Fish nest distribution design method based on positioning technology

By deploying sensors in the underwater environment and using advanced sonar image processing technology, combined with fish behavior pattern analysis, the problems of incomplete data and lack of personalization in fish nest distribution design in existing technologies have been solved, realizing scientific and effective fish nest distribution design, and improving the quality of fish habitats and the stability of the ecosystem.

CN118940894AInactive Publication Date: 2024-11-12CCCC THIRD HARBOR ENG CO LTD NINTH ENG CO LTD +1
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202410996888.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-24
Publication Date
2024-11-12
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

Existing technologies for fish nest distribution design suffer from incomplete data collection, outdated analysis methods, and a lack of personalized design, making it difficult to accurately reflect the complexity of the underwater environment and meet the specific needs of fish populations.

Method used

By deploying sensors to monitor aquatic environmental parameters and combining advanced sonar image processing technology with fish behavior pattern analysis, the scientific design of fish nest distribution can be achieved.

Benefits of technology

This improves the scientific validity and effectiveness of fish nest distribution design, meets the habitat needs of fish, and promotes the healthy development of aquatic ecosystems.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118940894A_ABST
    Figure CN118940894A_ABST
Patent Text Reader

Abstract

The invention discloses a fish nest distribution design method based on a positioning technology. The method comprises the steps that sensors are deployed in a local water area to monitor and record water environment parameters of all positions; a sonar monitoring system is used for measuring a local water area to obtain a sonar image, and the sonar image is analyzed to obtain a fish school positioning analysis result; activity distribution of the fish schools is determined based on a large number of fish school positioning analysis results, inhabitation behavior pattern analysis is conducted on the fish schools, and inhabitation distribution of the fish schools is judged according to analysis results; performing feature extraction on the water environment parameters of the dense inhabitation distribution points based on the inhabitation distribution of the fish schools, expanding to a larger target water area according to a feature extraction result, and screening position points suitable for inhabitation of the fish schools in the target water area; and carrying out distribution design on the fish nests according to screening results. According to the method, fish school positioning analysis can be carried out by comprehensively monitoring water environment parameters and utilizing an advanced sonar image processing technology, and scientific design of fish nest distribution is realized in combination with behavior pattern analysis of fish schools.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the technical field of ecological restoration, and in particular relates to a fish nest distribution design method based on positioning technology. Background Art

[0002] In the field of aquatic ecology and fishery resource management, the distribution of fish habitats is crucial for their reproduction, foraging and survival. Reasonable design of fish nest distribution is of great significance for protecting aquatic biodiversity, maintaining ecological balance and promoting sustainable development of fisheries. With the advancement of science and technology, the fish nest distribution design method based on positioning technology has gradually become the focus of research and practice.

[0003] Existing technologies usually rely on manual surveys and empirical judgment to determine the location and number of fish nests. This method is time-consuming and labor-intensive, and it is difficult to adapt to the complex and changing underwater environment. In recent years, some automated technologies have begun to be applied to the design of fish nest distribution, such as using sonar monitoring systems for underwater imaging and deploying sensor networks to monitor water environment parameters. These technologies can provide a certain degree of quantitative data, but there are still some limitations.

[0004] The existing technologies have the following defects: 1. Incomplete data collection: The existing technologies often focus on monitoring a single or a few parameters, which makes it difficult to fully reflect the complexity of the underwater environment, resulting in limited understanding of the habitat habits of fish schools. 2. Outdated analysis methods: Traditional sonar image analysis and data processing methods rely on fixed algorithms and thresholds, lack the ability to adapt to complex environmental changes, and affect the accuracy and reliability of fish positioning. 3. Lack of personalization in design: Existing technologies rarely consider the specific needs and behavior patterns of fish schools, resulting in a lack of pertinence in the design of fish nest distribution and an inability to maximize its ecological benefits.

[0005] The present invention aims at the defects of the prior art and provides a method for designing fish nest distribution based on positioning technology. The method realizes the scientific design of fish nest distribution by comprehensively monitoring water environment parameters, using advanced sonar image processing technology to perform fish school positioning analysis, and combining the behavior pattern analysis of fish schools. Summary of the invention

[0006] In view of this, the purpose of the present invention is to propose a fish nest distribution design method based on positioning technology, which can realize the scientific design of fish nest distribution by comprehensively monitoring water environment parameters, using advanced sonar image processing technology to perform fish positioning analysis, and combining with the behavior pattern analysis of fish schools.

