Fishing gear damage guiding and tracing method based on satellite remote sensing

Abstract

The invention discloses a fishing gear damage guiding traceability method based on satellite remote sensing, which comprises the following steps of: firstly, acquiring a remote sensing image of a water area in a historical time period by utilizing a satellite remote sensing technology, and identifying and positioning a preset fishing gear range; the method comprises the following steps: searching an adjacent camera device through a water area model to obtain a fishing gear image, extracting features by adopting a PCA analysis method, analyzing feature changes, screening out an analysis period in which the feature changes exceed an expected range, and extracting a damaged image feature set; a variational auto-encoder generation model is constructed, and analog data is generated through encoder and decoder learning training. Setting a first analysis period to obtain fishing gear image data, importing the generation model to generate damage traceability features, and using the damage traceability features as training data to optimize the fishing gear identification model. And in the second analysis period, performing identification and traceability positioning on the real-time fishing gear image by using the trained model. According to the invention, identification analysis and positioning traceability of the fishing gear can be realized, and efficient and accurate fishing gear identification and guiding positioning traceability can be realized.

Description

Technical Field

[0001] The present invention relates to the field of remote sensing data and fishery analysis, and more specifically, to a method for tracing the source of damaged fishing gear based on satellite remote sensing. Background Art

[0002] In fishery supervision and fishing gear identification, the traditional method is generally ground patrol. However, due to problems such as high cost and limited coverage area, it is difficult to meet the needs of modern fishery management. In recent years, satellite remote sensing technology has received extensive attention because it can provide large-area and high-frequency spatial observations. In the use of general fishing gear, it is easily affected by natural factors or human damage, resulting in deformation or damage of fishing gear equipment, which has a certain destructive effect on the ecological environment. And still, corresponding identification and supervision are required. However, for damaged fishing gear, especially for special fishing gear set by humans, it is difficult to carry out accurate positioning and traceability analysis, resulting in difficulties in discovering and warning fishing gear. Therefore, it is of great significance to develop a set of efficient and reliable systems to identify and locate such fishing gear. Summary of the Invention

[0003] The present invention overcomes the defects of the prior art and proposes a method for tracing the source of damaged fishing gear based on satellite remote sensing.

[0004] The first aspect of the present invention provides a method for tracing the source of damaged fishing gear based on satellite remote sensing, including:

[0005] Construct a three-dimensional water area model based on preset water area spatial information;

[0006] In a historical time period, obtain remote sensing image data of a preset water area through satellite remote sensing technology, identify general fishing gear through the remote sensing image data, and locate the fishing gear search range;

[0007] Through the fishing gear search range, search for adjacent camera devices in the water area model, and obtain multi-angle fishing gear image data;

[0008] Import the fishing gear image data into the fishing gear identification model, extract features from the fishing gear image data through PCA analysis method to form a fishing gear feature data set. Based on multiple analysis periods, perform feature change analysis on the fishing gear feature data set. The feature change is based on the difference analysis of feature vectors. Through a preset difference degree, screen out the analysis periods in which the feature change exceeds the expected range, mark them as change periods, and extract the data corresponding to the change periods from the fishing gear feature data to obtain a damaged image feature set;

[0009] Construct a generative model based on a variational autoencoder. The generative model includes an encoder and a decoder. Import the damaged image feature set into the encoder for feature learning. The encoder performs low-dimensional mapping on the imported data to form a latent variable z. The decoder maps the latent variable z to the original image space and obtains simulated data. The variational autoencoder is trained by minimizing the sum of the reconstruction loss and the KL divergence until the simulated data generated by the model meets the expected standard;

[0010] Set the first analysis period. In the first analysis period, obtain fishing gear image data and label it as the first image data. Import the first image data into the generative model for feature learning and simulated feature generation, and label the obtained simulated data as damaged traceability features;

[0011] Use the damaged traceability features as training data to import into the fishing gear recognition model for model training. In the second analysis period, based on the trained fishing gear recognition model, perform fishing gear recognition and traceability positioning on the fishing gear image data collected in real time.

[0012] In this solution, based on the preset water area space information, construct a three-dimensional water area model, specifically:

[0013] Obtain the preset water area space information, and the space information includes the area of the preset water area, the map contour, and the fishery resource distribution information;

[0014] Construct a three-dimensional visualized water area model through the space information, and perform dynamic monitoring of fishing boats and fishing gear through the water area model.

[0015] In this solution, within a historical time period, obtain the remote sensing image data of the preset water area through satellite remote sensing technology, and perform general fishing gear recognition through the remote sensing image data and locate the fishing gear search range, specifically:

[0016] Set a historical time period, and within a historical time period, obtain the remote sensing image data of the preset water area;

[0017] Perform image enhancement and standardization preprocessing on the remote sensing image data, and extract the analysis contour features of the image objects based on the Sobel operator;

[0018] Retrieve general fishing gear images from the image big data and obtain comparison fishing gear images;

[0019] Extract the contour features based on the Sobel operator through the comparison fishing gear images and compare the contour features;

[0020] Compare the feature similarity between the analysis contour features and the comparison contour features, identify the corresponding fishing gear, and locate and mark the fishing gear search range from the remote sensing image data.

[0021] In this solution, the method of searching for adjacent camera devices in the water area model based on the fishing gear search range and obtaining fishing gear image data from multiple angles is as follows:

[0022] Search for camera devices within a preset range in the water area model based on the fishing gear search range and mark them as adjacent camera devices;

[0023] Obtain fishing gear image data from multiple angles through the adjacent camera devices.

