A fish catch intelligent identification method and system

Abstract

The application discloses a kind of fish catch intelligent identification method and system, belong to fishery resources management technical field.The application constructs fish catch species multi-source identification vector containing multi-source data, then constructs spatial weight matrix with fishing site, determines fish catch weight gradient distribution by autocorrelation analysis, further determines weight propagation path, constructs weight propagation regional link, then based on fishing time and fishing site, obtain the spatial movement mode of fish catch node, to construct fishing propagation time sequence link, by fusing weight propagation regional link and fishing propagation time sequence link, generate multi-source propagation time sequence link, so that the generated fish catch multi-source data relationship diagram can fully couple fishing operation mode and fish catch life characteristics, realize intelligent identification to fish catch by graph feature extraction, improve the comprehensiveness and accuracy of fish catch identification, so that fish catch identification can provide basis for formulating fishing ban period, fishing quota and other fishery resources operation.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of fishery resource management, and particularly relates to a fish catch intelligent identification method and system. BACKGROUND

[0002] With the continuous development of fishery resources, realizing the sustainable development and management of fishery resources has become a crucial factor in the process of fishery resource development. Among them, by identifying the fish catch in the fish catch operation, the aggregation area of the fish catch, the abnormal fish catch behavior and other data can be judged, thereby providing data basis for fishery resource management, providing basis for formulating fishing prohibition period, fishing quota and other fishery resource operations, and realizing the sustainable development and management of fishery resources.

[0003] The current fish catch identification scheme is usually realized based on the combination of a neural network model and mathematical statistics. Specifically, the neural network model is used to extract features of images in the fish catch operation process, so as to identify the corresponding fish catch species, and other types of data recorded by sensors are combined to integrate data using a mathematical statistical method, so as to obtain the fish catch identification result. Although this method can quickly identify the fish catch, due to the influence of the large number of fish catch species in the actual fish catch operation process, different fish catches may present similar results on the image, which leads to a large error in the image processing result of the neural network model, and reduces the accuracy of the fish catch identification result. More seriously, due to the migratory characteristics of the fish catch, different fish catches have a complex coupling relationship in time and place, so that the traditional mathematical statistical method of fish catch identification statistics has a large data deviation, and cannot comprehensively and accurately analyze the multiple characteristics of the fish catch in the actual fishery environment. Therefore, there is an urgent need for a fish catch intelligent identification method and system to solve the defects of the prior art. SUMMARY

[0004] The present application aims to provide a fish catch intelligent identification method and system to solve the above technical problems, by constructing a weight propagation region link and a catch propagation time sequence link, fusing the weight propagation region link and the catch propagation time sequence link with a similarity matching matrix, and using a multi-source propagation time sequence link to realize the graph modeling of the fish catch operation, and using a fish catch multi-source data relationship graph to realize the intelligent identification of the fish catch, thereby improving the comprehensiveness and accuracy of the fish catch identification.

[0005] To solve the above technical problems, the present application embodiment provides a fish catch intelligent identification method, comprising:

[0006] Obtaining a plurality of fishery images, and the corresponding fishing time, fishery weight and fishing location of each fishery image, and then constructing a plurality of fishery species multi-source identification vectors; based on a preset fishery species identification neural network model, the fishery species multi-source identification vectors are subjected to feature recognition to obtain the fishery species corresponding to each fishery species multi-source identification vector;

[0007] Based on the fishery species and fishery species multi-source identification vector, a plurality of fishery clusters are initialized, each fishery cluster containing a plurality of fishery nodes of the same fishery species;

[0008] Based on the fishing location, a spatial weight matrix is constructed; the fishery weight is subjected to autocorrelation analysis to determine the fishery weight gradient distribution; the weight propagation path is constructed based on the spatial weight matrix and the fishery weight gradient distribution, and the fishery nodes in the fishery cluster are connected in turn to determine the weight propagation area link of each fishery cluster;

[0009] Based on the fishing time and fishing location, the spatial movement mode of each fishery node in each fishery cluster is identified, and based on the spatial movement mode, the fishing propagation time sequence link of each fishery cluster is constructed;

[0010] The similarity matching matrix between the weight propagation area link and the fishing propagation time sequence link of each fishery cluster is constructed, and based on the similarity matching matrix, the fishery nodes are screened from the weight propagation area link and the fishing propagation time sequence link respectively for fusion to form multi-source fishery nodes, and then the multi-source propagation time sequence link of each fishery cluster is determined;

[0011] Based on the multi-source propagation time sequence link, the fishery multi-source data relationship graph of each fishery cluster is constructed; the fishery multi-source data relationship graph is subjected to graph feature extraction to obtain the fishery intelligent recognition result.

[0012] It can be understood that the fish catch image, the fish catch weight, the fishing time and the fishing location are used to construct a fish catch species multi-source identification vector containing multi-source data, so that the inaccuracy of the traditional neural network model in predicting the fish catch species of a single image dimension is avoided; then the fish catch cluster is initialized and the fish catch nodes in the fish catch cluster have the same fish catch species, so that the identification result of the fish catch can be more targeted to the same type of fish catch; then the fishing location is used to construct a spatial weight matrix, and the fish catch weight is subjected to autocorrelation analysis to determine the fish catch weight gradient distribution, so as to determine the weight propagation path and construct the weight propagation regional link, and the coupling analysis of the fish catch in the weight dimension and the active region dimension is realized; then the fishing time and the fishing location are used to obtain the spatial movement mode of the fish catch node, so as to construct the fishing propagation time sequence link and realize the coupling analysis of the fish catch in the active region dimension and the fish catch operation time dimension; the weight propagation regional link and the fishing propagation time sequence link are fused by constructing a similarity matching matrix to generate a multi-source propagation time sequence link, so that the fish catch multi-source data relationship graph generated based on the multi-source propagation time sequence link can fully couple the fishing operation mode and the fish catch life characteristics, and then the relationship between the fish catch fishing operation and the fish catch habit is mined by extracting the graph features, the intelligent identification of the fish catch in the fish catch fishing operation process is realized, the comprehensiveness and accuracy of the fish catch identification are improved, the fish catch identification can provide a basis for formulating the fishing prohibition period, the fishing quota and other fishery resource operations, and the sustainable development and management of fishery resources are realized.

[0013] Correspondingly, the embodiment of the present application provides a fish catch intelligent identification system, comprising: a fish catch species identification module, a fish catch cluster construction module, a weight propagation regional link construction module, a fishing propagation time sequence link construction module, a multi-source propagation time sequence link construction module and a graph feature extraction module.

[0014] The fish catch species identification module is used to obtain a plurality of fish catch images, and the fishing time, the fish catch weight and the fishing location corresponding to each fish catch image, and then construct a plurality of fish catch species multi-source identification vectors; the fish catch species multi-source identification vectors are subjected to feature identification based on a preset fish catch species identification neural network model, and the fish catch species corresponding to each fish catch species multi-source identification vector is obtained.

[0015] The fish catch cluster construction module is used to initialize a plurality of fish catch clusters based on the fish catch species and the fish catch species multi-source identification vectors, and each fish catch cluster contains a plurality of fish catch nodes of the same fish catch species.

[0016] The weight propagation area link construction module is configured to construct a spatial weight matrix based on the fishing locations; perform autocorrelation analysis on the fishery weight to determine fishery weight gradient distribution; construct a weight propagation path based on the spatial weight matrix and the fishery weight gradient distribution, and sequentially connect the fishery nodes in the fishery cluster to determine the weight propagation area link of each fishery cluster;

[0017] The fishing propagation time sequence link construction module is configured to identify the spatial movement mode of each fishery node in each fishery cluster based on the fishing time and fishing location, and construct the fishing propagation time sequence link of each fishery cluster based on the spatial movement mode;

[0018] The multi-source propagation time sequence link construction module is configured to construct a similarity matching matrix between the weight propagation area link and the fishing propagation time sequence link of each fishery cluster, and based on the similarity matching matrix, filter and fuse fishery nodes from the weight propagation area link and the fishing propagation time sequence link respectively to form a multi-source fishery node, and further determine the multi-source propagation time sequence link of each fishery cluster;

[0019] The graph feature extraction module is configured to construct a fishery multi-source data relationship graph of each fishery cluster based on the multi-source propagation time sequence link; and perform graph feature extraction on the fishery multi-source data relationship graph to obtain a fishery intelligent recognition result.

[0020] Understandably, this system constructs a multi-source identification vector for fish species using fish images, fish weight, fishing time, and fishing location, thus avoiding the inaccuracy of traditional neural network models in predicting fish species based on a single image dimension. Next, it initializes fish clusters and ensures that fish nodes within the clusters belong to the same fish species, making the identification results more targeted to fish of the same type. Then, it constructs a spatial weight matrix based on fishing location, performs autocorrelation analysis on fish weight, determines the fish weight gradient distribution, thereby identifying the weight propagation path and constructing a weight propagation regional link, achieving coupled analysis of fish in the weight and activity region dimensions. Finally, based on fishing time and location, it obtains the fish node... Spatial movement patterns were used to construct a fishing propagation time-series link, enabling coupled analysis of catches in terms of activity area and fishing operation time. By constructing a similarity matching matrix, the regional and temporal propagation links were fused to generate a multi-source propagation time-series link. This allowed the multi-source data relationship graph of catches generated based on the multi-source propagation time-series link to comprehensively couple fishing operation patterns and catch characteristics. Subsequently, graph feature extraction was used to explore the relationship between fishing operations and catch habits, enabling intelligent identification of catches during the fishing operation process. This improved the comprehensiveness and accuracy of catch identification, providing a basis for fishery resource operations such as setting fishing bans and fishing quotas, and realizing the sustainable development and management of fishery resources. Attached Figure Description

[0021] Figure 1 A flowchart illustrating the steps of an intelligent fish catch identification method provided in an embodiment of the present invention;

[0022] Figure 2 This is a schematic diagram of the structure of an intelligent fish catch recognition system provided in an embodiment of the present invention. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] Example 1

[0025] Please refer to Figure 1 , Figure 1 The flowchart of a method for intelligent identification of fish catch provided in an embodiment of the present invention includes steps S101 to S106.

[0026] Step S101: Obtain a plurality of fishery images, and the fishing time, fishery weight and fishing location corresponding to each of the fishery images, and further construct a plurality of fishery species multi-source identification vectors; perform feature recognition on the fishery species multi-source identification vectors based on a preset fishery species identification neural network model to obtain the fishery species corresponding to each of the fishery species multi-source identification vectors.