[0007] In order to achieve the above object, the present invention provides the following technical solutions:

[0008] The present invention provides a fish nest distribution design method based on positioning technology, comprising:

[0009] Deploy sensors in selected local waters to monitor and record water environment parameters at various locations within the local waters;

[0010] The sonar monitoring system is used to measure the local water area to obtain a sonar image, and the sonar image is analyzed to obtain a positioning analysis result of the fish school;

[0011] Based on a large number of fish positioning analysis results, the activity distribution of the fish school is determined, and the habitat behavior pattern of the fish school is analyzed based on the activity distribution of the fish school, and the habitat distribution of the fish school is judged according to the activity distribution of the fish school and the behavior pattern analysis results;

[0012] Based on the habitat distribution of fish schools, feature extraction is performed on water environment parameters of dense habitat distribution points, and based on the feature extraction results, the feature extraction is expanded to a larger target water area and the location points suitable for fish schools to inhabit are screened to obtain screening results;

[0013] The fish nests are distributed and designed according to the screening results.

[0014] Preferably, sensors are deployed in the selected local waters to monitor and record water environment parameters at various locations in the local waters, including:

[0015] Determine any divided local water area, and arrange combined sensors at various locations in the local water area;

[0016] The combined sensors are used to monitor and record water environment parameters at various locations in the local water area, including water temperature, water pressure, water flow speed and direction, dissolved oxygen content, pH value, salinity and turbidity;

[0017] The position information of each combined sensor is determined, and the collected water environment parameters are associated with the combined sensor and its corresponding position information.

[0018] Preferably, the sonar monitoring system is used to measure the local water area to obtain a sonar image, and the sonar image is analyzed to obtain the positioning analysis result of the fish school, including:

[0019] Use the sonar monitoring system to conduct a comprehensive scan of the selected local waters, obtain sonar images of the underwater environment, and perform image denoising, enhancement and normalization processing on the obtained sonar images;

[0020] Image segmentation technology is used to distinguish fish schools from other underwater objects in the processed sonar images, and the target recognition algorithm is applied to analyze the segmented images to identify the position, size and shape of the fish schools to obtain the positioning analysis results of the fish schools.

[0021] Preferably, using a sonar monitoring system to comprehensively scan the selected local waters to obtain a sonar image of the underwater environment includes:

[0022] Use at least two groups of sonar equipment to conduct a comprehensive alternating scan of the selected local waters from different angles. During the alternating scanning process:

[0023] For the same area, a first group of sonar equipment is first started to scan the local waters to obtain a first scanning result, and a first interval time length of sonar signal feedback is determined according to the first scanning result;

[0024] Then, a second group of sonar equipment is started to scan the local waters to obtain a second scanning result, and a second interval time length of the sonar signal feedback is determined according to the second scanning result;

[0025] Determine the frequency of alternating operation between the first group of sonar devices and the second group of sonar devices based on the first interval duration and the second interval duration;

[0026] The sonar images obtained by alternately scanning the first group of sonar devices and the second group of sonar devices are paired according to time proximity, so as to obtain multiple groups of sonar images of the underwater environment.

[0027] Preferably, the fish school and other underwater objects in the processed sonar image are distinguished by using image segmentation technology, including:

[0028] For any group of sonar images obtained, after performing image denoising, enhancement and normalization processing on the group of sonar images, a grayscale image of the group of sonar images is obtained;

[0029] After the grayscale image is subjected to boundary erosion processing, for any pixel unit with a non-zero grayscale value in the image, the eight adjacent pixel units around the pixel unit are checked, and it is determined that a connected domain is formed between the adjacent pixel unit with a non-zero grayscale value and the pixel unit, thereby dividing the grayscale image into multiple connected domains;

[0030] A second threshold is set and the number of pixel units with non-zero gray values ​​in each connected domain is counted, and the connected domains in the gray image where the number of pixel units with non-zero gray values ​​is less than the second threshold are deleted to obtain multiple reserved areas of the gray image;

[0031] The original sonar image is segmented according to the retained area to obtain the segmentation result.