[0024] In this solution, the method of importing the fishing gear image data into the fishing gear recognition model, extracting features from the fishing gear image data through PCA analysis method to form a fishing gear feature data set, analyzing the feature changes of the fishing gear feature data set based on multiple analysis periods, analyzing the differences in features based on feature vectors, and screening out the analysis periods with feature changes exceeding the expected range through a preset difference degree, marking them as change periods, and extracting the data corresponding to the change periods from the fishing gear feature data to obtain a damaged image feature set is as follows:

[0025] Normalize the fishing gear image data and import it into the fishing gear recognition model;

[0026] Based on the PCA analysis method, extract the principal component features of the fishing gear image data to form a fishing gear feature data set;

[0027] Divide a historical time period into multiple analysis periods, divide the fishing gear feature data set into corresponding analysis period data to form multiple unit feature sets;

[0028] Convert each unit feature set into a feature vector representation to obtain a feature vector set;

[0029] Select the first feature vector set and the second feature vector set, set a vector space, calculate the first center point and the second center point in the vector space based on the first feature vector set and the second feature vector set respectively, calculate the distance between the first center point and the second center point based on the Manhattan distance, and use the calculation result as the data difference degree;

[0030] The first feature vector set and the second feature vector set are two consecutive analysis period corresponding vector sets randomly selected;

[0031] If the data difference degree is greater than the preset difference degree, mark the corresponding two consecutive analysis periods as change periods;

[0032] Discriminate all analysis periods and mark the change periods.

[0033] In this solution, a generative model based on a variational autoencoder is constructed. The generative model includes an encoder and a decoder. The damaged image feature set is imported into the encoder for feature learning. The encoder performs low-dimensional mapping on the imported data to form a latent variable z. The decoder maps the latent variable z to the original image space and obtains simulated data. By minimizing the sum of the reconstruction loss and the KL divergence, the variational autoencoder is trained until the simulated data generated by the model meets the expected standard. Specifically:

[0034] Initialize the generative model and set multiple convolutional layers for the encoder and the decoder respectively;

[0035] Clean and standardize the damaged image feature set;

[0036] Perform low-dimensional mapping on the damaged image feature set through the encoder, and output the latent variable z and the variable distribution;

[0037] The decoder maps the latent variable z to the original image space to obtain the output data, which is marked as simulated data;

[0038] Based on the preset reconstruction loss, judge the difference between the simulated data and the imported data, and based on the KL divergence, judge the difference between the variable distribution and the preset distribution. By minimizing the sum of the reconstruction loss and the KL divergence, the generative model is trained iteratively until the sum of the reconstruction loss and the KL divergence is within the preset range.

[0039] In this solution, a first analysis period is set. In the first analysis period, fishing gear image data is acquired and marked as the first image data. The first image data is imported into the generative model for feature learning and simulated feature generation. The obtained simulated data is marked as the damaged traceability feature. The damaged traceability feature is used as training data to import into the fishing gear recognition model for model training. Specifically:

[0040] Set the first analysis period, and in the first analysis period, acquire fishing gear image data in real time and mark it as the first image data;

[0041] Import the first image data into the generative model for feature learning and simulated feature generation, generate a preset amount of simulated data, and mark it as the damaged traceability feature;

[0042] Use the damaged traceability feature as a training set to import into the fishing gear recognition model for model training.

[0043] In the second aspect of the present invention, a fishing gear damage-oriented traceability system based on satellite remote sensing is also provided. The system includes: a memory and a processor. The memory includes a fishing gear damage-oriented traceability program based on satellite remote sensing. When the fishing gear damage-oriented traceability program based on satellite remote sensing is executed by the processor, the following steps are implemented:

[0044] Construct a three-dimensional water area model based on preset water area spatial information;

[0045] Within a historical time period, obtain remote sensing image data of the preset water area through satellite remote sensing technology, identify general fishing gear through the remote sensing image data, and locate the fishing gear search range;

[0046] Search for adjacent camera devices in the water area model through the fishing gear search range, and obtain multi-angle fishing gear image data;

[0047] Import the fishing gear image data into the fishing gear recognition model, extract features from the fishing gear image data through PCA analysis method to form a fishing gear feature data set. Based on multiple analysis cycles, perform feature change analysis on the fishing gear feature data set. The feature change is based on the difference analysis of feature vectors. Through a preset difference degree, screen out the analysis cycles in which the feature change exceeds the expected range, mark them as change cycles, and extract the data corresponding to the change cycles from the fishing gear feature data to obtain a damaged image feature set;

[0048] Construct a generative model based on a variational autoencoder. The generative model includes an encoder and a decoder. Import the damaged image feature set into the encoder for feature learning. Through the encoder, perform low-dimensional mapping on the imported data to form a latent variable z. Through the decoder, map the latent variable z to the original image space and obtain simulated data. Through minimizing the sum of the reconstruction loss and the KL divergence, perform learning and training on the variational autoencoder until the simulated data generated by the model reaches the expected standard;

[0049] Set a first analysis cycle. In the first analysis cycle, obtain fishing gear image data and mark it as first image data. Import the first image data into the generative model for feature learning and simulated feature generation, and mark the obtained simulated data as damaged traceability features;

[0050] Use the damaged traceability features as training data to import into the fishing gear recognition model for model training. In the second analysis cycle, perform fishing gear recognition and traceability positioning on the fishing gear image data collected in real time based on the trained fishing gear recognition model.

[0051] The third aspect of the present invention also provides a computer-readable storage medium, which includes a fishing gear damage-oriented traceability program based on satellite remote sensing. When the fishing gear damage-oriented traceability program based on satellite remote sensing is executed by a processor, it realizes the steps of the fishing gear damage-oriented traceability method according to any one of the above.

[0052] The present invention discloses a method for guiding and tracing the damage of fishing gear based on satellite remote sensing. First, satellite remote sensing technology is used to obtain remote sensing images of waters within a historical time period, and the preset fishing gear range is identified and located. Fishing gear images are obtained by searching for adjacent camera devices through a water area model. PCA analysis method is used to extract features, analyze the feature changes, screen out the analysis periods in which the feature changes exceed the expected range, and extract the damaged image feature set. A variational autoencoder generation model is constructed, and through the learning and training of the encoder and decoder, simulated data is generated. The fishing gear image data is obtained in the first analysis period, imported into the generation model to generate damage tracing features, and used as training data to optimize the fishing gear recognition model. In the second analysis period, the trained model is used to identify and trace the location of real-time fishing gear images. Through the present invention, the identification analysis and location tracing of fishing gear can be realized, and efficient and accurate fishing gear identification and guiding location tracing can be achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 FIG. shows a flowchart of a method for guiding and tracing the damage of fishing gear based on satellite remote sensing according to the present invention;

[0054] Figure 2 FIG. shows a block diagram of a system for guiding and tracing the damage of fishing gear based on satellite remote sensing according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0055] In order to more clearly understand the above objects, features and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, without conflict, the embodiments of the present application and the features in the embodiments can be combined with each other.