[0027] In the present embodiment, during each fishery fishing operation, each fishery obtained by fishing is photographed using a camera to obtain a plurality of fishery images, and the fishing time and fishing location of each fishery are recorded synchronously; the fishing time is set in the format of a time stamp, and the fishing location is set as a longitude and latitude coordinate; the fishery weight is measured by electronic scales, sensors and other devices.

[0028] In the present embodiment, the obtaining of a plurality of fishery images, and the fishing time, fishery weight and fishing location corresponding to each of the fishery images, and further the construction of a plurality of fishery species multi-source identification vectors; the feature recognition on the fishery species multi-source identification vectors based on a preset fishery species identification neural network model to obtain the fishery species corresponding to each of the fishery species multi-source identification vectors comprises: obtaining a plurality of fishery images during a fishery fishing operation, and the fishing time, fishery weight and fishing location corresponding to each of the fishery images; taking each pixel point in each of the fishery images as a neighborhood block center point, assigning a neighborhood block to each pixel point in each of the fishery images, and performing gray scale conversion on each of the fishery images based on the neighborhood blocks to obtain a gray scale fishery image of each of the fishery images; performing contour recognition on the gray scale fishery image of each of the fishery images based on a preset edge detection algorithm to determine a fishery contour point set of each of the fishery images; determining a fishery area and a fishery contour range of each of the fishery images based on the fishery contour point set; performing weighted average on the RGB color of the pixel points in the fishery contour range of each of the fishery images to determine a fishery color value of each of the fishery images; constructing a fishery species multi-source identification vector corresponding to each of the fishery images based on the fishery area, fishery color value, fishing time, fishery weight and fishing location of each of the fishery images; performing feature recognition on the fishery species multi-source identification vectors based on a preset fishery species identification neural network model to obtain a plurality of fishery species identification candidate results corresponding to each of the fishery species multi-source identification vectors; performing screening on the fishery species identification candidate results based on pre-acquired historical fishery fishing operation data to obtain the fishery species corresponding to each of the fishery species multi-source identification vectors.

[0029] In an optional embodiment, after the fish catch image is converted into a grayscale fish catch image, the fish catch image can highlight the fish catch in the background; the edge detection algorithm is set as a Canny edge detection algorithm, the contour of the grayscale fish catch image is recognized by the Canny edge detection algorithm, and the fish catch contour point set of each fish catch image is determined; the fish catch contour point set contains a plurality of pixel points; the contour surrounded by the pixel points in the fish catch contour point set is the fish catch contour range, and the fish catch area can be calculated by the coordinates of the pixel points in the fish catch contour point set; then the channel values of the RGB colors of the pixel points in the fish catch contour range are extracted, the channel values of the RGB colors of all the pixel points in the fish catch contour range are weighted and averaged, and the obtained value is taken as the fish catch color value of the fish catch image; then the fish catch area , the fish catch color value , the fishing time , the fish catch weight and the fishing location are used to construct a fish catch type multi-source identification vector, and the form of the fish catch type multi-source identification vector is set as: .

[0030] In this embodiment, each pixel point is assigned a neighborhood block to realize grayscale conversion in the pixel point dimension, which can reduce the influence of light changes and background interference on image quality, thereby improving the accuracy of subsequent contour recognition; then the fish catch area and the fish catch color value are recognized by the fish catch contour point set, and the fishing time, the fish catch weight and the fishing location are combined, so that the fish catch type multi-source identification vector constructed can integrate data in multiple dimensions, thereby comprehensively representing the characteristics of different fish catches and avoiding the inaccuracy of traditional neural network models in predicting the fish catch type of a single image dimension; then the multiple fish catch type identification candidate results output by the fish catch type identification neural network model are screened by the historical fish catch operation data, the historical data factor is introduced, and the identification accuracy of the fish catch type is further improved, which provides a data basis for subsequent fish catch identification and further improves the accuracy and comprehensiveness of fish catch identification.

[0031] In the embodiment, the assigning a neighborhood block for each pixel point in each of the fish catch images with each pixel point in each of the fish catch images as a neighborhood block center point and performing gray scale conversion on each of the fish catch images based on the neighborhood block to obtain a gray scale fish catch image of each of the fish catch images comprises: assigning a neighborhood block for each pixel point in each of the fish catch images with each pixel point in each of the fish catch images as a neighborhood block center point; calculating a pixel point gray scale mean and a pixel point gray scale standard deviation of all the pixel points in the neighborhood block; assigning a gray scale weight for each pixel point in the neighborhood block based on the pixel point gray scale mean and the pixel point gray scale standard deviation and in combination with a preset Gaussian function; calculating a gray scale threshold of the neighborhood block based on the gray scale weight and the gray scale value of each pixel point in the neighborhood block; setting a gray scale conversion value of the pixel point corresponding to the neighborhood block center point as a preset first gray scale value if the gray scale threshold is greater than the gray scale value of the neighborhood block center point; setting the gray scale conversion value of the pixel point corresponding to the neighborhood block center point as a preset second gray scale value if the gray scale threshold is less than or equal to the gray scale value of the neighborhood block center point; and performing gray scale conversion on each of the fish catch images based on the gray scale conversion value of each pixel point in each of the fish catch images to obtain a gray scale fish catch image of each of the fish catch images.

[0032] In an optional embodiment, the size of the neighborhood block is set to 3x3, i.e. a total of 9 unit grids, each of which can accommodate one pixel point; a neighborhood block is assigned to each pixel point, and each pixel point is at the center point (i.e. the unit grid) of the respective neighborhood block. In particular, in the edge region of the fish catch image, some unit grids in the neighborhood block do not have pixel points, in which case the gray scale value is set to 0. Then in each neighborhood block, the pixel point gray scale mean and the pixel point gray scale standard deviation of all the pixel points are calculated based on the gray scale values of the pixel points. Then the preset Gaussian function is set to a one-dimensional Gaussian function. The formula of the one-dimensional Gaussian function is: ; wherein, represents the gray scale value of the pixel point, is the gray scale weight, is the pixel point gray scale mean, is the pixel point gray scale standard deviation; and All are mathematical constants; the gray value of each pixel point in the neighborhood block is substituted into a one-dimensional Gaussian function to obtain the gray weight of each pixel point; then the gray value and the gray weight of each pixel point in the neighborhood block are weighted and summed and then a constant is subtracted to obtain the gray threshold of the neighborhood block; the constant is set to 40 in this embodiment, and the size of the gray threshold can be fine-tuned by subtracting the constant, and the size of the constant can be modified according to actual needs; after obtaining the gray threshold, if the gray threshold is greater than the gray value of the pixel point corresponding to the center point of the neighborhood block, the gray conversion value of the pixel point corresponding to the center point of the neighborhood block is set to a preset first gray value; if it is less than or equal to, it is set to a preset second gray value; the preset first gray value is 255; the preset second gray value is 0; and so on. The same operation is performed on each neighborhood block to obtain the gray conversion value of each pixel point, and the gray value of each pixel point is set through the gray conversion value, so that the fish catch image is converted into a gray fish catch image.

[0033] In this embodiment, by assigning a neighborhood block to each pixel point, then calculating the pixel gray mean value and the pixel gray standard deviation, the image local texture change of the pixel point dimension in the neighborhood block can be captured, the gray weight is assigned by the Gaussian function, which can ensure the smoothness and consistency of the gray weight, and reduce the edge blur or distortion; then the gray conversion value is dynamically set by comparing the gray threshold, which realizes the dynamic conversion of the gray image, effectively highlights the difference between the fish catch and the background, thereby improving the clarity and accuracy of the subsequent contour recognition, and further improving the accuracy of the fish catch recognition.

[0034] In this embodiment, the preset fish catch species identification neural network model is used to perform feature identification on the fish catch species multi-source identification vector, to obtain a plurality of fish catch species identification candidate results corresponding to each fish catch species multi-source identification vector; the fish catch species identification candidate results are screened based on the pre-acquired historical fish catch operation data, to obtain a fish catch species corresponding to each fish catch species multi-source identification vector, including: based on the preset fish catch species identification neural network model, performing feature identification on the fish catch species multi-source identification vector, to obtain a plurality of fish catch species identification candidate results of each fish catch species multi-source identification vector; based on the pre-acquired historical fish catch operation data, combining the fish catch species identification candidate results, constructing a fish catch reference vector of each fish catch species identification candidate result; wherein, the data dimensions of the fish catch reference vector and the fish catch species multi-source identification vector are the same; calculating the similarity between the fish catch reference vector corresponding to each fish catch species identification candidate result and the fish catch species multi-source identification vector; screening the fish catch species identification candidate result with the highest similarity between the fish catch reference vector and the fish catch species multi-source identification vector, as the fish catch species corresponding to each fish catch species multi-source identification vector.

[0035] In an optional embodiment, the fish catch species identification neural network model is set as a convolutional neural network (CNN). Since the training and prediction of the neural network model have been relatively mature in the current image recognition field, a mature convolutional neural network is used in this embodiment. The skilled person can select other neural network models according to actual needs. The historical fish catch fishing operation data can be obtained by consulting the database, which records the theoretical fish catch area, fish catch color value, fishing time, fish catch weight and fishing location corresponding to each fish catch species. Therefore, the fish catch reference vector includes five dimensions of fish catch area, fish catch color value, fishing time, fish catch weight and fishing location, and the similarity calculation can use Pearson correlation coefficient, Jaccard similarity coefficient, etc.

[0036] Step S102: initializing a plurality of fish catch clusters based on the fish catch species and the fish catch species multi-source identification vector, each fish catch cluster containing a plurality of fish catch nodes of the same fish catch species.

[0037] In this embodiment, initializing a plurality of fish catch clusters based on the fish catch species and the fish catch species multi-source identification vector, each fish catch cluster containing a plurality of fish catch nodes of the same fish catch species, includes: constructing a plurality of nodes based on the number of fish catch species multi-source identification vectors; taking the fishing time, fishing location and fish catch weight in the fish catch species multi-source identification vector, and the fish catch species corresponding to the fish catch species multi-source identification vector, as the attributes of the nodes, to obtain a plurality of fish catch nodes; based on the fish catch species of the fish catch nodes, classifying the fish catch nodes in the same fish catch cluster to contain only fish catch nodes of the same fish catch species, to obtain a plurality of fish catch clusters, each fish catch cluster containing a plurality of fish catch nodes of the same fish catch species.