[0032] Preferably, applying a target recognition algorithm to analyze the segmented image to identify the position, size and shape of the fish school to obtain the positioning analysis results of the fish school includes:

[0033] Each group of sonar images includes a first type of sonar images corresponding to a first group of sonar equipment and a second type of images corresponding to a second group of sonar equipment, each of which corresponds to a first type of segmentation result and a second type of segmentation result;

[0034] The first-category segmentation result and the second-category segmentation result are simultaneously input into the trained fish school recognition neural network to obtain the judgment result of fish school recognition, wherein the fish school recognition neural network is obtained in advance by the following method:

[0035] Build a basic YOLOv5s model, which includes four parts: Input, Backbone, Neck, and Head.

[0036] Among them, the Backbone network uses the CSPDarknet53 structure as the feature extraction end for training samples, which includes Conv convolution module, SPPF module and C3 module;

[0037] The neck, as the network connection part, adopts the feature pyramid network and aggregation network FPN-PAN structure. FPN adopts a top-down structure to generate the prediction feature map by upsampling and fusing the underlying feature information, and PAN adopts a bottom-up structure to supplement the prediction feature map to improve the network's recognition ability of different feature levels.

[0038] The head, as the network output part, generates three prediction layers to predict and regress the target;

[0039] During the training process, a large number of groups of sonar image samples obtained by detecting fish schools are obtained in advance and a plurality of groups of first-class segmentation results and second-class segmentation results are obtained. Training samples and verification samples are generated using the plurality of groups of first-class segmentation results and second-class segmentation results, and the training samples are annotated to form a labeled training data set.

[0040] Use the labeled training data set to train the YOLOv5s model, and continuously optimize the network parameters through the back propagation algorithm until the model reaches the preset recognition accuracy on the verification sample;

[0041] During training, data augmentation techniques including scaling and translation are used to improve the generalization and robustness of the model;

[0042] After training is completed, the model is evaluated and tested until it meets the preset requirements.

[0043] Preferably, the Backbone network uses deformable convolution, which learns offsets through parallel convolution layers to increase the offsets of sampling points. The calculation process of the deformable convolution includes:

[0044] A regular network is used to sample the mapping x on the feature map;

[0045] The convolution and w are used to calculate the weights of the sample points, and the range and expansion degree of the receptive field are specified through the rule network;

[0046] Define a convolution kernel R with a size of 3×3 and an expansion rate of 1:

[0047] R={(―1,―1),(―1,0),…,(0,1),(1,1)}

[0048] Set the variable convolution formula:

[0049]

[0050] Among them, y represents the output feature map, P0 represents each position in y, and P n represents the enumeration of the positions listed in R, {ΔP n │n=1,2,…,N} represents the offset of R.

[0051] The present invention has achieved at least the following beneficial effects:

[0052] 1. It can realize the scientific design of fish nest distribution by comprehensively monitoring water environment parameters, using advanced sonar image processing technology to locate fish schools, and combining with the analysis of fish behavior patterns.

[0053] Other advantages, objectives and features of the present invention will be described in the following description and will be apparent to those skilled in the art to some extent, or those skilled in the art may be taught from the practice of the present invention. The objectives and other advantages of the present invention may be realized and obtained through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to make the purpose, technical solution and beneficial effects of the present invention clearer, the present invention provides the following drawings for illustration:

[0055] Figure 1 A flow chart of the steps of a method for designing fish nest distribution based on positioning technology in an embodiment of the present invention;

[0056] Figure 2 This is a schematic diagram of the structure of the YOLOv5 network in an embodiment of the present invention;

[0057] Figure 3 The figure is a flow chart of variable convolution calculation in an embodiment of the present invention. DETAILED DESCRIPTION

[0058] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0059] The present invention provides a method for designing fish nest distribution based on positioning technology, referring to Figure 1 ,include:

[0060] Deploy sensors in selected local waters to monitor and record water environment parameters at various locations within the local waters;

[0061] The sonar monitoring system is used to measure the local water area to obtain a sonar image, and the sonar image is analyzed to obtain a positioning analysis result of the fish school;

[0062] Based on a large number of fish positioning analysis results, the activity distribution of the fish school is determined, and the habitat behavior pattern of the fish school is analyzed based on the activity distribution of the fish school, and the habitat distribution of the fish school is judged according to the activity distribution of the fish school and the behavior pattern analysis results;

[0063] Based on the habitat distribution of fish schools, feature extraction is performed on water environment parameters of dense habitat distribution points, and based on the feature extraction results, the feature extraction is expanded to a larger target water area and the location points suitable for fish schools to inhabit are screened to obtain screening results;

[0064] The fish nests are distributed and designed according to the screening results.

