Fishing boat behavior intelligent identification method and device, electronic equipment and storage medium

Abstract

The invention relates to a fishing boat behavior intelligent identification method and device, electronic equipment and a storage medium, and belongs to the technical field of fishery, and the method comprises the steps: carrying out the feature extraction of the fishing boat trajectory data of a target fishing boat, and obtaining the trajectory features; wherein the trajectory features comprise global statistical features of the trajectory characterization quantity, local differential features corresponding to each grid region, global semantic features, local semantic features, statistical features of the local semantic features and interaction features of the local semantic features and the global semantic features; and inputting the trajectory features into a trained fishing boat behavior identification model to obtain behavior information of the target fishing boat. According to the method, the multi-dimensional trajectory characteristics of the fishing boat are extracted, and the micro maneuvering behavior and the macro operation mode of the fishing boat and the semantic association between the micro maneuvering behavior and the macro operation mode are comprehensively captured, so that the behavior information of the fishing boat can be predicted more accurately through prediction based on the trajectory characteristics by the fishing boat behavior identification model.

Description

Technical Field

[0001] This invention relates to the field of fisheries technology, and in particular to a method, device, electronic device, and storage medium for intelligent identification of fishing vessel behavior. Background Technology

[0002] The Automatic Identification System (AIS) plays an indispensable role as a crucial tool for global maritime surveillance. It broadcasts real-time dynamic and static data such as ship speed, course, and position via VHF radio frequencies. This information not only provides navigational safety assurance for shipping companies but also offers vital support to maritime regulatory authorities and emergency response agencies. However, despite its excellent performance in data collection and transmission, the AIS system still has some fundamental limitations.

[0003] The most significant problem is that the AIS system cannot directly perceive the operational status of a vessel. For example, while AIS can provide basic navigation information about a vessel, it cannot determine whether the vessel is engaged in specific operational activities such as fishing, transportation, or rest. This creates a technological bottleneck for fisheries management and supervision, particularly in identifying the specific operational behaviors of fishing vessels (such as trawls, purse seines, and gillnets), where intelligent analysis relying on AIS data streams becomes especially important.

[0004] In this field, although existing research has attempted to improve the accuracy of job behavior classification through methods such as density clustering and deep learning, it still faces multiple challenges. For example, traditional feature engineering struggles to effectively represent the semantic relationship between local trajectory points and macroscopic behavioral patterns when extracting data features, resulting in insufficient model recognition capabilities for different job behaviors. Summary of the Invention

[0005] In view of this, it is necessary to provide a method, device, electronic device and storage medium for intelligent identification of fishing vessel behavior, so as to solve the problem that the existing technology is insufficient in identifying the behavior of fishing vessels.

[0006] To address the aforementioned problems, in a first aspect, the present invention provides a method for intelligent identification of fishing vessel behavior, comprising: Feature extraction is performed on the trajectory data of the target fishing vessel to obtain trajectory features; wherein, the trajectory features include global statistical features of trajectory representation, local differential features corresponding to each grid region, global semantic features, local semantic features, statistical features of the local semantic features, and interaction features between the local semantic features and the global semantic features; The trajectory features are input into the trained fishing vessel behavior recognition model to obtain the behavior information of the target fishing vessel.

[0007] In one possible implementation, the local differential feature is obtained by the following formula:

[0008] In the formula, For the set of continuous trajectory segments corresponding to the grid region, For the local differential features corresponding to the fragment set, and For the first fragment in the collection k The and the first k+ 1 trajectory point, This is the time decay weighting coefficient. , and The first k The and the first k+1 The time for each trajectory point and The first k The and the first k +1 longitude of the trajectory point, and The first k The and the first k +1 latitude of trajectory points, ⊕ is the feature splicing operation.

[0009] In one possible implementation, the local semantic features are a coupled embedding representation of the velocity and position of the trajectory points corresponding to the sliding window. The feature extraction of the target fishing vessel's trajectory data to obtain trajectory features includes: By using a word embedding model, a sliding window is used to process the trajectory point sequence, resulting in a coupled embedding representation of the velocity and position of the trajectory points corresponding to each sliding window.

[0010] In one possible implementation, the fishing vessel behavior recognition model is trained in the following manner: Label smoothing is performed on the fishing vessel behavior category labels in the training samples to obtain soft labels for fishing vessel behavior categories; The fishing vessel behavior identification model makes predictions based on the sample trajectory features in the training samples, and trains the model according to the predicted category label and the soft label of the fishing vessel behavior category until the model converges.

[0011] In one possible implementation, the step of smoothing the labels of fishing vessel behavior categories in the training samples to obtain soft labels for fishing vessel behavior categories includes: Label smoothing is performed using the following formula:

[0012] In the formula, Indicates the firstk Soft tags for fishing vessel behavior categories Indicates the first k Individual fishing vessel behavior category tags Represents the smoothing coefficient. K This indicates the number of categories of fishing vessel behavior.