[0038] This embodiment can simplify the data expression of fish catch species, fishing time, fishing location and fish catch weight by constructing fish catch nodes, and provide a data basis for the construction of weight propagation regional link, fishing propagation time sequence link and multi-source propagation time sequence link in the subsequent process; constructing fish catch clusters by fish catch species can make the identification result of fish catch more targeted to fish catch of the same fish catch species, avoid analysis interference between different fish catch species, and thus improve the accuracy of fish catch identification.

[0039] Step S103: constructing a spatial weight matrix based on the fishing locations; performing autocorrelation analysis on the fishery weight to determine fishery weight gradient distribution; constructing a weight propagation path based on the spatial weight matrix and the fishery weight gradient distribution, sequentially connecting the fishery nodes in the fishery cluster to determine the weight propagation regional link of each fishery cluster.

[0040] In the embodiment, the constructing a spatial weight matrix based on the fishing locations; performing autocorrelation analysis on the fishery weight to determine fishery weight gradient distribution; constructing a weight propagation path based on the spatial weight matrix and the fishery weight gradient distribution, sequentially connecting the fishery nodes in the fishery cluster to determine the weight propagation regional link of each fishery cluster includes: in each fishery cluster, performing cluster analysis on the fishery nodes based on the fishing locations of the fishery nodes to obtain a plurality of fishery clusters of each fishery cluster; if two fishery nodes belong to the same fishery cluster, calculating the Euclidean distance between the fishing locations of the two fishery nodes as the spatial weight of the two fishery nodes; if the two fishery nodes do not belong to the same fishery cluster, setting the spatial weight of the two fishery nodes as a preset negative value; forming a spatial weight matrix of each fishery cluster based on the spatial weight of each two fishery nodes in each fishery cluster; performing autocorrelation analysis on the fishery weight of each fishery cluster based on the spatial weight matrix and the fishery weight to determine the global correlation and the local correlation set of each fishery cluster; determining the fishery weight gradient distribution of each fishery cluster based on the global correlation and the local correlation set; constructing a weight propagation path of each fishery cluster based on the spatial weight matrix and the fishery weight gradient distribution; sequentially connecting the fishery nodes in each fishery cluster based on the weight propagation path to determine the weight propagation regional link of each fishery cluster.

[0041] In an optional embodiment, since the fishing locations use latitude and longitude coordinates, the embodiment uses a K-means clustering algorithm to process the geographic location data to divide the areas according to the geographic location data; the fishing locations of the fish catch nodes are processed by the K-means clustering algorithm to obtain a plurality of fish catch clusters, and each fish catch cluster contains a plurality of fish catch nodes; the fish catch nodes belonging to the same fish catch cluster have great similarity in the fishing locations, and thus the fish catch cluster can be identified as a fishing area; if two fish catch nodes belong to the same fish catch cluster, the Euclidean distance of their fishing locations is calculated as a spatial weight; if two fish catch nodes do not belong to the same fish catch cluster, the spatial weight of the fish catch nodes is set to -1 (i.e., a preset negative value, and the specific negative value can be set by experimenters); since the Euclidean distance is a value greater than or equal to 0, the negative value can be set to determine whether two fish catch nodes belong to the same fish catch cluster; then, a spatial weight matrix can be constructed through the spatial weight between each two fish catch nodes; the elements of the spatial weight matrix are the spatial weights between two fish catch nodes.

[0042] The embodiment clusters the fish catch nodes by the fishing locations, groups the nodes close in space into clusters, introduces the natural aggregation characteristics of the fish catch, and constructs the spatial weight based on the fish catch cluster and the Euclidean distance, so that the spatial weight is more consistent with the actual distribution characteristics of the fish catch; then, the global correlation and the local correlation set are obtained through the autocorrelation analysis, the aggregation or diffusion trend of the fish catch weight in space can be identified, so that the fish catch weight gradient distribution can accurately reflect the weight trend of the fish catch at different distribution locations, and thus the weight propagation regional link of the constructed fish catch cluster can not only consider the mutation of the fish catch in the weight dimension, but also introduce the actual distribution factors of the fish catch, so as to improve the coupling accuracy of the weight propagation regional link in the two dimensions of the location and the weight of the fish catch, and thus improve the accuracy of the subsequent multi-source propagation timing link and the accuracy of the fish catch identification.

[0043] In the embodiment, the autocorrelation analysis of the fish weight of each fish cluster in each fish group of each fish cluster is performed based on the spatial weight matrix and the fish weight, the global correlation and the set of local correlations of each fish cluster in each fish group of each fish cluster are determined, the fish weight gradient distribution of each fish cluster in each fish cluster of each fish cluster is determined based on the global correlation and the set of local correlations, and the fish weight gradient distribution of each fish cluster in each fish cluster of each fish cluster is formed based on the spatial weight matrix and the fish weight of each fish node in each fish cluster of each fish cluster.

[0044] It should be noted that the Moran index is a statistical index for measuring spatial autocorrelation, and the Moran index is divided into global Moran's I and local Moran's I. The global Moran's I is used to measure the overall degree of spatial autocorrelation in the entire study area, and the local Moran's I is used to identify the spatial autocorrelation pattern of a specific area in the study area. The commonly used calculation formula of the global Moran's I is: The commonly used calculation formula of the local Moran's I is: wherein, GloMor is the global Moran's I, N is the total number of spatial units; W is the sum of all elements of the spatial weight matrix; and is the attribute value of spatial unit and spatial unit respectively; is the mean of all spatial unit attribute values; is the spatial weight of spatial unit and spatial unit respectively; GloMor is the local Moran's I of spatial unit ; is the variance of spatial unit attribute values.

[0045] In an optional embodiment, based on the calculation idea of Moran's I, the embodiment defines the spatial unit as the fish catch node; the attribute value of the spatial unit is the fish catch weight of the fish catch node; for the calculation of the global Moran's I of any one fish catch cluster, the embodiment takes the total number of all fish catch nodes in the fish catch cluster as the total number of spatial units ; then the spatial weight between any two fish catch nodes in the fish catch cluster is accumulated and summed to obtain the sum of all elements of the spatial weight matrix ; the fish catch weights of all fish catch nodes in the fish catch cluster are accumulated and averaged to obtain the mean of all spatial unit attribute values ; then based on the spatial weight matrix calculated above, the spatial weight of spatial unit and spatial unit can be obtained; by substituting these data into the commonly used calculation formula of the global Moran's I, the global Moran's I of any one fish catch cluster can be obtained; for the calculation of the local Moran's I of any one fish catch node in any one fish catch cluster, the embodiment takes the variance of the fish catch weights of all fish catch nodes in the fish catch cluster as the variance of spatial unit attribute values ; the fish catch weights of all fish catch nodes in the fish catch cluster are accumulated and averaged to obtain the mean of all spatial unit attribute values ; then based on the spatial weight matrix calculated above, the spatial weight of spatial unit and spatial unit can be obtained.weight; the data is substituted into the commonly used calculation formula of the global Moran index, and the local Moran index of any fish catch node in the fish catch cluster is obtained; then the global Moran index of the fish catch cluster is taken as the global correlation of the fish catch cluster, and the local Moran index of all the fish catch nodes in the fish catch cluster forms a local correlation set of the fish catch cluster; the weight gradient direction distribution principle of the embodiment is defined as follows: if the numerical value is from large to small, a directed edge is set, and the direction of the directed edge is from the large numerical value to the small numerical value; if the numerical value is the same, a bidirectional edge is set; assuming that the global Moran index of one fish catch cluster is 0.87, and the global Moran index of another fish catch cluster is 0.43, the fish catch cluster weight gradient direction between the two fish catch clusters is set as a directed edge from the fish catch cluster with the global Moran index of 0.87 to the fish catch cluster with the global Moran index of 0.43; at this time, the fish catch cluster weight gradient strength is set as the absolute value of the difference between the two global Moran indexes, that is, the absolute value of the difference between 0.87 and 0.43, which is 0.44. Assuming that the global Moran indexes of the two fish catch clusters are both 0.2, the fish catch cluster weight gradient direction between the two fish catch clusters with the global Moran index of 0.2 is set as a bidirectional edge; at this time, the fish catch cluster weight gradient strength is set as the absolute value of the difference between the two global Moran indexes, that is, the difference between 0.2 and 0.2, which is 0. For the fish catch nodes in the fish catch cluster, the fish catch node weight gradient direction between the fish catch nodes in the same fish catch cluster is constructed by comparing the local Moran indexes of the fish catch nodes and using the weight gradient direction distribution principle. Assuming that the local Moran index of one fish catch node is 0.48, and the local Moran index of another fish catch node is 0.21, the fish catch node weight gradient direction between the two fish catch nodes is set as a directed edge from the fish catch node with the local Moran index of 0.48 to the fish catch node with the local Moran index of 0.21; at this time, the fish catch node weight gradient strength is also the absolute value of the difference between the two local Moran indexes, that is, 0.27. If the local Moran indexes of the two fish catch nodes are both 0.3, the fish catch node weight gradient direction between the two fish catch nodes is set as a bidirectional edge, and the fish catch node weight gradient strength is calculated as 0. In particular, for the fish catch nodes between different fish catch clusters, no fish catch node weight gradient direction and fish catch node weight gradient strength are set between them; then the fish catch cluster weight gradient strength and the fish catch cluster weight gradient direction between the two fish catch clusters, and the fish catch node weight gradient strength and the fish catch node weight gradient direction between the fish catch nodes are integrated to form the fish catch weight gradient distribution of the fish catch cluster.

[0046] The embodiment calculates the global Morant index and the local Morant index by using the spatial weight matrix and the fish catch weight, evaluates the weight spatial correlation strength of each fish catch cluster by using the global Morant index, and quantifies the fish catch weight difference of the fish catch nodes in each fish catch cluster by using the local Morant index, so as to comprehensively mine the variation rule of the fish catch weight in space; then, based on the global Morant index and the local Morant index, the weight gradient strength of the fish catch cluster, the weight gradient direction of the fish catch cluster, the weight gradient strength of the fish catch node, and the weight gradient direction of the fish catch node are determined by using the weight gradient direction distribution principle, so as to ensure the rationality and continuity of the fish catch weight gradient distribution, and then the fish catch weight propagation dynamics in space can be simulated more accurately, so as to improve the accuracy of the weight propagation regional link, and then the accuracy of the fish catch identification is improved.