[0065] The working principle and beneficial effects of the above technical solution are as follows: the technical solution of the present invention can accurately monitor and record water environment parameters and obtain sonar images by deploying sensors and using a sonar monitoring system in local waters. Through in-depth analysis of these data, not only can the fish school be accurately located, but also the activity distribution and habitat behavior patterns of the fish school can be understood. This precise positioning and behavior analysis provides a scientific basis for the reasonable distribution design of fish nests, which helps to improve the attractiveness of fish nests and the survival rate of fish. Using a large number of positioning analysis results, the technical solution can extract the characteristics of the habitat distribution of fish schools and expand it to larger target waters. By extracting and screening the characteristics of water environment parameters, the location points suitable for fish schools to inhabit can be efficiently identified, thereby optimizing the distribution design of fish nests. This method not only improves the efficiency of habitat assessment, but also ensures the scientificity and effectiveness of fish nest layout, and helps to build a more stable and diverse aquatic ecosystem. Based on the screening results, the technical solution of the present invention allows customized fish nest distribution design according to specific water characteristics and fish school needs. This design takes into account the dense habitat points and water environment parameters of fish schools, and can maximize the habitat needs of different fish schools and improve the use efficiency of fish nests. In addition, the customized design also helps to reduce the disturbance to the aquatic environment, protect the natural ecological balance, and provide a more comfortable and safe habitat for fish. By implementing the technical solution of the present invention, the optimization of the fish nest distribution design can be achieved, the quality of the fish habitat can be improved, and the healthy development of the aquatic ecosystem can be promoted.

[0066] In a preferred embodiment, sensors are deployed in selected local waters to monitor and record water environment parameters at various locations in the local waters, including:

[0067] Determine any divided local water area, and arrange combined sensors at various locations in the local water area;

[0068] The combined sensors are used to monitor and record water environment parameters at various locations in the local water area, including water temperature, water pressure, water flow speed and direction, dissolved oxygen content, pH value, salinity and turbidity;

[0069] The position information of each combined sensor is determined, and the collected water environment parameters are associated with the combined sensor and its corresponding position information.

[0070] The working principle and beneficial effects of the above technical solution are as follows: This technical solution can comprehensively monitor and record key water environment parameters, such as water temperature, water pressure, water flow speed and direction, dissolved oxygen content, pH value, salinity and turbidity, by deploying combined sensors in local waters. The combined sensors deployed in local waters have a highly specialized design and can adapt to different underwater environments and conditions. These sensors integrate advanced data acquisition technologies, such as high-precision temperature and pressure sensing elements, and Doppler flowmeters that can accurately measure flow velocity and flow direction. In addition, the sensors are also equipped with optical sensors to measure dissolved oxygen content and turbidity, as well as electrochemical sensors for pH and salinity. This comprehensive monitoring capability provides accurate data support for understanding the water environment and helps to evaluate and optimize the habitat conditions of fish and other aquatic organisms. The location information of each combined sensor is determined by a high-precision global positioning system (GPS) and underwater positioning technology, and the collected water environment parameters are associated with the sensor and its location information. This solution realizes efficient management and analysis of data. This association method not only improves the efficiency of data processing, but also enhances the insight into the trend of water environment changes, making it easier for researchers or managers to make more timely and accurate decisions. Using the detailed water environment data obtained from the sensors, the distribution of fish nests can be precisely designed to ensure that the location of the fish nests can maximize the fish's needs for an ideal habitat. For example, the optimal location of the fish nest can be determined based on the distribution of water temperature and water flow speed, thereby increasing the success rate of fish reproduction and growth. In addition, monitoring of dissolved oxygen content and pH value in water helps to assess water quality and ensure fish health, while salinity and turbidity data can be used to study the impact of environmental factors on fish behavior.