[0056] In the following description, many specific details are set forth in order to fully understand the present invention. However, the present invention can also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited by the specific embodiments disclosed below.

[0057] Figure 1 FIG. shows a flowchart of a method for guiding and tracing the damage of fishing gear based on satellite remote sensing according to the present invention.

[0058] As Figure 1 shown, in the first aspect of the present invention, a method for guiding and tracing the damage of fishing gear based on satellite remote sensing is provided, including:

[0059] S102. Based on the preset water area spatial information, a three-dimensional water area model is constructed;

[0060] S104. Within a historical time period, remote sensing image data of a preset water area is obtained through satellite remote sensing technology, general fishing gear is identified through the remote sensing image data, and the fishing gear search range is located;

[0061] S106, search for adjacent camera devices in the water area model based on the fishing gear search range, and obtain fishing gear image data from multiple angles;

[0062] S108, import the fishing gear image data into the fishing gear recognition model, extract features from the fishing gear image data through PCA analysis to form a fishing gear feature data set, perform feature change analysis on the fishing gear feature data set based on multiple analysis periods, perform difference analysis on the feature changes based on feature vectors, and filter out the analysis periods in which the feature changes exceed the expected range through a preset difference degree, mark them as change periods, and extract the data corresponding to the change periods from the fishing gear feature data to obtain a damaged image feature set;

[0063] S110, construct a generative model based on a variational autoencoder. The generative model includes an encoder and a decoder. Import the damaged image feature set into the encoder for feature learning. Perform low-dimensional mapping on the imported data through the encoder to form a latent variable z. Map the latent variable z to the original image space through the decoder and obtain simulated data. Learn and train the variational autoencoder by minimizing the sum of the reconstruction loss and the KL divergence until the simulated data generated by the model reaches the expected standard;

[0064] S112, set the first analysis period, obtain fishing gear image data in the first analysis period and mark it as the first image data, import the first image data into the generative model for feature learning and simulated feature generation, and mark the obtained simulated data as damaged traceability features;

[0065] S114, import the damaged traceability features as training data into the fishing gear recognition model for model training. In the second analysis period, perform fishing gear recognition and traceability positioning on the fishing gear image data collected in real time based on the trained fishing gear recognition model.

[0066] According to the embodiments of the present invention, constructing a three-dimensional water area model based on preset water area spatial information specifically includes:

[0067] Obtain preset water area spatial information, where the spatial information includes the area of the preset water area, the map contour, and the fishery resource distribution information;

[0068] Construct a three-dimensional visualization-based water area model through the spatial information, and perform dynamic monitoring of fishing boats and fishing gear through the water area model.

[0069] It should be noted that the water area model is a visualization model, which monitors the water area through satellite remote sensing and camera devices, and identifies and visualizes the positions of corresponding fishing boats and fishing gear.

[0070] In the identification of general fishing gear, the fishing gear is general fishing gear, including net fishing gear, fish stakes, fish baskets, etc. Especially for some fishing gear set artificially, there may be situations of illegal use. When the fishing gear is in a damaged state, it is difficult to conduct identification and analysis. Through the embodiments of the present invention, effective traceability analysis can be carried out, which helps to conduct early warning analysis or positioning analysis on relevant fishing gear.

[0071] According to the embodiments of the present invention, within a historical time period, remote sensing image data of a preset water area is obtained through satellite remote sensing technology, and general fishing gear is identified through the remote sensing image data, and the fishing gear search range is located. Specifically:

[0072] Set a historical time period, and within a historical time period, obtain remote sensing image data of a preset water area;

[0073] Perform image enhancement and standardization preprocessing on the remote sensing image data, and extract the analysis contour features of the image objects based on the Sobel operator;

[0074] Retrieve general fishing gear images from the image big data and obtain comparison fishing gear images;

[0075] Extract contour features based on the Sobel operator through the comparison fishing gear images and compare the contour features;

[0076] Compare the analysis contour features with the comparison contour features, identify the corresponding fishing gear, and mark the fishing gear search range in the remote sensing image data with a map.

[0077] It should be noted that the image big data is an identification image storing remote sensing images corresponding to various fishing gears, which is used for comparative analysis. Net fishing gear or other artificially set fishing gear devices are a special type of fishing gear, whose shape and size are specific, and there are different shapes due to the influence of artificial layout. In traditional fishing gear identification, it is difficult to effectively identify the corresponding fishing gear. Satellite remote sensing is a high-definition image remote sensing technology that can be widely applied to fisheries. In the embodiments of the present invention, through remote sensing technology, image recognition of water areas is carried out from a large range perspective, extensive feature search and identification positioning of specific fishing gear are carried out, and based on the positioning range, further camera image recognition is carried out to achieve efficient identification of specific fishing gear.

[0078] According to the embodiments of the present invention, through the fishing gear search range, adjacent camera devices are searched in the water area model, and multi-angle fishing gear image data is obtained. Specifically:

[0079] Based on the fishing gear search range, search for camera devices within a preset range in the water area model and mark them as adjacent camera devices;

[0080] Obtain multi-angle fishing gear image data through the adjacent camera devices.

[0081] It should be noted that the imaging device is a high-definition imaging device and is disposed in a preset water area.