[0047] In the embodiment, the weight propagation path of each fish catch cluster is constructed by using the spatial weight matrix and the fish catch weight gradient distribution; the fish catch nodes in each fish catch cluster are sequentially connected based on the weight propagation path, and the weight propagation regional link of each fish catch cluster is determined, including: in each fish catch cluster of each fish catch cluster, the fish catch node weight gradient direction between each two fish catch nodes is taken as the fish catch node directed edge direction, the weighted sum of the fish catch node weight gradient strength and the spatial weight between each two fish catch nodes is taken as the fish catch node directed edge weight, the fish catch node connection path of each fish catch cluster of each fish catch cluster is solved by using a preset greedy algorithm with the minimum fish catch node directed edge weight sum as the solving target; in each fish catch cluster, the fish catch cluster weight gradient direction between each two fish catch clusters is taken as the fish catch cluster directed edge direction, and the fish catch cluster weight gradient strength between each two fish catch clusters is taken as the fish catch cluster directed edge weight, the fish catch cluster connection path of each fish catch cluster is solved by using a preset greedy algorithm with the minimum fish catch cluster directed edge weight sum as the solving target; the weight propagation path direction of each fish catch cluster is formed based on the fish catch node connection path and the fish catch cluster connection path; in each fish catch cluster, the fish catch nodes in each fish catch cluster are sequentially connected based on the fish catch node connection path of each fish catch cluster, and the starting fish catch node and the terminal fish catch node of each fish catch cluster are marked; the starting fish catch node or the terminal fish catch node of each two fish catch clusters is connected based on the fish catch cluster connection path of the fish catch cluster, so as to form the weight propagation regional link of each fish catch cluster.

[0048] In an optional embodiment, since the fish catch cluster weight gradient direction and the fish catch cluster weight gradient strength of each two fish catch clusters have been constructed above; and the fish catch cluster weight gradient direction is a directed edge or a bidirectional edge, the fish catch cluster weight gradient direction is taken as the fish catch cluster directed edge direction, and the fish catch cluster weight gradient strength is taken as the fish catch cluster directed edge weight, so as to form a fish catch cluster directed edge weighted graph; at this time, the minimum sum of fish catch cluster directed edge weights is the solving target, and the greedy algorithm is used to solve the fish catch cluster connection path of the fish catch cluster set; similarly, since the fish catch node weight gradient direction, the fish catch node weight gradient strength and the spatial weight of each two fish catch nodes in each fish catch cluster have been constructed, the fish catch node weight gradient direction is taken as the fish catch node directed edge direction of each two fish catch nodes in the fish catch cluster, and the weight of the fish catch node weight gradient strength and the spatial weight is set to be 0.5 respectively; the weighted sum of the fish catch node weight gradient strength and the spatial weight is taken as the fish catch node directed edge weight; the minimum sum of fish catch node directed edge weights is the solving target, and the greedy algorithm is used to solve the fish catch node connection path of each fish catch cluster; based on the fish catch node connection path and the fish catch cluster connection path, the weight propagation path direction is formed; it is assumed that there are four fish catch clusters, which are 、 、 and ; there are three fish catch nodes in the fish catch cluster , , and ; there are three fish catch nodes in the fish catch cluster , , and ; there are two fish catch nodes in the fish catch cluster , , ; there are four fish catch nodes in the fish catch cluster , , , and ; the fish catch cluster connection path obtained by solving is assumed to be ; the fish catch node connection path of the fish catch cluster is , so as to connect the fish catch nodes in the fish catch cluster , the starting fish catch node is , and the terminal fish catch node is ; the fish catch node connection path of the fish catch cluster is , so as to connect the fish catch nodes in the fish catch cluster the fish catch nodes in the fish catch cluster, the starting fish catch node of which is , and the ending fish catch node of which is ; the fish catch node connection path of the fish catch cluster is , thereby connecting the fish catch nodes in the fish catch cluster , the starting fish catch node of which is , and the ending fish catch node of which is ; the fish catch node connection path of the fish catch cluster is , thereby connecting the fish catch nodes in the fish catch cluster , the starting fish catch node of which is , and the ending fish catch node of which is ; and then, according to the fish catch cluster connection path , the starting fish catch node or the ending fish catch node of the fish catch cluster is connected, that is, the ending fish catch node of the fish catch cluster is connected with the starting fish catch node of the fish catch cluster , the ending fish catch node of the fish catch cluster is connected with the starting fish catch node of the fish catch cluster , and the ending fish catch node of the fish catch cluster is connected with the starting fish catch node of the fish catch cluster , so that the weight propagation regional link of the fish catch cluster is .

[0049] In this embodiment, the fish catch node directed edge direction and the fish catch cluster directed edge direction are respectively constructed according to the fish catch cluster weight gradient direction and the fish catch node weight gradient direction in the fish catch weight gradient distribution, so as to ensure the consistency of the fish catch node connection path and the fish catch cluster connection path with the actual weight propagation trend, and make the weight propagation regional link more consistent with the actual fish catch weight of the fish catch; by connecting the fish catch nodes in each fish catch cluster first and then connecting the fish catch nodes between different fish catch clusters, the consistency of the local path and the overall path is ensured, the accuracy and continuity of the weight propagation regional link are improved, and thus the accuracy of subsequent fish catch identification is improved.

[0050] Step S104: identifying the spatial movement mode of each fish catch node in each fish catch cluster based on the fishing time and the fishing location, and constructing the fishing propagation time sequence link of each fish catch cluster based on the spatial movement mode.

[0051] In the embodiment, the spatial movement pattern of each fish catch node in each fish catch cluster is identified based on the fishing time and fishing location, and a fishing propagation time sequence link of each fish catch cluster is constructed based on the spatial movement pattern, including: in each fish catch cluster, a fishing time difference between each two fish catch nodes is calculated based on the fishing time of each two fish catch nodes; a fishing time similarity between each two fish catch nodes is generated by mapping the fishing time difference between each two fish catch nodes based on a preset exponential decay function; a fishing distance between each two fish catch nodes is calculated based on the fishing location of each two fish catch nodes; a distance similarity between each two fish catch nodes is generated by mapping the fishing distance between each two fish catch nodes based on a preset Gaussian kernel function; a fishing similarity between each two fish catch nodes in each fish catch cluster is determined by weighted sum of the time similarity and distance similarity; if the fishing similarity between two fish catch nodes is greater than a preset fishing similarity threshold, a fishing undirected edge is set between the two fish catch nodes, and the spatial movement pattern of the two fish catch nodes is identified as a fixed movement pattern; a fish catch node without fishing undirected edge in the fish catch cluster is identified, and the spatial movement pattern of the fish catch node without fishing undirected edge is identified as a random movement pattern; the fish catch nodes with fixed movement pattern are connected in order of fishing time from small to large to form a fish catch fixed link; a central fishing location of the fish catch fixed link is determined based on the fishing location of each fish catch node in the fish catch fixed link; the distance between each fish catch node with random movement pattern and the central fishing location is calculated, and the fish catch nodes with random movement pattern are connected in order of distance from small to large to form a fish catch random link; the fish catch random link is connected behind the fish catch fixed link to form a fishing propagation time sequence link of each fish catch cluster.

[0052] In an optional embodiment, the fishing time difference between fish catch nodes is calculated based on the fishing time of fish catch nodes, and then substituted into an exponential decay function to map the fishing time difference, thereby generating a time similarity between each two fish catch nodes (the value range is ); similarly, the fishing distance between fish catch nodes is calculated, and the Euclidean distance is used as the fishing distance in the embodiment, and then substituted into a Gaussian kernel function to map the fishing distance, thereby generating a distance similarity between each two fish catch nodes (the value range is the weight of the distance similarity and the time similarity is set as 0.5, the time similarity and the distance similarity are weighted and summed to determine the fishing similarity between each two fish catch nodes; then the fishing similarity threshold is set as 0.5; if the fishing similarity is greater than 0.5, a fishing undirected edge is set between the two fish catch nodes, and the spatial movement mode of the two fish catch nodes is identified as a fixed movement mode; then the fish catch nodes that are not connected by any fishing undirected edge are set to have a spatial movement mode identified as a random movement mode; then the fish catch nodes with a fixed movement mode are connected in the order of fishing time from small to large to form a fish catch fixed link; then the center of the fishing locations (i.e. the centroid) of the fish catch nodes (i.e. the fish catch nodes with a fixed movement mode) in the fish catch fixed link is calculated based on the fishing locations (i.e. the latitude and longitude coordinates) of the fish catch nodes, to obtain a center fishing location; then the distance (here still using the Euclidean distance) between each fish catch node with a random movement mode and the center fishing location is calculated, and the fish catch nodes with a random movement mode are connected in the order of distance from small to large to form a fish catch random link; then the fish catch random link is connected behind the fish catch fixed link to form a fishing propagation time sequence link of the fish catch cluster; assuming that there are 、 、 、 、 、 、 、 、 ; the fish catch nodes with a random movement mode are 、 、 ; the fish catch fixed link is ; the fish catch random link is ; and thus the fishing propagation time sequence link of the fish catch cluster is .

[0053] The embodiment can realize similarity comparison of different fish catch nodes in time dimension and place dimension by calculating the fishing time difference and the fishing distance and converting them into time similarity and distance similarity respectively; can comprehensively and accurately evaluate the relevance between different fish catch nodes in fishing operation by obtaining the fishing similarity through weighted summation; can identify the spatial movement mode of the fish catch nodes through the fishing similarity threshold, and the fish catch fixed link reflects the long-term trend of the fishing operation, and the fish catch random link reflects the mutation characteristics of the fishing operation; so that the fishing propagation time sequence link can accurately reflect the actual working condition of the fishing operation; and further improve the accuracy of the subsequent multi-source propagation time sequence link and the accuracy of the intelligent identification of the fish catch.

[0054] Step S105: constructing a similarity matching matrix between the weight propagation area link and the fishing propagation time sequence link of each fish catch cluster, and based on the similarity matching matrix, filtering and fusing fish catch nodes from the weight propagation area link and the fishing propagation time sequence link respectively to form a multi-source fish catch node, and further determining a multi-source propagation time sequence link of each fish catch cluster.