[0071] In a preferred embodiment, the sonar monitoring system is used to measure the local waters to obtain a sonar image, and the sonar image is analyzed to obtain the location analysis result of the fish school, including:

[0072] Use the sonar monitoring system to conduct a comprehensive scan of the selected local waters, obtain sonar images of the underwater environment, and perform image denoising, enhancement and normalization processing on the obtained sonar images;

[0073] Image segmentation technology is used to distinguish fish schools from other underwater objects in the processed sonar images, and the target recognition algorithm is applied to analyze the segmented images to identify the position, size and shape of the fish schools to obtain the positioning analysis results of the fish schools.

[0074] The working principle and beneficial effects of the above technical solution are as follows: by using a sonar monitoring system to conduct a comprehensive scan of the local waters, and performing image denoising, enhancement and normalization processing, the quality of the sonar image can be significantly improved. These preprocessing steps help to eliminate background noise and interference, making the features of the fish and other underwater objects in the image clearer, thereby providing more accurate input data for subsequent image segmentation and target recognition. The image segmentation technology is used to distinguish the fish and other underwater objects in the processed sonar image, and combined with the target recognition algorithm, the position, size and shape of the fish can be accurately identified. This precise positioning and feature recognition is of great significance for understanding the behavior patterns, distribution density and population structure of fish schools, and helps researchers and fishery managers make more scientific decisions.

[0075] In a preferred embodiment, using a sonar monitoring system to comprehensively scan a selected local water area, obtaining a sonar image of an underwater environment includes:

[0076] Use at least two groups of sonar equipment to conduct a comprehensive alternating scan of the selected local waters from different angles. During the alternating scanning process:

[0077] For the same area, a first group of sonar equipment is first started to scan the local waters to obtain a first scanning result, and a first interval time length of sonar signal feedback is determined according to the first scanning result;

[0078] Then, a second group of sonar equipment is started to scan the local waters to obtain a second scanning result, and a second interval time length of the sonar signal feedback is determined according to the second scanning result;

[0079] Determine the frequency of alternating operation between the first group of sonar devices and the second group of sonar devices based on the first interval duration and the second interval duration;

[0080] The sonar images obtained by alternately scanning the first group of sonar devices and the second group of sonar devices are paired according to time proximity, so as to obtain multiple groups of sonar images of the underwater environment.

[0081] The working principle and beneficial effects of the above technical solution are as follows: using two groups of sonar devices to work alternately, scanning from different angles can increase the observation angle of the underwater environment, similar to the principle of stereoscopic vision, which helps to improve the spatial resolution of the target object. At the same time, this multi-angle coverage can increase the monitoring range of local waters, ensure that there is no missed area, and provide technical guarantee for obtaining more comprehensive underwater fish information. Since each group of sonar works at different time intervals, the probability of signal overlap and mutual interference is reduced, thereby ensuring the clarity and accuracy of the sonar signal. The multiple groups of sonar images obtained by alternating scanning can be analyzed in time series. By comparing the images at different time points, the dynamic changes of underwater objects, such as the movement and behavior patterns of fish, can be more accurately captured. In addition, the comparison and analysis of multiple groups of data can improve the reliability of the results, eliminate accidental errors, and provide a solid data basis for the accurate assessment of the underwater environment. The technical solution of the present invention adopts the technical solution of two groups of sonar devices for alternating scanning, which provides an efficient and accurate solution for the detection and identification of underwater fish by reducing signal interference, improving spatial resolution, enhancing data accuracy and optimizing resource utilization.

[0082] In a preferred embodiment, the fish school and other underwater objects in the processed sonar image are distinguished by image segmentation technology, including:

[0083] For any group of sonar images obtained, after performing image denoising, enhancement and normalization processing on the group of sonar images, a grayscale image of the group of sonar images is obtained;

[0084] After the grayscale image is subjected to boundary erosion processing, for any pixel unit with a non-zero grayscale value in the image, the eight adjacent pixel units around the pixel unit are checked, and it is determined that a connected domain is formed between the adjacent pixel unit with a non-zero grayscale value and the pixel unit, thereby dividing the grayscale image into multiple connected domains;

[0085] A second threshold is set and the number of pixel units with non-zero gray values ​​in each connected domain is counted, and the connected domains in the gray image where the number of pixel units with non-zero gray values ​​is less than the second threshold are deleted to obtain multiple reserved areas of the gray image;

[0086] The original sonar image is segmented according to the retained area to obtain the segmentation result.