[0013] In one possible implementation, the fishing vessel behavior identification model is a gradient boosting decision tree model based on the LightGBM framework. The step of training the fishing vessel behavior identification model based on the predicted category label and the soft label of the fishing vessel behavior category until the model converges includes: The symmetric cross-entropy loss is calculated based on the predicted category label and the soft label of the fishing vessel behavior category, and the parameters of the gradient boosting decision tree model are updated based on the symmetric cross-entropy loss until the model converges.

[0014] In one possible implementation, the method further includes: The SHAP values ​​of the trajectory features are generated using the kernel SHAP algorithm and the tree SHAP algorithm. The SHAP value is then visualized.

[0015] Secondly, the present invention also provides a fishing vessel behavior intelligent identification device, comprising: The feature extraction module is used to extract features from the trajectory data of the target fishing vessel to obtain trajectory features; wherein, the trajectory features include global statistical features of trajectory representation, local differential features corresponding to each grid region, global semantic features, local semantic features, statistical features of the local semantic features, and interaction features between the local semantic features and the global semantic features. The fishing vessel behavior recognition module is used to input the trajectory features into the trained fishing vessel behavior recognition model to obtain the behavior information of the target fishing vessel.

[0016] Thirdly, the present invention also provides an electronic device, including a memory and a processor, wherein the memory is used to store a program; the processor is coupled to the memory and is used to execute the program stored in the memory to implement the steps in the fishing vessel behavior intelligent identification method described in any of the above claims.

[0017] Fourthly, the present invention also provides a computer-readable storage medium for storing a computer-readable program, wherein the program or instructions, when executed by a processor, are capable of implementing the steps in any of the above-described intelligent identification method for fishing vessel behavior.

[0018] The beneficial effects of this invention are: This invention constructs multi-dimensional trajectory features by building global statistical features of fishing vessel trajectories, local differential features corresponding to grid regions, multi-level semantic features, and interactive features of multi-level semantic features. This allows for a more comprehensive capture of the micro-maneuvering behavior and macro-operational patterns of fishing vessels, and effectively achieves semantic correlation between micro-maneuvering behavior and macro-operational patterns. Inputting these trajectory features into a pre-trained fishing vessel behavior recognition model yields more accurate behavioral information about the fishing vessel. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 A flowchart illustrating an embodiment of the intelligent identification method for fishing vessel behavior provided by the present invention; Figure 2 A framework diagram of an intelligent identification method for fishing vessel behavior provided by the present invention; Figure 3 A schematic diagram of the structure of an embodiment of the intelligent identification device for fishing vessel behavior provided by the present invention; Figure 4 A schematic diagram of an embodiment of the electronic device provided by the present invention. Detailed Implementation

[0021] 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 a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0022] In the description of the embodiments of this invention, unless otherwise stated, "a plurality of" means two or more. The terms "first," "second," etc., used in the embodiments of this invention are used to distinguish similar objects, and are not used to describe a specific order or sequence, nor to indicate or imply their relative importance or implicitly specify the number of indicated technical features. It should be understood that such data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class, and the number of objects is not limited; for example, a first object can be one or more.

[0023] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0024] Reference Figure 1 The diagram illustrates a flowchart of an embodiment of the intelligent identification method for fishing vessel behavior provided by the present invention. The method includes: S101, extract features from the trajectory data of the target fishing vessel to obtain trajectory features; wherein, the trajectory features include global statistical features of trajectory representation, local differential features corresponding to each grid region, global semantic features, local semantic features, statistical features of local semantic features, and interaction features between local semantic features and global semantic features.

[0025] Trajectory representations can include velocity, heading, longitude, latitude, acceleration, etc., of trajectory points. Global statistical characteristics refer to the statistical calculation of the overall distribution characteristics of trajectory representations, which can be achieved using statistical measures such as mean, variance, and extreme values, and are used to characterize the behavior patterns of fishing vessels over a global time span.

[0026] The grid regions are pre-divided according to geographic location. Local differential features refer to the features obtained by quantifying the instantaneous change rate of the trajectory representation quantity of each grid region, which are used to capture the dynamic change patterns between trajectory points.

[0027] Global semantic features refer to the vector representation obtained by encoding the complete trajectory sequence through a semantic model. Specifically, they can be implemented using recurrent neural networks or attention mechanisms, and are used to extract long-term behavioral intentions.

[0028] Local semantic features refer to the semantic embedding of each trajectory segment divided from the complete trajectory sequence. Specifically, word embedding models can be used to process the coupling information of speed and position, which is used to reflect short-term behavioral patterns.

[0029] Statistical features of local semantic features refer to the aggregation analysis of local semantic features, which can be achieved by mean pooling or max pooling operations to enhance the stability of local patterns.

[0030] Interaction features between local and global semantic features refer to establishing a connection between the two through feature concatenation or cross-attention mechanisms. Specifically, matrix multiplication or cascading operations can be used to fuse semantic information at different time scales.

[0031] S102, input the trajectory features into the trained fishing vessel behavior recognition model to obtain the behavior information of the target fishing vessel.

[0032] Fishing vessel behavior identification models can be decision tree models, gradient boosting models, neural network models, etc., and can be trained using labeled historical fishing vessel trajectory data.