[0082] According to an embodiment of the present invention, the fishing gear image data is imported into a fishing gear recognition model, and through PCA analysis, feature extraction is performed on the fishing gear image data to form a fishing gear feature data set. Based on multiple analysis periods, feature change analysis is performed on the fishing gear feature data set. The feature change is based on difference analysis of feature vectors. Through a preset difference degree, the analysis periods in which the feature change exceeds the expected range are screened out and marked as change periods, and the data corresponding to the change periods is extracted from the fishing gear feature data to obtain a damaged image feature set. Specifically:

[0083] The fishing gear image data is subjected to image standardization and imported into the fishing gear recognition model;

[0084] Based on PCA analysis, principal component feature extraction is performed on the fishing gear image data to form a fishing gear feature data set;

[0085] A historical time period is divided into multiple analysis periods, and corresponding analysis period data division is performed on the fishing gear feature data set to form multiple unit feature sets;

[0086] Each unit feature set is converted into a feature vector representation to obtain a feature vector set;

[0087] A first feature vector set and a second feature vector set are selected, a vector space is set, and a first center point and a second center point are respectively calculated in the vector space based on the first feature vector set and the second feature vector set. Based on the Manhattan distance, the distance between the first center point and the second center point is calculated, and the calculation result is used as the data difference degree;

[0088] The first feature vector set and the second feature vector set are two consecutive analysis period corresponding vector sets randomly selected;

[0089] If the data difference degree is greater than the preset difference degree, the corresponding two consecutive analysis periods are marked as change periods;

[0090] All analysis periods are discriminated and the change periods are marked.

[0091] It should be noted that the fishing gear recognition model is specifically an image recognition model based on deep learning, and the recognition rate is improved based on the training process. The first center point is specifically the one with the shortest sum of distances from each vector in the first feature vector set, and the second center point is specifically the one with the shortest sum of distances from each vector in the second feature vector set. The data difference degree is a distance value, and the larger this value is, the greater the data difference.

[0092] It is worth mentioning that the net-shaped or artificially set fishing gear is a special type of fishing gear. During the monitoring of fishing in waters, it needs to be supervised and used. Therefore, efficient fishing gear identification and positioning analysis are required. For artificially set fishing gear devices, their use process is easily affected by water flow and may be damaged. For fishing gear with damage, it is generally difficult to identify and locate, and it is also impossible to conduct fishing gear traceability analysis. Based on this, the present invention collects fishing gear characteristics, performs vectorization and change analysis based on multiple cycles, combines and mines change characteristics. By the distance change of the feature set in the vector space, the corresponding change cycle is screened out. The corresponding change cycle is the corresponding cycle when the fishing gear has damage changes in the image features. Through this process, the corresponding damage features are marked and collected. Subsequently, a variational autoencoder is introduced to learn the damage features and generate simulated features to perform damage simulation on real-time data and generate training data to improve the recognition accuracy, and further realize fishing gear positioning traceability and identification analysis, achieving efficient and accurate fishing gear identification and guiding positioning traceability.

[0093] Based on the Sobel operator, in the analysis of the contour features of the image object, specifically, the remote sensing image data is used to identify the object part of the image, and the corresponding object image part is converted to gray scale. After conversion, the image is applied to a 3x3 matrix to perform edge detection in the horizontal direction (X-axis) and vertical direction (Y-axis) respectively, and the gradient of each pixel is calculated. Further edge tracking is carried out to further form contour information, and all contour information is extracted to obtain the analysis contour features; in addition, the same Sobel operator method is adopted in the comparison of fishing gear images to extract contour features.

[0094] Based on the PCA analysis method, the principal component features of the fishing gear image data are extracted to form a fishing gear feature dataset. Specifically, the fishing gear image data needs to be preprocessed, including steps such as resizing, grayscale conversion, and normalization, to ensure that all images have the same size and format. All preprocessed images are flattened into vectors, and a large matrix is constructed, where each row represents the data of one image. Then, the covariance matrix is calculated based on this matrix. The eigenvalue decomposition is performed on the covariance matrix to obtain the eigenvalues and the corresponding eigenvectors. Further, the principal components are selected. According to the eigenvalue sorting result, the first K largest eigenvalues and their corresponding eigenvectors are selected as the principal components, where k is the preset target dimension number. Finally, the original image data is projected onto the selected principal components to form a feature dataset. The main purpose of this process is to screen the main feature data and reduce redundant features.

[0095] According to an embodiment of the present invention, the variational autoencoder-based generation model is constructed. The generation model includes an encoder and a decoder. The damaged image feature set is imported into the encoder for feature learning. The encoder performs low-dimensional mapping on the imported data to form a latent variable z. The decoder maps the latent variable z to the original image space and obtains simulated data. The variational autoencoder is trained by minimizing the sum of the reconstruction loss and the KL divergence until the simulated data generated by the model reaches the expected standard. Specifically:

[0096] Initialize the generation model, and set multiple convolutional layers for the encoder and the decoder respectively;

[0097] Perform data cleaning and standardization processing on the damaged image feature set;

[0098] Perform low-dimensional mapping on the damaged image feature set through the encoder, and output the latent variable z and the variable distribution;

[0099] The decoder maps the latent variable z to the original image space to obtain the output data, which is marked as simulated data;

[0100] Based on the preset reconstruction loss, distinguish the difference between the simulated data and the imported data, and based on the KL divergence, distinguish the difference between the variable distribution and the preset distribution. By minimizing the sum of the reconstruction loss and the KL divergence, the generation model is trained iteratively until the sum of the reconstruction loss and the KL divergence is within the preset range.

[0101] It should be noted that the decoder is a multi-layer transposed convolution. The preset distribution can be compared using a Gaussian distribution. The preset reconstruction loss can be calculated based on the mean square error or cross entropy.

[0102] According to an embodiment of the present invention, the first analysis period is set. In the first analysis period, fishing gear image data is acquired and marked as the first image data. The first image data is imported into the generation model for feature learning and simulated feature generation. The obtained simulated data is marked as the damaged traceability feature. The damaged traceability feature is used as training data to import into the fishing gear recognition model for model training. Specifically:

[0103] Set the first analysis period, and in the first analysis period, acquire fishing gear image data in real time and mark it as the first image data;

[0104] Import the first image data into the generation model for feature learning and simulated feature generation, generate a preset amount of simulated data, and mark it as the damaged traceability feature;

[0105] Use the damaged traceability feature as a training set to import into the fishing gear recognition model for model training.

[0106] It should be noted that the first analysis period is a real-time analysis period, and the second analysis period is the next consecutive period of the first analysis period. Based on the trained model, it can be applied to subsequent recognition tasks.