[0087] The working principle and beneficial effects of the above technical solution are as follows: through image denoising, enhancement and normalization processing, the grayscale image of the sonar image is optimized, the background noise is reduced, the useful information in the image is enhanced, and the characteristics of the fish school and other underwater objects are more obvious. This provides a clearer image basis for subsequent image segmentation, thereby improving the accuracy of segmentation. Boundary erosion processing helps to eliminate small objects or noise in the grayscale image, while connected domain analysis ensures that large objects or groups of objects in the image are retained. By setting a second threshold and counting the number of pixel units with non-zero grayscale values ​​in the connected domain, the main objects in the image, such as fish schools, can be effectively identified and separated from other smaller or unimportant suspended debris in the water. By deleting the connected domain where the number of pixel units with non-zero grayscale values ​​is less than the second threshold, those unimportant or noise-formed areas can be removed, thereby reducing unnecessary computational burdens in subsequent processing steps. This not only optimizes the segmentation results, but also improves the efficiency of the algorithm, especially when processing large-scale or high-resolution sonar images. The segmentation results obtained through the above processing steps can more accurately reflect the real situation of the underwater environment and provide a reliable data basis for subsequent analysis and applications, such as fish population assessment, behavior analysis or underwater structure identification.

[0088] In a preferred embodiment, referring to Figure 2 , the target recognition algorithm is used to analyze the segmented image, identify the position, size and shape of the fish school, and obtain the positioning analysis results of the fish school, including:

[0089] Each group of sonar images includes a first type of sonar images corresponding to a first group of sonar equipment and a second type of images corresponding to a second group of sonar equipment, each of which corresponds to a first type of segmentation result and a second type of segmentation result;

[0090] The first-category segmentation result and the second-category segmentation result are simultaneously input into the trained fish school recognition neural network to obtain the judgment result of fish school recognition, wherein the fish school recognition neural network is obtained in advance by the following method:

[0091] Build a basic YOLOv5s model, which includes four parts: Input, Backbone, Neck, and Head.

[0092] Among them, the Backbone network uses the CSPDarknet53 structure as the feature extraction end for training samples, which includes Conv convolution module, SPPF module and C3 module;

[0093] The neck, as the network connection part, adopts the feature pyramid network and aggregation network FPN-PAN structure. FPN adopts a top-down structure to generate the prediction feature map by upsampling and fusing the underlying feature information, and PAN adopts a bottom-up structure to supplement the prediction feature map to improve the network's recognition ability of different feature levels.

[0094] The head, as the network output part, generates three prediction layers to predict and regress the target;

[0095] During the training process, a large number of groups of sonar image samples obtained by detecting fish schools are obtained in advance and a plurality of groups of first-class segmentation results and second-class segmentation results are obtained. Training samples and verification samples are generated using the plurality of groups of first-class segmentation results and second-class segmentation results, and the training samples are annotated to form a labeled training data set.

[0096] Use the labeled training data set to train the YOLOv5s model, and continuously optimize the network parameters through the back propagation algorithm until the model reaches the preset recognition accuracy on the verification sample;

[0097] During training, data augmentation techniques including scaling and translation are used to improve the generalization and robustness of the model;

[0098] After training is completed, the model is evaluated and tested until it meets the preset requirements.

[0099] The working principle and beneficial effects of the above technical solution are as follows: by applying the target recognition algorithm, especially the well-trained YOLOv5s neural network model, the technical solution can accurately identify and analyze the fish schools in the segmented sonar images. This deep learning-based method can quickly process image data while providing highly accurate recognition results, including the location, size and morphology of the fish schools. The design of the YOLOv5s model, especially its Backbone network and Neck structure, enhances the ability to extract image features, thereby improving the performance of target detection. The use of data enhancement techniques such as scaling and translation during the training process helps the model learn to identify fish schools from different angles and conditions, thereby improving the generalization ability and robustness of the model in practical applications. This approach ensures that the model does not overfit the training data, but is able to adapt to various underwater environments and changes in sonar images, and maintains a stable recognition effect even in the face of complex or undesirable sonar images. The technical solution reduces the workload and subjective errors of manual analysis of sonar images through an automated target recognition process. The automated neural network model can continuously and efficiently analyze large amounts of sonar image data, providing real-time and accurate information for underwater fish monitoring. In addition, by continuously optimizing and evaluating the model, it can ensure that the monitoring results meet the preset accuracy requirements.