[0055] In the embodiment, the similarity matching matrix between the weight propagation area link and the fishing propagation time sequence link of each fishery cluster is constructed, and based on the similarity matching matrix, fishery nodes are screened from the weight propagation area link and the fishing propagation time sequence link respectively to be fused to form multi-source fishery nodes, and then the multi-source propagation time sequence link of each fishery cluster is determined, including: in each fishery cluster, the position sequence number of any fishery node in the weight propagation area link is subtracted from the position sequence number of any fishery node in the fishing propagation time sequence link, and the absolute value is taken, and then divided by the total number of fishery nodes to determine the link position cost of the two fishery nodes; the fishery weight of any fishery node in the weight propagation area link is subtracted from the fishery weight of any fishery node in the fishing propagation time sequence link, and the absolute value is taken, and then divided by the maximum value of the fishery weight of the two fishery nodes to determine the fishery weight cost of the two fishery nodes; the Euclidean distance between the fishing site of any fishery node in the weight propagation area link and the fishing site of any fishery node in the fishing propagation time sequence link is calculated as the fishing distance cost of the two fishery nodes; the link position cost, the fishery weight cost and the fishing distance cost of the two fishery nodes are weighted and summed to obtain the similarity matching cost between any fishery node in the weight propagation area link and any fishery node in the fishing propagation time sequence link; based on the similarity matching cost of any fishery node in the weight propagation area link and any fishery node in the fishing propagation time sequence link, the similarity matching matrix between the weight propagation area link and the fishing propagation time sequence link of each fishery cluster is constructed; according to a preset link matching algorithm, the similarity matching matrix is solved to construct an optimal fishery node matching matrix of each fishery cluster; based on the optimal fishery node matching matrix, fishery nodes are screened from the weight propagation area link and the fishing propagation time sequence link respectively, the position sequence numbers of the two screened fishery nodes are weighted and summed to obtain the weighted average position of the multi-source fishery node; the fishery weight, the fishing time and the fishing site of the two screened fishery nodes are weighted and summed respectively to form the weighted fishery weight, the weighted fishing time and the weighted fishing site of the multi-source fishery node respectively; the multi-source fishery nodes are sorted in ascending order of the weighted average position, and the multi-source fishery nodes are connected in turn to determine the multi-source propagation time sequence link of each fishery cluster.

[0056] In an optional embodiment, the weight propagation area link is ; the fishing propagation time sequence link is ; at this time, define any one of the fishery nodes in the weight propagation area link as , and any one of the fishery nodes in the fishing propagation time sequence link as ; subtract the position serial number of the fishery node from the position serial number of the fishery node , take the absolute value, and divide by the absolute value of the total number of fishery nodes, and the link position cost is obtained; for example, assuming that the fishery node is , and its position serial number is 9; the fishery node is , and its position serial number is 3; the total number of fishery nodes is 12, so the link position cost is 0.5; subtract the fishery weight of the fishery node from the fishery weight of the fishery node , take the absolute value, and divide by the maximum value of the fishery weight of the two fishery nodes, to determine the fishery weight cost of the two fishery nodes; assuming that the fishery weight of the fishery node is 0.3 kg, and the fishery weight of the fishery node is 0.5 kg; the maximum value of the fishery weight of the two is 0.5 kg; therefore, the fishery weight cost is 0.4; then calculate the Euclidean distance between the fishing locations of the fishery node and the fishery node as the fishing distance cost; set the weight of the link position cost to 0.4, the weight of the fishery weight cost to 0.5, and the weight of the fishing distance cost to 0.1, and perform a weighted sum of the link position cost, the fishery weight cost and the fishing distance cost to obtain the similarity matching cost; repeat the above operation to calculate the similarity matching cost between any one of the fishery nodes in the weight propagation area link and any one of the fishery nodes in the fishing propagation time sequence link, and construct a similarity matching matrix; the rows of the similarity matching matrix are set to the fishery nodes of the weight propagation area link, and the columns of the similarity matching matrix are set to the fishery nodes of the fishing propagation time sequence link; for example, the first row of the similarity matching matrix corresponds to the first fishery node of the weight propagation area link; the second column of the similarity matching matrix corresponds to the second fishery node of the fishing propagation time sequence link;

[0057] The link matching algorithm is set as the Hungarian Algorithm, the Hungarian Algorithm is used to iterate the similarity matching matrix, so that the optimal fish catch node matching matrix has multiple zero elements; then, based on the row corresponding to the zero element in the optimal fish catch node matching matrix, the corresponding fish catch node is screened out from the weight propagation area link, based on the column corresponding to the zero element in the optimal fish catch node matching matrix, the corresponding fish catch node is screened out from the fishing propagation time sequence link, the unknown serial number is weighted and summed to obtain the average (the weight is set to 0.5), and the weighted average position of the multi-source fish catch node is obtained; similarly, the fish catch weight, fishing time and fishing place of the two screened fish catch nodes are weighted and summed to obtain the average (the weight is set to 0.5), and the weighted fish catch weight, weighted fishing time and weighted fishing place of the multi-source fish catch node are obtained; after obtaining the weighted average position, weighted fish catch weight, weighted fishing time and weighted fishing place, the multi-source fish catch node is formed; then, the multi-source fish catch node is sorted in the order of the weighted average position from small to large and then connected in turn, and the multi-source propagation time sequence link is formed.

[0058] In the embodiment, the link position cost, fish catch weight cost and fishing distance cost of the fish catch nodes in the weight propagation area link and the fishing propagation time sequence link are identified, the differences of the fish catch nodes in the link, weight and place are comprehensively considered, the representation accuracy and comprehensiveness of the similarity matching matrix are improved, then the similarity matching matrix is solved by the link matching algorithm, so that the optimal fish catch node matching matrix can realize accurate matching between the weight propagation area link and the fishing propagation time sequence link, then the weighted average position, weighted fish catch weight, weighted fishing time and weighted fishing place of the source fish catch node are calculated by the weighted method, the effective fusion of the links in the weight and time sequence dimensions is realized, the fishing operation mode and the fish catch life characteristics can be fully coupled, and the accuracy and comprehensiveness of subsequent fish catch identification are improved.

[0059] Step S106: based on the multi-source propagation time sequence link, a fish catch multi-source data relationship graph of each fish catch cluster is constructed; graph feature extraction is performed on the fish catch multi-source data relationship graph to obtain a fish catch intelligent identification result.

[0060] In the embodiment, the fishery multi-source data relationship graph of each fishery cluster is constructed based on the multi-source propagation time sequence link; and the fishery intelligent recognition result is obtained by performing graph feature extraction on the fishery multi-source data relationship graph, including: constructing the fishery multi-source data relationship graph of each fishery cluster based on the multi-source propagation time sequence link; dividing the fishery multi-source data relationship graph into regions according to a preset label propagation algorithm to determine the fishing hotspot region of each fishery species; identifying the abnormal multi-source fishery node in the fishery multi-source data relationship graph according to a preset isolation forest algorithm, and forming the abnormal fishery catch operation event of each fishery species based on the weighted fishery weight, the weighted fishing time and the weighted fishing location of the abnormal multi-source fishery node; and forming the fishery intelligent recognition result based on the fishing hotspot region and the abnormal fishery catch operation event of each fishery species.

[0061] In an optional embodiment, the multi-source propagation time sequence link includes the multi-source fishery nodes and the connection between the multi-source fishery nodes, so that it can be directly used as the fishery multi-source data relationship graph; the label propagation algorithm can identify the hotspot region of the fishery multi-source data relationship graph, i.e., the fishing hotspot region of the fishery species; the Louvain algorithm, Girvan-Newman algorithm and the like can also be used to identify the hotspot region of the fishery multi-source data relationship graph, which will not be described in detail in the embodiment. The isolation forest algorithm can identify the abnormal multi-source fishery node as the abnormal fishery catch operation event; since the related algorithm for graph feature extraction has been relatively maturely applied, the embodiment will not be expanded in detail; in particular, the skilled person can select different graph feature extraction algorithms to extract the features of the fishery multi-source data relationship graph according to the actual fishery recognition requirements, so as to realize the fishery intelligent recognition.

[0062] The embodiment identifies the fishing hotspot region of each fishery species by the label propagation algorithm, so as to identify the popular fishery catch region in the fishery catch operation process, and identifies the abnormal fishery catch operation event in the fishery catch operation process by the isolation forest algorithm, so as to realize the comprehensive and accurate recognition of the fishery, and make the fishery recognition provide the basis for formulating the fishing prohibition period, the fishing quota and the like fishery resource operation, and realize the sustainable development and management of the fishery resources.

[0063] Embodiment Two

[0064] Please refer to Figure 2 , Figure 2A structural schematic diagram of a fish catch intelligent identification system provided for an embodiment of the application, comprising: a fish catch species identification module 201, a fish catch cluster construction module 202, a weight propagation region link construction module 203, a fishing propagation time sequence link construction module 204, a multi-source propagation time sequence link construction module 205, and a graph feature extraction module 206;

[0065] The fish catch species identification module 201 is configured to obtain a plurality of fish catch images, and the fishing time, the fish catch weight, and the fishing location corresponding to each fish catch image, and then construct a plurality of fish catch species multi-source identification vectors; based on a preset fish catch species identification neural network model, the fish catch species multi-source identification vectors are subjected to feature recognition to obtain the fish catch species corresponding to each fish catch species multi-source identification vector; the fish catch cluster construction module 202 is configured to initialize a plurality of fish catch clusters based on the fish catch species and the fish catch species multi-source identification vectors, each fish catch cluster containing a plurality of fish catch nodes of the same fish catch species; the weight propagation region link construction module 203 is configured to construct a spatial weight matrix based on the fishing location; the fish catch weight is subjected to autocorrelation analysis to determine the fish catch weight gradient distribution; the weight propagation path is constructed with the spatial weight matrix and the fish catch weight gradient distribution, and the fish catch nodes in the fish catch cluster are sequentially connected to determine the weight propagation region link of each fish catch cluster; the fishing propagation time sequence link construction module 204 is configured to identify the spatial movement mode of each fish catch node in each fish catch cluster based on the fishing time and the fishing location, and construct the fishing propagation time sequence link of each fish catch cluster based on the spatial movement mode; the multi-source propagation time sequence link construction module 205 is configured to construct a similarity matching matrix between the weight propagation region link and the fishing propagation time sequence link of each fish catch cluster, and based on the similarity matching matrix, fish catch nodes are screened from the weight propagation region link and the fishing propagation time sequence link respectively for fusion to form multi-source fish catch nodes, and then the multi-source propagation time sequence link of each fish catch cluster is determined; the graph feature extraction module 206 is configured to construct the fish catch multi-source data relationship graph of each fish catch cluster based on the multi-source propagation time sequence link; the fish catch multi-source data relationship graph is subjected to graph feature extraction to obtain the fish catch intelligent identification result.