[0107] According to an embodiment of the present invention, it further includes:

[0108] Dividing a preset water area into multiple unit areas;

[0109] Within multiple analysis periods, the position of the fishing gear is located by satellite remote sensing, and the positioning information of different analysis periods is recorded to obtain multiple positioning data;

[0110] Based on one unit area for analysis, the distribution characteristics of the fishing gear are analyzed according to multiple positioning data. The distribution characteristics analysis includes the change in the number of fishing gear identifications, position changes, the maximum and minimum values within the area, and the information characteristics of the usage frequency of the fishing gear, and distribution characteristic information is generated;

[0111] Based on the distribution characteristic information, a two-dimensional distribution characteristic matrix is constructed;

[0112] Analyze the distribution characteristic matrix of each unit area;

[0113] Select a median value from the fishing gear identification data corresponding to multiple unit areas, and mark the corresponding unit area as the first area based on the median value;

[0114] Mark the distribution characteristic matrix corresponding to the first area as the first characteristic matrix;

[0115] Perform a difference degree analysis on each distribution characteristic matrix and the first characteristic matrix. The difference degree analysis is judged based on the sum of the eigenvalue distance and eigenvector distance of the two characteristic matrices;

[0116] Through the difference degree, a fishing gear warning priority table based on multiple unit areas is generated. The greater the difference degree, the higher the warning priority.

[0117] It should be noted that the position of the fishing gear is the general or artificially set position of the fishing gear. Satellite remote sensing has the advantage of identifying fishing gear positioning information on a large scale, but the existing technology lacks effective distribution feature analysis and early warning level allocation for the fishing gear positioning identified by satellite remote sensing. Real-time early warning often relies on manual viewing of the identification results and fishery early warning, with a low level of informatization. Therefore, based on the remote sensing identification and positioning information, the present invention processes the distribution features through the positioning information, analyzes the distribution change differences of multiple unit areas in the form of a feature matrix, screens out the unit areas with large changes, and generates an early warning priority table based on the deviation degree (difference degree analysis) between the unit area and the overall area. The greater the difference degree, the greater the deviation of the fishing gear identification distribution in this area, and stronger fishery supervision is required. The eigenvalues and eigenvectors of the feature matrix can effectively reflect the characteristics of the matrix, and the difference degree analysis between matrices is carried out based on both of them. The eigenvector distance is calculated based on the standard Euclidean distance, and the eigenvalue distance is represented by the difference between two values. The difference degree is the sum of the eigenvalue distance and the eigenvector distance of two feature matrices, and can be calculated by the sum of the corresponding weights based on the analysis requirements. The fishing gear identification data is the corresponding quantity data.

[0118] Figure 2 Fig. shows a block diagram of a fishing gear damage-oriented traceability system based on satellite remote sensing according to the present invention.

[0119] In a second aspect, the present invention also provides a fishing gear damage-oriented traceability system 2 based on satellite remote sensing. The system includes: a memory 21 and a processor 22. The memory 21 includes a fishing gear damage-oriented traceability program based on satellite remote sensing. When the fishing gear damage-oriented traceability program based on satellite remote sensing is executed by the processor 22, the following steps are implemented:

[0120] Based on the preset water area space information, construct a three-dimensional water area model;

[0121] Within a historical time period, obtain the remote sensing image data of the preset water area through satellite remote sensing technology, identify general fishing gear through the remote sensing image data, and locate the fishing gear search range;

[0122] Through the fishing gear search range, search for adjacent camera devices in the water area model, and obtain multi-angle fishing gear image data;

[0123] Import the fishing gear image data into the fishing gear identification model, extract features from the fishing gear image data through PCA analysis method to form a fishing gear feature data set. Based on multiple analysis periods, perform feature change analysis on the fishing gear feature data set. The feature change is based on the difference analysis of eigenvectors. Through the preset difference degree, screen out the analysis periods in which the feature change exceeds the expected range, mark them as change periods, and extract the data corresponding to the change periods from the fishing gear feature data to obtain a damaged image feature set;

[0124] Construct a generative model based on a variational autoencoder. The generative model includes an encoder and a decoder. Import the damaged image feature set into the encoder for feature learning. The encoder performs low-dimensional mapping on the imported data to form a latent variable z. The decoder maps the latent variable z to the original image space to obtain simulated data. The variational autoencoder is trained by minimizing the sum of the reconstruction loss and the KL divergence until the simulated data generated by the model meets the expected standard.

[0125] Set a first analysis period. In the first analysis period, obtain fishing gear image data and label it as the first image data. Import the first image data into the generative model for feature learning and simulated feature generation, and label the obtained simulated data as damaged traceability features.

[0126] Use the damaged traceability features as training data to import into the fishing gear recognition model for model training. In the second analysis period, based on the trained fishing gear recognition model, perform fishing gear recognition and traceability positioning on the fishing gear image data collected in real time.

[0127] According to an embodiment of the present invention, based on the preset water area space information, construct a three-dimensional water area model, specifically:

[0128] Obtain the preset water area space information, and the space information includes the area of the preset water area, the map contour, and the fishery resource distribution information.

[0129] Construct a three-dimensional visualization-based water area model through the space information, and through the water area model, perform dynamic monitoring of fishing boats and fishing gear.

[0130] It should be noted that the water area model is a visualization model, which monitors the water area through satellite remote sensing and camera devices, and identifies and visualizes the positions of corresponding fishing boats and fishing gear.

[0131] According to an embodiment of the present invention, within a historical time period, obtain the remote sensing image data of the preset water area through satellite remote sensing technology, and perform general fishing gear recognition through the remote sensing image data and locate the fishing gear search range, specifically:

[0132] Set a historical time period, and within a historical time period, obtain the remote sensing image data of the preset water area.

[0133] Perform image enhancement and standardization preprocessing on the remote sensing image data, and extract the analysis contour features of the image objects based on the Sobel operator.

[0134] Retrieve general fishing gear images from the image big data and obtain comparison fishing gear images.

[0135] Extract the contour features based on the Sobel operator through the comparison fishing gear images and compare the contour features.

[0136] Compare the feature similarity between the analyzed contour features and the comparison contour features, identify the corresponding fishing gear, and locate the fishing gear search range marked on the map from the remote sensing image data.