[0100] In a preferred embodiment, referring to Figure 3 , the Backbone network uses deformable convolution, which learns offset offsets through parallel convolution layers to increase the offset of sampling points. The calculation process of deformable convolution includes:

[0101] A regular network is used to sample the mapping x on the feature map;

[0102] The convolution and w are used to calculate the weights of the sample points, and the range and expansion degree of the receptive field are specified through the rule network;

[0103] Define a convolution kernel R with a size of 3×3 and an expansion rate of 1:

[0104] R={(―1,―1),(―1,0),…,(0,1),(1,1)}

[0105] Set the variable convolution formula:

[0106]

[0107] Among them, y represents the output feature map, P0 represents each position in y, and P n represents the enumeration of the positions listed in R, {ΔP n │n=1,2,…,N} represents the offset of R.

[0108] The working principle and beneficial effects of the above technical solution are as follows: by adopting deformable convolution, fish schools in different forms can be better identified and multi-scale feature extraction of image models can be realized. It is worth noting that variable convolution introduces an offset for each position based on traditional convolution. The offset is generated by the input feature map and another convolution, so the value of x needs to be calculated by bilinear interpolation.

[0109] Finally, it should be noted that the above preferred embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail through the above preferred embodiments, those skilled in the art should understand that various changes can be made in form and details without departing from the scope defined by the claims of the present invention.

Claims

1. A fish nest distribution design method based on positioning technology, characterized in that: include: Deploy sensors in selected local waters to monitor and record water environment parameters at various locations within the local waters; The sonar monitoring system is used to measure the local water area to obtain a sonar image, and the sonar image is analyzed to obtain a positioning analysis result of the fish school; Based on a large number of fish positioning analysis results, the activity distribution of the fish school is determined, and the habitat behavior pattern of the fish school is analyzed based on the activity distribution of the fish school, and the habitat distribution of the fish school is judged according to the activity distribution of the fish school and the behavior pattern analysis results; Based on the habitat distribution of fish schools, feature extraction is performed on water environment parameters of dense habitat distribution points, and based on the feature extraction results, the feature extraction is expanded to a larger target water area and the location points suitable for fish schools to inhabit are screened to obtain screening results; The fish nests are distributed and designed according to the screening results.

2. A fish nest distribution design method based on positioning technology according to claim 1, characterized in that: The deployment of sensors in the selected local waters to monitor and record water environment parameters at various locations in the local waters includes: Determine any divided local water area, and arrange combined sensors at various locations in the local water area; The combined sensors are used to monitor and record water environment parameters at various locations in the local water area, including water temperature, water pressure, water flow speed and direction, dissolved oxygen content, pH value, salinity and turbidity; The position information of each combined sensor is determined, and the collected water environment parameters are associated with the combined sensor and its corresponding position information.

3. The method for designing fish nest distribution based on positioning technology according to claim 1, characterized in that: The method of using a sonar monitoring system to measure the local waters to obtain a sonar image, and analyzing the sonar image to obtain a positioning analysis result of the fish school includes: Use the sonar monitoring system to conduct a comprehensive scan of the selected local waters, obtain sonar images of the underwater environment, and perform image denoising, enhancement and normalization processing on the obtained sonar images; Image segmentation technology is used to distinguish fish schools from other underwater objects in the processed sonar images, and the target recognition algorithm is applied to analyze the segmented images to identify the position, size and shape of the fish schools to obtain the positioning analysis results of the fish schools.