[0066] In the embodiment, the fish catch species identification module 201 comprises a fish catch species identification unit. The fish catch species identification unit is configured to acquire a plurality of fish catch images in a fish catch operation process, and a corresponding fishing time, fish catch weight and fishing location of each fish catch image. Each pixel point in each fish catch image is taken as a neighborhood block center point, and a neighborhood block is assigned to each pixel point in each fish catch image. Each fish catch image is converted to grayscale based on the neighborhood block, and a grayscale fish catch image of each fish catch image is obtained. The contour of each grayscale fish catch image is identified based on a preset edge detection algorithm, and a fish catch contour point set of each fish catch image is determined. The fish catch area and fish catch contour range of each fish catch image are determined based on the fish catch contour point set. The RGB color of a pixel point in the fish catch contour range of each fish catch image is weighted and averaged, and a fish catch color value of each fish catch image is determined. A fish catch species multi-source identification vector corresponding to each fish catch image is constructed based on the fish catch area, fish catch color value, fishing time, fish catch weight and fishing location of each fish catch image. Feature recognition is performed on the fish catch species multi-source identification vector based on a preset fish catch species identification neural network model, and a plurality of fish catch species identification candidate results corresponding to each fish catch species multi-source identification vector are obtained. The fish catch species identification candidate results are filtered based on pre-acquired historical fish catch operation data, and a fish catch species corresponding to each fish catch species multi-source identification vector is obtained.

[0067] In the embodiment, the fish catch species identification unit comprises a grayscale fish catch image acquisition subunit. Each pixel point in each fish catch image is taken as a neighborhood block center point, and a neighborhood block is assigned to each pixel point in each fish catch image. In each neighborhood block, the pixel point grayscale mean and pixel point grayscale standard deviation of all pixel points in the neighborhood block are calculated. A grayscale weight is assigned to each pixel point in the neighborhood block based on the pixel point grayscale mean and pixel point grayscale standard deviation and in combination with a preset Gaussian function. A grayscale threshold of the neighborhood block is calculated based on the grayscale weight and grayscale value of each pixel point in the neighborhood block. If the grayscale threshold is greater than the grayscale value of the neighborhood block center point, the grayscale conversion value of the pixel point corresponding to the neighborhood block center point is set as a preset first grayscale value. If the grayscale threshold is less than or equal to the grayscale value of the neighborhood block center point, the grayscale conversion value of the pixel point corresponding to the neighborhood block center point is set as a preset second grayscale value. Each fish catch image is converted to grayscale based on the grayscale conversion value of each pixel point in each fish catch image, and a grayscale fish catch image of each fish catch image is obtained.

[0068] In the embodiment, the weight propagation regional link construction module 203 comprises a weight propagation regional link construction unit. The weight propagation regional link construction unit is configured to: in each fish catch cluster, perform cluster analysis on the fish catch nodes based on the fishing locations of the fish catch nodes, to obtain a plurality of fish catch clusters of each fish catch cluster; if two fish catch nodes belong to the same fish catch cluster, calculate the Euclidean distance between the fishing locations of the two fish catch nodes as the spatial weight of the two fish catch nodes; if two fish catch nodes do not belong to the same fish catch cluster, set the spatial weight of the two fish catch nodes as a preset negative value; based on the spatial weight of each two fish catch nodes in each fish catch cluster, form a spatial weight matrix of each fish catch cluster; based on the spatial weight matrix and the fish catch weight, perform autocorrelation analysis on the fish catch weight of each fish catch cluster to determine the global correlation and the local correlation set of each fish catch cluster; determine the fish catch weight gradient distribution of each fish catch cluster based on the global correlation and the local correlation set; construct the weight propagation path of each fish catch cluster based on the spatial weight matrix and the fish catch weight gradient distribution; and sequentially connect the fish catch nodes in each fish catch cluster based on the weight propagation path, to determine the weight propagation regional link of each fish catch cluster.

[0069] In the embodiment, the weight propagation area link construction unit comprises a fish catch weight gradient distribution acquisition subunit. The fish catch weight gradient distribution acquisition subunit is configured to calculate a global Moran's index of each fish catch cluster based on the spatial weight matrix and the fish catch weight of each fish catch node in each fish catch cluster of each fish catch cluster; calculate a local Moran's index of each fish catch node in each fish catch cluster of each fish catch cluster based on the spatial weight matrix and the fish catch weight of each fish catch node in each fish catch cluster of each fish catch cluster; take the global Moran's index of each fish catch cluster as a global correlation of each fish catch cluster; construct a local correlation set of each fish catch cluster of each fish catch cluster based on the local Moran's index of all fish catch nodes in each fish catch cluster; in each fish catch cluster, determine a fish catch cluster weight gradient direction between each two fish catch clusters based on the size of the global Moran's index of each fish catch cluster and in combination with a preset weight gradient direction allocation principle; determine a fish catch cluster weight gradient strength between each two fish catch clusters based on the absolute value of the difference between the global Moran's indexes of each two fish catch clusters; determine a fish catch node weight gradient direction between each two fish catch nodes in each fish catch cluster based on the size of the local Moran's index of each fish catch node in each fish catch cluster and in combination with the preset weight gradient direction allocation principle; determine a fish catch node weight gradient strength between each two fish catch nodes based on the absolute value of the difference between the local Moran's indexes of each two fish catch nodes; and form a fish catch weight gradient distribution of each fish catch cluster based on the fish catch cluster weight gradient strength and the fish catch cluster weight gradient direction between each two fish catch clusters of each fish catch cluster and the fish catch node weight gradient strength and the fish catch node weight gradient direction between each two fish catch nodes in each fish catch cluster of each fish catch cluster.

[0070] In the embodiment, the weight propagation regional link construction unit comprises a weight propagation regional link construction subunit; the weight propagation regional link construction subunit is used for taking the weight gradient direction between each two fishery nodes in each fishery cluster as the fishery node directed edge direction, taking the weighted sum of the weight gradient strength and the spatial weight between each two fishery nodes as the fishery node directed edge weight, taking the minimum of the fishery node directed edge weight sum as the solving target, and using a preset greedy algorithm to solve the fishery node connection path of each fishery cluster; in each fishery cluster, the weight gradient direction between each two fishery clusters is taken as the fishery cluster directed edge direction, the weight gradient strength between each two fishery clusters is taken as the fishery cluster directed edge weight, the minimum of the fishery cluster directed edge weight sum is taken as the solving target, and a preset greedy algorithm is used to solve the fishery cluster connection path of each fishery cluster; based on the fishery node connection path and the fishery cluster connection path, the weight propagation path direction of each fishery cluster is formed; in each fishery cluster, based on the fishery node connection path of each fishery cluster, the fishery nodes in each fishery cluster are sequentially connected, and the starting fishery node and the terminal fishery node of each fishery cluster are marked; based on the fishery cluster connection path of the fishery cluster, the starting fishery node or the terminal fishery node of each two fishery clusters is connected to form the weight propagation regional link of each fishery cluster.

[0071] In the embodiment, the fishing propagation time sequence link construction module 204 comprises a fishing propagation time sequence link construction unit. The fishing propagation time sequence link construction unit is configured to: in each of the fish catch cluster, calculate a fishing time difference between each two fish catch nodes based on fishing times of the fish catch nodes; map the fishing time difference between each two fish catch nodes based on a preset exponential decay function to generate a time similarity between each two fish catch nodes; calculate a fishing distance between each two fish catch nodes based on fishing locations of the fish catch nodes; map the fishing distance between each two fish catch nodes based on a preset Gaussian kernel function to generate a distance similarity between each two fish catch nodes; perform weighted summation on the time similarity and the distance similarity to determine a fishing similarity between each two fish catch nodes in the fish catch cluster; if the fishing similarity between two fish catch nodes is greater than a preset fishing similarity threshold, set a fishing undirected edge between the two fish catch nodes, and identify a spatial movement mode of the two fish catch nodes as a fixed movement mode; identify fish catch nodes without fishing undirected edges in the fish catch cluster, and identify a spatial movement mode of the fish catch nodes without fishing undirected edges as a random movement mode; connect the fish catch nodes with the fixed movement mode in an order of fishing times from small to large to form a fish catch fishing fixed link; determine a central fishing location of the fish catch fishing fixed link based on fishing locations of each fish catch node in the fish catch fishing fixed link; calculate distances between each fish catch node with the random movement mode and the central fishing location, and sequentially connect the fish catch nodes with the random movement mode in an order of distances from small to large to form a fish catch fishing random link; and connect the fish catch fishing random link behind the fish catch fishing fixed link to form a fishing propagation time sequence link of each fish catch cluster.

[0072] In the embodiment, the multi-source propagation time sequence link construction module 205 comprises a multi-source propagation time sequence link construction unit. The multi-source propagation time sequence link construction unit is configured to: in each fish catch cluster, determine a link position cost of two fish catch nodes by subtracting a position sequence number of any one fish catch node in the weight propagation regional link from a position sequence number of any one fish catch node in the fishing propagation time sequence link and taking an absolute value, and dividing by a total number of fish catch nodes; determine a fish catch weight cost of the two fish catch nodes by subtracting a fish catch weight of any one fish catch node in the weight propagation regional link from a fish catch weight of any one fish catch node in the fishing propagation time sequence link and taking an absolute value, and dividing by a maximum value of the fish catch weights of the two fish catch nodes; calculate a fishing distance cost of the two fish catch nodes by calculating a Euclidean distance between a fishing location of any one fish catch node in the weight propagation regional link and a fishing location of any one fish catch node in the fishing propagation time sequence link; perform weighted summation on the link position cost, the fish catch weight cost, and the fishing distance cost of the two fish catch nodes to obtain a similarity matching cost between any one fish catch node in the weight propagation regional link and any one fish catch node in the fishing propagation time sequence link; construct a similarity matching matrix between the weight propagation regional link and the fishing propagation time sequence link of each fish catch cluster based on the similarity matching cost of any one fish catch node in the weight propagation regional link and any one fish catch node in the fishing propagation time sequence link; solve the similarity matching matrix according to a preset link matching algorithm to construct an optimal fish catch node matching matrix of each fish catch cluster; based on the optimal fish catch node matching matrix, filter fish catch nodes from the weight propagation regional link and the fishing propagation time sequence link respectively, perform weighted summation and averaging on the position sequence numbers of the two filtered fish catch nodes to obtain a weighted average position of a multi-source fish catch node, and perform weighted summation and averaging on the fish catch weights, fishing times, and fishing locations of the two filtered fish catch nodes respectively to form a weighted fish catch weight, a weighted fishing time, and a weighted fishing location of the multi-source fish catch node respectively; sort the multi-source fish catch nodes in ascending order of the weighted average position, and sequentially connect the multi-source fish catch nodes to determine a multi-source propagation time sequence link of each fish catch cluster.