[0137] It should be noted that the image big data is an identification image storing remote sensing images corresponding to various fishing gears, which is used for comparative analysis. The net fishing gear or other artificially set fishing gear devices are a special type of fishing gear, whose shape and size are specific, and there are different shapes due to the influence of artificial layout. In traditional fishing gear identification, it is difficult to effectively identify the corresponding fishing gear. Satellite remote sensing is a high-definition image remote sensing technology that can be widely applied to fisheries. In the embodiments of the present invention, through remote sensing technology, image recognition of waters is carried out from a large-scale perspective, extensive feature search and identification positioning of specific fishing gears are carried out, and based on the positioning range, further camera image recognition is carried out to achieve efficient identification of specific fishing gears.

[0138] According to the embodiments of the present invention, searching for adjacent camera devices in the water area model based on the fishing gear search range and obtaining fishing gear image data from multiple angles specifically includes:

[0139] Search for camera devices within a preset range in the water area model based on the fishing gear search range and mark them as adjacent camera devices;

[0140] Obtain fishing gear image data from multiple angles through the adjacent camera devices.

[0141] It should be noted that the camera device is a high-definition camera device set in a preset water area.

[0142] According to the embodiments of the present invention, importing the fishing gear image data into the fishing gear recognition model, extracting features from the fishing gear image data through PCA analysis method to form a fishing gear feature data set, performing feature change analysis on the fishing gear feature data set based on multiple analysis periods, analyzing the feature change based on feature vectors, screening out the analysis periods whose feature change exceeds the expected range through a preset difference degree, marking them as change periods, and extracting the data corresponding to the change periods from the fishing gear feature data to obtain a damaged image feature set, specifically including:

[0143] Standardize the fishing gear image data and import it into the fishing gear recognition model;

[0144] Based on the PCA analysis method, perform principal component feature extraction on the fishing gear image data to form a fishing gear feature data set;

[0145] Divide a historical time period into multiple analysis periods, perform corresponding analysis period data division on the fishing gear feature data set to form multiple unit feature sets;

[0146] Convert each unit feature set into a feature vector representation to obtain a feature vector set;

[0147] Select a first feature vector set and a second feature vector set, set a vector space, calculate a first center point and a second center point in the vector space based on the first feature vector set and the second feature vector set respectively, calculate the distance between the first center point and the second center point based on the Manhattan distance, and use the calculation result as the data difference degree;

[0148] The first feature vector set and the second feature vector set are two consecutive analysis period corresponding vector sets randomly selected;

[0149] If the data difference degree is greater than a preset difference degree, mark the corresponding two consecutive analysis periods as change periods;

[0150] Discriminate all analysis periods and mark the change periods.

[0151] It should be noted that the fishing gear recognition model is specifically an image recognition model based on deep learning, and the recognition rate is improved based on the training process. The first center point is specifically the one with the shortest sum of distances to each vector in the first feature vector set, and the second center point is specifically the one with the shortest sum of distances to each vector in the second feature vector set. The data difference degree is a distance value, and the larger this value is, the greater the data difference.

[0152] It is worth mentioning that the net-like or artificially set fishing gear is a special kind of fishing gear. In the monitoring of water area fishing, it needs to be supervised and used. Therefore, efficient fishing gear recognition and positioning analysis are required. For the artificially set fishing gear device, its use process is easily affected by water flow and may be damaged. And for the fishing gear with damaged conditions, it is generally difficult to identify and locate, and it is also impossible to conduct fishing gear traceability analysis. Based on this, the present invention collects fishing gear features, performs vectorization and change analysis based on multiple periods, combines and mines change features, screens out corresponding change periods through the distance change of the feature set in the vector space. The corresponding change period is the corresponding period when the fishing gear has a damaged change in the image features. Through this process, mark and collect corresponding damaged features. Subsequently, introduce a variational autoencoder to learn the damaged features and generate simulated features to perform damaged simulation on real-time data and generate training data to improve the recognition accuracy, and further realize fishing gear positioning traceability and recognition analysis, and achieve efficient and accurate fishing gear recognition and guiding positioning traceability.

[0153] According to an embodiment of the present invention, the variational autoencoder-based generation model is constructed. The generation model includes an encoder and a decoder. The damaged image feature set is imported into the encoder for feature learning. The encoder performs low-dimensional mapping on the imported data to form a latent variable z. The decoder maps the latent variable z to the original image space and obtains simulated data. The variational autoencoder is trained by minimizing the sum of the reconstruction loss and the KL divergence until the simulated data generated by the model reaches the expected standard. Specifically:

[0154] Initialize the generation model and set multiple convolutional layers for the encoder and the decoder respectively;

[0155] Perform data cleaning and standardization processing on the damaged image feature set;

[0156] Perform low-dimensional mapping on the damaged image feature set through the encoder, and output the latent variable z and the variable distribution;

[0157] The decoder maps the latent variable z to the original image space to obtain the output data, which is marked as simulated data;

[0158] Based on the preset reconstruction loss, judge the difference between the simulated data and the imported data. Based on the KL divergence, judge the difference between the variable distribution and the preset distribution. By minimizing the sum of the reconstruction loss and the KL divergence, the generation model is cyclically trained until the sum of the reconstruction loss and the KL divergence is within the preset range.

[0159] It should be noted that the decoder is a multi-layer transposed convolution. The preset distribution can be compared using a Gaussian distribution. The preset reconstruction loss can be calculated based on the mean square error or cross entropy.

[0160] According to an embodiment of the present invention, the first analysis period is set. In the first analysis period, fishing gear image data is acquired and marked as the first image data. The first image data is imported into the generation model for feature learning and simulated feature generation. The obtained simulated data is marked as the damaged traceability feature. The damaged traceability feature is used as training data to import into the fishing gear recognition model for model training. Specifically:

[0161] Set the first analysis period, and in the first analysis period, acquire fishing gear image data in real time and mark it as the first image data;

[0162] Import the first image data into the generation model for feature learning and simulated feature generation, generate the simulated data of the preset data volume, and mark it as the damaged traceability feature;

[0163] Use the damaged traceability feature as the training set to import into the fishing gear recognition model for model training.

[0164] It should be noted that the first analysis period is a real-time analysis period, and the second analysis period is the next consecutive period of the first analysis period. Based on the trained model, it can be applied to subsequent recognition tasks.