4. The method for designing fish nest distribution based on positioning technology according to claim 3, characterized in that: The method of using the sonar monitoring system to comprehensively scan the selected local waters and obtain the sonar image of the underwater environment includes: Use at least two groups of sonar equipment to conduct a comprehensive alternating scan of the selected local waters from different angles. During the alternating scanning process: For the same area, a first group of sonar equipment is first started to scan the local waters to obtain a first scanning result, and a first interval time length of sonar signal feedback is determined according to the first scanning result; Then, a second group of sonar equipment is started to scan the local waters to obtain a second scanning result, and a second interval time length of the sonar signal feedback is determined according to the second scanning result; Determine the frequency of alternating operation between the first group of sonar devices and the second group of sonar devices based on the first interval duration and the second interval duration; The sonar images obtained by alternately scanning the first group of sonar devices and the second group of sonar devices are paired according to time proximity, so as to obtain multiple groups of sonar images of the underwater environment.

5. The method for designing fish nest distribution based on positioning technology according to claim 4, characterized in that: The method of using image segmentation technology to distinguish the fish school and other underwater objects in the processed sonar image includes: For any group of sonar images obtained, after performing image denoising, enhancement and normalization processing on the group of sonar images, a grayscale image of the group of sonar images is obtained; After the grayscale image is subjected to boundary erosion processing, for any pixel unit with a non-zero grayscale value in the image, the eight adjacent pixel units around the pixel unit are checked, and it is determined that a connected domain is formed between the adjacent pixel unit with a non-zero grayscale value and the pixel unit, thereby dividing the grayscale image into multiple connected domains; A second threshold is set and the number of pixel units with non-zero gray values ​​in each connected domain is counted, and the connected domains in the gray image where the number of pixel units with non-zero gray values ​​is less than the second threshold are deleted to obtain multiple reserved areas of the gray image; The original sonar image is segmented according to the retained area to obtain the segmentation result.

6. The method for designing fish nest distribution based on positioning technology according to claim 5, characterized in that: The target recognition algorithm is used to analyze the segmented image to identify the position, size and shape of the fish school to obtain the positioning analysis results of the fish school, including: Each group of sonar images includes a first type of sonar images corresponding to a first group of sonar equipment and a second type of images corresponding to a second group of sonar equipment, each of which corresponds to a first type of segmentation result and a second type of segmentation result; The first-category segmentation result and the second-category segmentation result are simultaneously input into the trained fish school recognition neural network to obtain the judgment result of fish school recognition, wherein the fish school recognition neural network is obtained in advance by the following method: Build a basic YOLOv5s model, which includes four parts: Input, Backbone, Neck, and Head. Among them, the Backbone network uses the CSPDarknet53 structure as the feature extraction end for training samples, which includes Conv convolution module, SPPF module and C3 module; The neck, as the network connection part, adopts the feature pyramid network and aggregation network FPN-PAN structure. FPN adopts a top-down structure to generate the prediction feature map by upsampling and fusing the underlying feature information, and PAN adopts a bottom-up structure to supplement the prediction feature map to improve the network's recognition ability of different feature levels. The head, as the network output part, generates three prediction layers to predict and regress the target; During the training process, a large number of groups of sonar image samples obtained by detecting fish schools are obtained in advance and a plurality of groups of first-class segmentation results and second-class segmentation results are obtained. Training samples and verification samples are generated using the plurality of groups of first-class segmentation results and second-class segmentation results, and the training samples are annotated to form a labeled training data set. Use the labeled training data set to train the YOLOv5s model, and continuously optimize the network parameters through the back propagation algorithm until the model reaches the preset recognition accuracy on the verification sample; During training, data augmentation techniques including scaling and translation are used to improve the generalization and robustness of the model; After training is completed, the model is evaluated and tested until it meets the preset requirements.

7. The method for designing fish nest distribution based on positioning technology according to claim 6, characterized in that: The Backbone network uses deformable convolution, which learns offsets through parallel convolutional layers to increase the offset of the sampling points. The calculation process of the deformable convolution includes: A regular network is used to sample the mapping x on the feature map; Convolution and w are used to calculate the weights of sample points, and the range and expansion degree of the receptive field are specified through the rule network; Define a convolution kernel R with a size of 3×3 and an expansion rate of 1: R={(―1,―1),(―1,0),…,(0,1),(1,1)} Set the variable convolution formula: Among them, y represents the output feature map, P0 represents each position in y, and P n represents the enumeration of the positions listed in R, {ΔP n │n=1,2,…,N} represents the offset of R.

Citation Information

Cited By

  • Ecological bank protection intelligent irrigation control system and method based on multi-source data fusion

    CN121366055A