[0073] In the embodiment, the graph feature extraction module 206 comprises a graph feature extraction unit; the graph feature extraction unit is configured to construct a fish catch multi-source data relationship graph of each fish catch cluster based on the multi-source propagation time sequence link; perform regional division on the fish catch multi-source data relationship graph according to a preset label propagation algorithm, and determine a fishing hotspot area of each fish catch species; identify an abnormal multi-source fish catch node in the fish catch multi-source data relationship graph according to a preset isolation forest algorithm, form an abnormal fish catch operation event of each fish catch species based on a weighted fish catch weight, a weighted fishing time and a weighted fishing location of the abnormal multi-source fish catch node, and form a fish catch intelligent recognition result based on the fishing hotspot area of each fish catch species and the abnormal fish catch operation event.

[0074] In summary, the embodiment of the present application constructs a fish catch species multi-source identification vector containing multi-source data by using a fish catch image, a fish catch weight, a fishing time and a fishing location, thereby avoiding the inaccuracy of a traditional neural network model in predicting a fish catch species from a single image dimension; then a fish catch cluster is initialized and fish catch nodes in the fish catch cluster are made to have the same fish catch species, so that the recognition result of the fish catch can be more targeted to fish catches of the same type; then a spatial weight matrix is constructed based on the fishing location, a self-correlation analysis is performed on the fish catch weight, and a fish catch weight gradient distribution is determined, thereby determining a weight propagation path, constructing a weight propagation area link and realizing coupled analysis of the fish catch in the weight dimension and the active area dimension; then a spatial movement pattern of the fish catch node is obtained based on the fishing time and the fishing location, thereby constructing a fishing propagation time sequence link and realizing coupled analysis of the fish catch in the active area dimension and the fish catch operation time dimension; the weight propagation area link and the fishing propagation time sequence link are fused by constructing a similarity matching matrix to generate a multi-source propagation time sequence link, so that the fish catch multi-source data relationship graph generated based on the multi-source propagation time sequence link can comprehensively couple the fishing operation mode and the fish catch life characteristics; then the relationship between the fish catch operation and the fish catch habit is mined by graph feature extraction, the intelligent recognition of the fish catch during the fishing operation process is realized, the comprehensiveness and accuracy of the fish catch recognition are improved, the fish catch recognition can provide a basis for formulating a fishing ban period, a fishing quota and other fishery resource operations, and the sustainable development and management of fishery resources are realized.

[0075] The above-described specific embodiments further illustrate the purpose, technical solutions and advantages of the present application, and it should be understood that the above-described embodiments are merely specific embodiments of the present application and are not intended to limit the protection scope of the present application. It should be particularly pointed out that any modifications, equivalent replacements, improvements, etc. made by those skilled in the art within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A fish catch intelligent recognition method, characterized in that, The method comprises the following steps: obtaining a plurality of fish catch images, and the corresponding fishing time, fish catch weight and fishing location of each fish catch image, and then constructing a plurality of fish catch species multi-source identification vectors; based on a preset fish catch species identification neural network model, performing feature identification on the fish catch species multi-source identification vectors to obtain the fish catch species corresponding to each fish catch species multi-source identification vector; based on the fish catch species and the fish catch species multi-source identification vector, initializing a plurality of fish catch clusters, each fish catch cluster containing a plurality of fish catch nodes of the same fish catch species; based on the fishing location, constructing a spatial weight matrix; performing autocorrelation analysis on the fish catch weight to determine the fish catch weight gradient distribution; constructing a weight propagation path based on the spatial weight matrix and the fish catch weight gradient distribution, sequentially connecting the fish catch nodes in the fish catch cluster to determine the weight propagation regional link of each fish catch cluster; based on the fishing time and the fishing location, identifying the spatial movement mode of each fish catch node in each fish catch cluster, and based on the spatial movement mode, constructing the fishing propagation time sequence link of each fish catch cluster; constructing a similarity matching matrix between the weight propagation regional link and the fishing propagation time sequence link of each fish catch cluster, and based on the similarity matching matrix, filtering and fusing fish catch nodes from the weight propagation regional link and the fishing propagation time sequence link respectively to form multi-source fish catch nodes, and then determining the multi-source propagation time sequence link of each fish catch cluster; based on the multi-source propagation time sequence link, constructing a fish catch multi-source data relationship graph for each fish catch cluster; performing graph feature extraction on the fish catch multi-source data relationship graph to obtain a fish catch intelligent identification result; the method comprises the following steps: based on the fishing location, constructing a spatial weight matrix; performing autocorrelation analysis on the fish catch weight to determine the fish catch weight gradient distribution; constructing a weight propagation path based on the spatial weight matrix and the fish catch weight gradient distribution, sequentially connecting the fish catch nodes in the fish catch cluster to determine the weight propagation regional link of each fish catch cluster, comprising: in each fish catch cluster, based on the fishing location of the fish catch node, performing cluster analysis on the fish catch node to obtain a plurality of fish catch clusters in each fish catch cluster; if two fish catch nodes belong to the same fish catch cluster, the Euclidean distance of the fishing locations of the two fish catch nodes is calculated as the spatial weight of the two fish catch nodes; if two fish catch nodes do not belong to the same fish catch cluster, the spatial weight of the two fish catch nodes is set to a preset negative value; based on the spatial weight of each two fish catch nodes in each fish catch cluster, a spatial weight matrix of each fish catch cluster is formed; perform autocorrelation analysis on the fish weight of each fish cluster of each fish catch cluster based on the spatial weight matrix and the fish weight, to determine a global correlation and a set of local correlations of each fish cluster of each fish catch cluster; determine a fish weight gradient distribution of each fish catch cluster based on the global correlation and the set of local correlations; construct a weight propagation path of each fish catch cluster based on the spatial weight matrix and the fish weight gradient distribution; sequentially connect the fish nodes in each fish catch cluster based on the weight propagation path to determine a weight propagation regional link of each fish catch cluster.

2. The fish capture smart identification method of claim 1, wherein, The fish catch image and the corresponding fishing time, fish weight and fishing location of each fish catch image are obtained, and a fish catch species multi-source identification vector is constructed; a fish catch species is obtained corresponding to each fish catch species multi-source identification vector by performing feature recognition on the fish catch species multi-source identification vector based on a preset fish catch species identification neural network model, including: obtaining a plurality of fish catch images during the fish catch operation process, and the fishing time, fish weight and fishing location corresponding to each fish catch image; each pixel point in each fish catch image is taken as a neighborhood block center point, and a neighborhood block is assigned to each pixel point in each fish catch image, and a gray scale conversion is performed on each fish catch image based on the neighborhood block to obtain a gray scale fish catch image of each fish catch image; based on a preset edge detection algorithm, the contour of each fish catch image is identified based on the gray scale fish catch image to determine a fish catch contour point set of each fish catch image; based on the fish catch contour point set, the fish catch area and the fish catch contour range of each fish catch image are determined; the RGB color of the pixel points in the fish catch contour range of each fish catch image is weighted and averaged to determine the fish catch color value of each fish catch image; based on the fish catch area, the fish catch color value, the fishing time, the fish weight and the fishing location of each fish catch image, a fish catch species multi-source identification vector corresponding to each fish catch image is constructed; based on a preset fish catch species identification neural network model, feature recognition is performed on the fish catch species multi-source identification vector to obtain a plurality of fish catch species identification candidate results corresponding to each fish catch species multi-source identification vector; the fish catch species identification candidate results are filtered based on the pre-acquired historical fish catch operation data to obtain the fish catch species corresponding to each fish catch species multi-source identification vector.

3. The fish capture smart identification method of claim 2, wherein, each pixel point in each fish catch image is taken as a neighborhood block center point, and a neighborhood block is assigned to each pixel point in each fish catch image, and a gray scale conversion is performed on each fish catch image based on the neighborhood block to obtain a gray scale fish catch image of each fish catch image, including: each pixel point in each fish catch image is taken as a neighborhood block center point, and a neighborhood block is assigned to each pixel point in each fish catch image; In each of the neighborhood blocks, the average pixel gray value and the standard deviation of pixel gray value of all the pixel points in the neighborhood block are calculated; Based on the average pixel gray value and the standard deviation of pixel gray value, a preset Gaussian function is combined to assign a gray weight to each pixel point in the neighborhood block; Based on the gray weight and the gray value of each pixel point in the neighborhood block, the gray threshold value of the neighborhood block is calculated; If the gray threshold value is greater than the gray value of the center point of the neighborhood block, the gray conversion value of the pixel point corresponding to the center point of the neighborhood block is set as a preset first gray value; If the gray threshold value is less than or equal to the gray value of the center point of the neighborhood block, the gray conversion value of the pixel point corresponding to the center point of the neighborhood block is set as a preset second gray value; Based on the gray conversion value of each pixel point of each fish catch image, the gray conversion of each fish catch image is performed to obtain a gray fish catch image of each fish catch image.

4. The fish capture smart identification method of claim 1, wherein, Based on the spatial weight matrix and the fish catch weight, the fish catch weight of each fish catch cluster is subjected to autocorrelation analysis to determine the global correlation and the local correlation set of each fish catch cluster; Based on the global correlation and the local correlation set, the fish catch weight gradient distribution of each fish catch cluster is determined, including: Based on the spatial weight matrix and the fish catch weight of each fish catch node in each fish catch cluster, the global Moran index of each fish catch cluster is calculated; Based on the spatial weight matrix and the fish catch weight of each fish catch node in each fish catch cluster, the local Moran index of each fish catch node in each fish catch cluster is calculated; The global Moran index of each fish catch cluster is taken as the global correlation of each fish catch cluster, and the local correlation set of each fish catch cluster of each fish catch cluster is constructed based on the local Moran index of all the fish catch nodes in each fish catch cluster; In each fish catch cluster, based on the size of the global Moran index of each fish catch cluster, a preset weight gradient direction allocation principle is combined to determine the fish catch cluster weight gradient direction between each two fish catch clusters; Based on the absolute value of the difference value of the global Moran index of each two fish catch clusters, the fish catch cluster weight gradient strength between each two fish catch clusters is determined; Based on the size of the local Moran index of each fish catch node in each fish catch cluster, a preset weight gradient direction allocation principle is combined to determine the fish catch node weight gradient direction between each two fish catch nodes in each fish catch cluster; Based on the absolute value of the difference value of the local Moran index of each two fish catch nodes, the fish catch node weight gradient strength between each two fish catch nodes is determined; form a fish weight gradient distribution of each of the fish catch clusters based on the fish catch cluster weight gradient strength and the fish catch cluster weight gradient direction between each two of the fish catch clusters of each of the fish catch clusters, and the fish catch node weight gradient strength and the fish catch node weight gradient direction between each two of the fish catch nodes within each of the fish catch clusters.