[0165] The third aspect of the present invention also provides a computer-readable storage medium, which includes a satellite remote sensing-based fishing gear damage-oriented traceability program. When the satellite remote sensing-based fishing gear damage-oriented traceability program is executed by a processor, the steps of the satellite remote sensing-based fishing gear damage-oriented traceability method as described in any one of the above are realized.

[0166] The present invention discloses a satellite remote sensing-based fishing gear damage-oriented traceability method. First, satellite remote sensing technology is used to obtain remote sensing images of waters in a historical time period, and a preset fishing gear range is identified and located. Fishing gear images are obtained by searching for adjacent imaging devices through a water area model, features are extracted using PCA analysis, feature changes are analyzed, analysis periods with feature changes exceeding the expected range are screened out, and a damaged image feature set is extracted. A variational autoencoder generation model is constructed, and through the learning and training of the encoder and decoder, simulated data is generated. The fishing gear image data is obtained in the first analysis period, imported into the generation model to generate damage traceability features, and used as training data to optimize the fishing gear recognition model. In the second analysis period, the trained model is used to identify and trace the real-time fishing gear image. Through the present invention, the identification analysis and positioning traceability of fishing gear can be realized, and efficient and accurate fishing gear identification and guiding positioning traceability can be achieved.

[0167] In several embodiments provided by the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of the units is only a logical function division, and there may be other division methods in actual implementation. For example, multiple units or components can be combined, or integrated into another system, or some features can be ignored, or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed with each other can be through some interfaces. The indirect coupling or communication connection of devices or units can be electrical, mechanical, or other forms.

[0168] The units described above as separate components may or may not be physically separated. The components shown as units may or may not be physical units; they can be located in one place or distributed to multiple network units; some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0169] In addition, each functional unit in the embodiments of the present invention may all be integrated into one processing unit, or each unit may be separately taken as one unit, or two or more units may be integrated into one unit; the above-mentioned integrated unit may be implemented in the form of hardware, or in the form of a hardware plus software functional unit.

[0170] Those of ordinary skill in the art can understand that all or part of the steps of implementing the above method embodiments can be completed by hardware related to program instructions. The foregoing program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps including the above method embodiments; and the foregoing storage medium includes: removable storage devices, read-only memory (ROM), random access memory (RAM), magnetic disks or optical discs and other various media that can store program codes.

[0171] Alternatively, if the above-mentioned integrated unit of the present invention is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the embodiments of the present invention, in essence, or the part that contributes to the prior art can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the methods described in the embodiments of the present invention. And the foregoing storage medium includes: removable storage devices, ROM, RAM, magnetic disks or optical discs and other various media that can store program codes.

[0172] The above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should all be covered by the protection scope of the present invention.

Claims

1. A method for tracing damaged fishing gear based on satellite remote sensing, characterized in that: include: Based on the preset water area spatial information, a three-dimensional water area model is constructed; In a historical period, remote sensing image data of preset waters are obtained through satellite remote sensing technology, common fishing gear is identified through remote sensing image data, and the fishing gear search range is located; Through the fishing gear search range, search for nearby camera devices in the water area model and obtain fishing gear image data from multiple angles; The fishing gear image data is imported into the fishing gear recognition model. The features of the fishing gear image data are extracted through the PCA analysis method to form a fishing gear feature data set. Based on multiple analysis cycles, feature change analysis is performed on the fishing gear feature data set. The feature change is analyzed based on the feature vector. The analysis cycle whose feature change exceeds the expected range is screened out through the preset difference degree and marked as a change cycle. The corresponding data of the change cycle is extracted from the fishing gear feature data to obtain a damaged image feature set. Construct a generative model based on a variational autoencoder. The generative model includes an encoder and a decoder. The feature set of the damaged image is imported into the encoder for feature learning. The encoder performs low-dimensional mapping on the imported data to form a latent variable z. The decoder maps the latent variable z to the original image space and obtains simulated data. The variational autoencoder is trained by minimizing the sum of the reconstruction loss and the KL divergence until the simulated data generated by the model meets the expected standard. A first analysis cycle is set, fishing gear image data is obtained in the first analysis cycle and marked as first image data, the first image data is imported into the generation model for feature learning and simulation feature generation, and the obtained simulation data is marked as damage traceability features; The damaged traceability features are imported as training data into the fishing gear recognition model for model training. In the second analysis cycle, the fishing gear recognition and traceability positioning are performed on the real-time collected fishing gear image data based on the trained fishing gear recognition model.

2. According to claim 1, a method for tracing damaged fishing gear based on satellite remote sensing is characterized in that: The three-dimensional water area model is constructed based on the preset water area spatial information, specifically: Acquire spatial information of a preset water area, wherein the spatial information includes the area, map outline, and fishery resource distribution information of the preset water area; A water area model based on three-dimensional visualization is constructed through spatial information, and dynamic monitoring of fishing boats and fishing gear is carried out through the water area model.

3. The method for tracing damaged fishing gear based on satellite remote sensing according to claim 1, characterized in that: In a historical period, remote sensing image data of a preset water area is obtained by satellite remote sensing technology, common fishing gear is identified by remote sensing image data, and the fishing gear search range is located, specifically: Set a historical time period and obtain remote sensing image data of a preset water area within the historical time period; Perform image enhancement and standardization preprocessing on remote sensing image data, and extract the analytical contour features of image objects based on the Sobel operator; Retrieve common fishing gear images from image big data and obtain comparative fishing gear images; By comparing the fishing gear images, contour features are extracted based on the Sobel operator and contour features are compared; The feature similarity between the analyzed contour features and the compared contour features is compared, and the corresponding fishing gear is identified, and the fishing gear search range is marked from the positioning map in the remote sensing image data.

4. The method for tracing damaged fishing gear based on satellite remote sensing according to claim 3 is characterized in that: The method searches for adjacent camera devices in the water area model through the fishing gear search range, and obtains fishing gear image data from multiple angles, specifically: Based on the fishing gear search range, searching for camera devices within a preset range in the water area model and marking them as adjacent camera devices; The fishing gear image data at multiple angles is acquired through the proximity camera device.

