estimates for correction.

6. The fish population estimation method based on forward-looking sonar data according to claim 2, characterized in that, The step of correcting the initial quantity estimate based on the clustering features includes: When the ratio of the total coverage area to the effective monitoring area is greater than the first threshold, it is determined that the fish are in a highly concentrated state. In a highly clustered state, a nonlinear mapping function is used to map the initial quantity estimate to a higher value, which is then used as the corrected fish population estimate.

7. The fish population estimation method based on forward-looking sonar data according to claim 3, characterized in that, The step of correcting the initial quantity estimate based on the clustering features includes: Count the number of overlapping pixels between target regions or calculate the average distance between the centers of target regions; When the number of overlapping pixels exceeds the second threshold or the average distance is less than the third threshold, the correction weight of the initial quantity estimate is increased.

8. The fish population estimation method based on forward-looking sonar data according to any one of claims 1-3, characterized in that, The step of correcting the initial quantity estimate based on the clustering features includes: Obtain at least two different clustering features; Each clustering feature is assigned a weight based on its numerical value. Based on each aggregation feature and its weight, a modified sub-result is calculated. By combining the results of each correction sub-result, the final corrected estimate of the fish population is generated.

9. A fish population estimation device based on forward-looking sonar data, characterized in that, The device includes: The construction module is used to acquire sonar monitoring data obtained by forward-looking sonar equipment from monitoring the target water area, and to construct an effective monitoring area in the sonar monitoring data according to the field of view parameters of the forward-looking sonar equipment. The preprocessing module is used to preprocess the sonar monitoring data located within the effective monitoring area to enhance the distinguishability of fish targets. The generation module is used to identify fish targets located within the effective monitoring area based on preprocessed sonar monitoring data, and generate an initial quantity estimate based on the number of identified fish targets. The acquisition module is used to acquire aggregation characteristics that reflect the aggregation status of fish in the effective monitoring area; The correction module is used to correct the initial quantity estimate based on the clustering features to obtain a corrected fish population estimate.

10. A computer device, the computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the fish population estimation method based on forward-looking sonar data as described in any one of claims 1 to 8.

11. A storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the fish population estimation method based on forward-looking sonar data as described in any one of claims 1 to 8.

Images