5. The fish capture smart identification method of claim 4, wherein, construct a weight propagation path of each of the fish catch clusters based on the spatial weight matrix and the fish weight gradient distribution; connect the fish catch nodes within each of the fish catch clusters in sequence based on the weight propagation path, and determine a weight propagation regional link of each of the fish catch clusters, including: in each of the fish catch clusters, take the fish catch node weight gradient direction between each two of the fish catch nodes as a fish catch node directed edge direction, and take a weighted sum of the fish catch node weight gradient strength and the spatial weight between each two of the fish catch nodes as a fish catch node directed edge weight, and use a preset greedy algorithm to solve a fish catch node connection path of each of the fish catch clusters as a solution target with a minimum fish catch node directed edge weight sum; in each of the fish catch clusters, take the fish catch cluster weight gradient direction between each two of the fish catch clusters as a fish catch cluster directed edge direction, and take the fish catch cluster weight gradient strength between each two of the fish catch clusters as a fish catch cluster directed edge weight, and use a preset greedy algorithm to solve a fish catch cluster connection path of each of the fish catch clusters as a solution target with a minimum fish catch cluster directed edge weight sum; form a weight propagation path direction of each of the fish catch clusters based on the fish catch node connection path and the fish catch cluster connection path; in each of the fish catch clusters, connect the fish catch nodes within each of the fish catch clusters in sequence based on the fish catch node connection path of each of the fish catch clusters, and mark a starting fish catch node and a terminal fish catch node of each of the fish catch clusters; connect the starting fish catch node or the terminal fish catch node of each two of the fish catch clusters based on the fish catch cluster connection path of the fish catch clusters to form a weight propagation regional link of each of the fish catch clusters.

6. The fish capture smart identification method of claim 1, wherein, identify a spatial movement pattern of each of the fish catch nodes within each of the fish catch clusters based on the fishing time and the fishing location, and construct a fishing propagation time sequence link of each of the fish catch clusters based on the spatial movement pattern, including: in each of the fish catch clusters, calculate a fishing time difference between each two of the fish catch nodes based on the fishing time of each two of the fish catch nodes, and map the fishing time difference between each two of the fish catch nodes based on a preset exponential decay function to generate a time similarity between each two of the fish catch nodes; calculate a fishing distance between each two of the fish catch nodes based on the fishing location of each two of the fish catch nodes, and map the fishing distance between each two of the fish catch nodes based on a preset Gaussian kernel function to generate a distance similarity between each two of the fish catch nodes; calculate a fishing distance between each two of the fish catch nodes based on the fishing location of each two of the fish catch nodes, and map the fishing distance between each two of the fish catch nodes based on a preset Gaussian kernel function to generate a distance similarity between each two of the fish catch nodes; weighting and summing the time similarity and the distance similarity to determine a fishing similarity between each two fishery nodes in each fishery cluster; if the fishing similarity between two fishery nodes is greater than a preset fishing similarity threshold, setting a fishing undirected edge between the two fishery nodes, and identifying the spatial movement mode of the two fishery nodes as a fixed movement mode; identifying fishery nodes without fishing undirected edges in the fishery cluster, and identifying the spatial movement mode of the fishery nodes without fishing undirected edges as a random movement mode; connecting fishery nodes with fixed movement modes in order of fishing time from small to large to form fishery fishing fixed links; determining a central fishing location of the fishery fishing fixed links based on the fishing locations of each fishery node in the fishery fishing fixed links; calculating the distance between each fishery node with a random movement mode and the central fishing location, and sequentially connecting fishery nodes with a random movement mode in order of distance from small to large to form fishery fishing random links; connecting the fishery fishing random links behind the fishery fishing fixed links to form a fishing propagation time sequence link of each fishery cluster.

7. The fish capture smart identification method of claim 1, wherein, The similarity matching matrix between the weight propagation region link and the fishing propagation time sequence link of each fishery cluster is constructed, and based on the similarity matching matrix, fishery nodes are screened and fused from the weight propagation region link and the fishing propagation time sequence link respectively to form multi-source fishery nodes, and then a multi-source propagation time sequence link of each fishery cluster is determined, including: In each fishery cluster, the link position cost of two fishery nodes is determined by subtracting the position sequence number of any fishery node in the weight propagation region link from the position sequence number of any fishery node in the fishing propagation time sequence link, taking the absolute value, and dividing by the total number of fishery nodes; The fishery weight cost of two fishery nodes is determined by subtracting the fishery weight of any fishery node in the weight propagation region link from the fishery weight of any fishery node in the fishing propagation time sequence link, taking the absolute value, and dividing by the maximum value of the fishery weights of the two fishery nodes; The fishing distance cost of two fishery nodes is determined by calculating the Euclidean distance between the fishing locations of any fishery node in the weight propagation region link and any fishery node in the fishing propagation time sequence link; The similarity matching cost between any fishery node in the weight propagation region link and any fishery node in the fishing propagation time sequence link is obtained by weighting and summing the link position cost, the fishery weight cost and the fishing distance cost of the two fishery nodes. constructing a similarity matching matrix between the weight propagation regional link and the fishing propagation time sequence link of each fishery cluster based on the similarity matching cost of any one of the fishery nodes in the weight propagation regional link and any one of the fishery nodes in the fishing propagation time sequence link; solving the similarity matching matrix according to a preset link matching algorithm to construct an optimal fishery node matching matrix of each fishery cluster; based on the optimal fishery node matching matrix, screening fishery nodes from the weight propagation regional link and the fishing propagation time sequence link respectively, and performing weighted summation and averaging on the position serial numbers of the two screened fishery nodes to obtain a weighted average position of a multi-source fishery node; performing weighted summation and averaging on the fishery weight, fishing time and fishing location of the two screened fishery nodes respectively to form a weighted fishery weight, a weighted fishing time and a weighted fishing location of the multi-source fishery node respectively; sorting the multi-source fishery nodes in ascending order of the weighted average position, and sequentially connecting the multi-source fishery nodes to determine a multi-source propagation time sequence link of each fishery cluster.

8. The fish catch intelligent recognition method according to any one of claims 1 to 7, characterized in that, constructing a fishery multi-source data relationship graph of each fishery cluster based on the multi-source propagation time sequence link; performing graph feature extraction on the fishery multi-source data relationship graph to obtain a fishery intelligent recognition result, including: constructing a fishery multi-source data relationship graph of each fishery cluster based on the multi-source propagation time sequence link; dividing the fishery multi-source data relationship graph according to a preset label propagation algorithm to determine a fishing hotspot area of each fishery species; identifying abnormal multi-source fishery nodes in the fishery multi-source data relationship graph according to a preset isolation forest algorithm, and forming an abnormal fishery fishing operation event of each fishery species based on the weighted fishery weight, the weighted fishing time and the weighted fishing location of the abnormal multi-source fishery nodes; forming a fishery intelligent recognition result based on the fishing hotspot area and the abnormal fishery fishing operation event of each fishery species.

9. A fish catch intelligent recognition system, characterized by, including: a fishery species identification module, a fishery cluster construction module, a weight propagation regional link construction module, a fishing propagation time sequence link construction module, a multi-source propagation time sequence link construction module and a graph feature extraction module; wherein the fishery species identification module is configured to obtain a plurality of fishery images, and the fishing time, fishery weight and fishing location corresponding to each fishery image, and then construct a plurality of fishery species multi-source identification vectors; and perform feature recognition on the fishery species multi-source identification vectors based on a preset fishery species identification neural network model to obtain a fishery species corresponding to each fishery species multi-source identification vector; the fishery cluster construction module is configured to initialize a plurality of fishery clusters based on the fishery species and the fishery species multi-source identification vectors, and each fishery cluster contains a plurality of fishery nodes of the same fishery species; The weight propagation area link construction module is configured to construct a spatial weight matrix based on the fishing locations; perform autocorrelation analysis on the fish weight to determine fish weight gradient distribution; construct a weight propagation path based on the spatial weight matrix and the fish weight gradient distribution; and sequentially connect the fish nodes in the fish cluster to determine the weight propagation area link of each fish cluster. The fishing propagation time sequence link construction module is configured to identify the spatial movement mode of each fish node in each fish cluster based on the fishing time and the fishing location, and construct a fishing propagation time sequence link of each fish cluster based on the spatial movement mode. The multi-source propagation time sequence link construction module is configured to construct a similarity matching matrix between the weight propagation area link and the fishing propagation time sequence link of each fish cluster, and filter fish nodes from the weight propagation area link and the fishing propagation time sequence link based on the similarity matching matrix to form multi-source fish nodes, and further determine a multi-source propagation time sequence link of each fish cluster. The graph feature extraction module is configured to construct a fish multi-source data relationship graph of each fish cluster based on the multi-source propagation time sequence link, and perform graph feature extraction on the fish multi-source data relationship graph to obtain fish intelligent recognition results. The weight propagation area link construction module is configured to construct a spatial weight matrix based on the fishing locations; perform autocorrelation analysis on the fish weight to determine fish weight gradient distribution; construct a weight propagation path based on the spatial weight matrix and the fish weight gradient distribution; and sequentially connect the fish nodes in the fish cluster to determine the weight propagation area link of each fish cluster. In each fish cluster, the fish nodes are clustered based on the fishing locations of the fish nodes to obtain a plurality of fish clusters in each fish cluster. If two fish nodes belong to the same fish cluster, the Euclidean distance between the fishing locations of the two fish nodes is calculated as the spatial weight of the two fish nodes. If two fish nodes do not belong to the same fish cluster, the spatial weight of the two fish nodes is set to a preset negative value. Based on the spatial weight of each two fish nodes in each fish cluster, a spatial weight matrix of each fish cluster is formed. Based on the spatial weight matrix and the fish weight, autocorrelation analysis is performed on the fish weight of each fish cluster to determine the global correlation and the local correlation set of each fish cluster; and the fish weight gradient distribution of each fish cluster is determined based on the global correlation and the local correlation set. The weight propagation path of each fish cluster is constructed based on the spatial weight matrix and the fish weight gradient distribution; and the fish nodes in each fish cluster are sequentially connected based on the weight propagation path to determine the weight propagation area link of each fish cluster.

Images