5. The method for tracing damaged fishing gear based on satellite remote sensing according to claim 4 is characterized in that: The fishing gear image data is imported into the fishing gear recognition model, and the fishing gear image data is subjected to feature extraction by PCA analysis to form a fishing gear feature data set. Based on multiple analysis cycles, feature change analysis is performed on the fishing gear feature data set. The feature change is subjected to difference analysis based on the feature vector. By presetting the difference degree, the analysis cycle whose feature change exceeds the expected range is screened out and marked as a change cycle. The data corresponding to the change cycle is extracted from the fishing gear feature data to obtain a damaged image feature set, which is specifically: The fishing gear image data is standardized and imported into the fishing gear recognition model; Based on the PCA analysis method, the principal component features of the fishing gear image data are extracted to form a fishing gear feature data set; Divide a historical time period into multiple analysis periods, divide the fishing gear feature data set into corresponding analysis period data, and form multiple unit feature sets; Convert each unit feature set into a feature vector representation to obtain a feature vector set; Selecting a first eigenvector set and a second eigenvector set, setting a vector space, respectively calculating a first center point and a second center point based on the first eigenvector set and the second eigenvector set in the vector space, calculating a distance between the first center point and the second center point based on the Manhattan distance, and using the calculation result as the data difference; The first eigenvector set and the second eigenvector set are randomly selected vector sets corresponding to two consecutive analysis periods; If the data difference is greater than the preset difference, the corresponding two consecutive analysis periods are marked as change periods; Identify all analysis periods and mark the change periods.

6. The method for tracing damaged fishing gear based on satellite remote sensing according to claim 5, characterized in that: The construction is based on a generative model of a variational autoencoder, the generative model includes an encoder and a decoder, a damaged image feature set is imported into the encoder for feature learning, the imported data is low-dimensionally mapped by the encoder to form a latent variable z, the latent variable z is mapped to the original image space by the decoder, and simulated data is obtained, and the variational autoencoder is trained by minimizing the sum of the reconstruction loss and the KL divergence until the simulated data generated by the model reaches the expected standard, specifically: Initialize the generative model and set multiple convolutional layers for the encoder and decoder respectively; Perform data cleaning and standardization on the damaged image feature set; The encoder performs low-dimensional mapping on the feature set of the damaged image and outputs the latent variable z and variable distribution; The decoder maps the latent variable z to the original image space and obtains the output data, which is marked as simulated data; Based on the preset reconstruction loss, the difference between the simulated data and the imported data is determined. Based on the KL divergence, the difference between the variable distribution and the preset distribution is determined. By minimizing the sum of the reconstruction loss and the KL divergence, the model is trained cyclically until the sum of the reconstruction loss and the KL divergence is within the preset range.

7. The method for tracing damaged fishing gear based on satellite remote sensing according to claim 6, characterized in that: The first analysis cycle is set, fishing gear image data is obtained in the first analysis cycle and marked as first image data, the first image data is imported into the generation model for feature learning and simulation feature generation, and the obtained simulation data is marked as damage traceability features; the damage traceability features are imported as training data into the fishing gear recognition model for model training, specifically: Setting a first analysis period, acquiring fishing gear image data in real time during the first analysis period and marking the data as first image data; Importing the first image data into the generation model, performing feature learning and simulation feature generation, generating a preset amount of simulation data, and marking it as a damage tracing feature; The damage traceability features are imported as training sets into the fishing gear recognition model for model training.

8. A fishing gear damage guidance and tracing system based on satellite remote sensing, characterized in that: The system includes: a memory and a processor, wherein the memory includes a fishing gear damage guidance tracing program based on satellite remote sensing, and the fishing gear damage guidance tracing program based on satellite remote sensing is executed by the processor to implement the following steps: Based on the preset water area spatial information, a three-dimensional water area model is constructed; In a historical period, remote sensing image data of preset waters are obtained through satellite remote sensing technology, common fishing gear is identified through remote sensing image data, and the fishing gear search range is located; Through the fishing gear search range, search for nearby camera devices in the water area model and obtain fishing gear image data from multiple angles; The fishing gear image data is imported into the fishing gear recognition model. The features of the fishing gear image data are extracted through the PCA analysis method to form a fishing gear feature data set. Based on multiple analysis cycles, feature change analysis is performed on the fishing gear feature data set. The feature change is analyzed based on the feature vector. The analysis cycle whose feature change exceeds the expected range is screened out through the preset difference degree and marked as a change cycle. The corresponding data of the change cycle is extracted from the fishing gear feature data to obtain a damaged image feature set. Construct a generative model based on a variational autoencoder. The generative model includes an encoder and a decoder. The feature set of the damaged image is imported into the encoder for feature learning. The encoder performs low-dimensional mapping on the imported data to form a latent variable z. The decoder maps the latent variable z to the original image space and obtains simulated data. The variational autoencoder is trained by minimizing the sum of the reconstruction loss and the KL divergence until the simulated data generated by the model meets the expected standard. A first analysis cycle is set, fishing gear image data is obtained in the first analysis cycle and marked as first image data, the first image data is imported into the generation model for feature learning and simulation feature generation, and the obtained simulation data is marked as damage traceability features; The damaged traceability features are imported as training data into the fishing gear recognition model for model training. In the second analysis cycle, the fishing gear recognition and traceability positioning are performed on the real-time collected fishing gear image data based on the trained fishing gear recognition model.

9. The fishing gear damage guidance and tracing system based on satellite remote sensing according to claim 1 is characterized in that: The three-dimensional water area model is constructed based on the preset water area spatial information, specifically: Acquire spatial information of a preset water area, wherein the spatial information includes the area, map outline, and fishery resource distribution information of the preset water area; A water area model based on three-dimensional visualization is constructed through spatial information, and dynamic monitoring of fishing boats and fishing gear is carried out through the water area model.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a fishing gear damage guidance and tracing program based on satellite remote sensing. When the fishing gear damage guidance and tracing program based on satellite remote sensing is executed by a processor, the steps of fishing gear damage guidance and tracing based on satellite remote sensing as described in any one of claims 1 to 7 are implemented.

Images