[0033] By inputting the trajectory features of the target fishing vessel into the trained fishing vessel behavior recognition model, the behavior information of the target fishing vessel can be obtained. The behavior information can include behavior category, behavior time, etc. Among them, behavior category includes trawl, purse seine, gillnet, etc.

[0034] The intelligent identification method for fishing vessel behavior provided in this embodiment can be applied to an intelligent identification system for fishing vessel behavior, which can be a software system running on a terminal device. The terminal device can be a tablet computer, shipborne equipment, augmented reality (AR) / virtual reality (VR) device, laptop computer, ultra-mobile personal computer (UMPC), netbook, personal digital assistant (PDA), mobile phone, etc. This embodiment does not impose any restrictions on the specific type of terminal device.

[0035] Traditional methods rely solely on single-level statistical features or shallow semantic analysis, making it difficult to capture the correlation between local dynamics and global intent in fishing vessel trajectory data. This embodiment constructs global statistical features, local differential features, multi-level semantic features, and interactive features of multi-level semantic features to form multi-dimensional trajectory features. This allows for a more comprehensive capture of the micro-maneuvering behavior and macro-operational patterns of fishing vessels, effectively achieving semantic correlation between micro-maneuvering behavior and macro-operational patterns. Inputting these trajectory features into a pre-trained fishing vessel behavior recognition model yields more accurate behavioral information about the fishing vessel.

[0036] In some embodiments of the present invention, when constructing trajectory features, a set of fishing boat trajectory representation quantities can be used. ,in, Represents the velocity of the trajectory point. Indicates the heading of the trajectory point. Represents the longitude of the trajectory point. Representing the latitude of the trajectory points, global-grid dual-scale statistical modeling is performed to obtain global statistical features and local differential features. The global statistical features are represented as follows:

[0037] In the formula, Represents a set X The Middle i The trajectory representation of a trajectory point is specifically: , , and , Represents a set X middle maximum value Represents a set X middle minimum value Represents a set X In mean Represents a set X middle The quantity.

[0038] Local differential features are constructed by geospatial grids The trajectory fragment set is divided according to the grid region, and then the differential features of the trajectory fragment set are calculated. The calculation formula is as follows:

[0039] In the formula, For the set of trajectory segments corresponding to the grid region, For the local differential features corresponding to the fragment set, and For the first fragment in the collection k The and the first k+ 1 trajectory point, This is the time decay weighting coefficient. exp represents the exponential function. and The first k The and the first k+1 The time for each trajectory point and The first k The and the first k +1 longitude of the trajectory point, and The first k The and the first k +1 latitude of trajectory points, ⊕ is the feature splicing operation.

[0040] The time decay weighting coefficient is used to reduce the impact of adjacent trajectory points with excessively large time intervals on local differential features, and to suppress the interference of abnormal trajectory points with long time intervals on local features. By combining the dynamic coupling relationship between latitude and longitude changes and time decay, local differential features can more accurately represent the instantaneous behavioral changes of fishing vessels under complex sea conditions. Global semantic features enable the ALBERT model to process complete trajectory point sequences. The semantic representation of the dynamic behavior output afterwards:

[0041] Local semantic features can be a coupled embedding representation of the velocity and position of the trajectory points corresponding to the sliding window, which can be obtained by processing the trajectory point sequence with a sliding window using a word embedding model. Specifically, the word embedding model can be a Word2Vec model, using the Skip-gram architecture of Word2Vec to process the trajectory point sequence with a sliding window, learning a distributed representation of the joint velocity-position distribution of the trajectory points in the sliding window:

[0042] In the formula, For velocity semantic vectors, For geocoding vectors, Parameters representing the Skip-gram architecture, c For a single context trajectory point, p Represents probability. C t Center trajectory point t The set of context trajectory points, W is the trajectory point centered at. t Based on this, a sequence of local trajectory points is constructed using a sliding window mechanism. This formula means that the spliced ​​features (the spliced ​​vector of velocity semantic vector and geocoding vector) of each trajectory point in the local trajectory point sequence constructed through the sliding window mechanism are input into the Skip-gram model, and then the semantic features corresponding to the local trajectory point sequence are obtained, that is, the local semantic features.

[0043] By combining a sliding window mechanism with a word embedding model, the implicit relationship between velocity and position in local trajectory segments can be automatically learned, generating coupled embedding features with semantic expressive power. Through the embedded representation of coupled velocity and position information, the model can distinguish the spatiotemporal distribution differences of trajectory points under different operation modes, such as the continuous low-speed movement feature of trawl operations and the intermittent mooring feature of gillnet operations, thus providing more discriminative feature inputs for behavior classification models.

[0044] The statistical features of local semantic features can be represented as follows:

[0045] In the formula, , , , These represent the maximum, minimum, mean, and standard deviation of the local semantic features in each dimension, respectively.

[0046] The interaction features between local semantic features and global semantic features can be represented as follows:

[0047] In some embodiments of the present invention, the extracted trajectory features can be laterally fused to form a feature library. Specifically, the trajectory features are merged with the original trajectory data to form the final feature library. To improve storage efficiency, a sparse matrix storage structure can be used to represent non-zero elements in the high-dimensional feature space, as follows:

[0048] This embodiment can effectively store the spatiotemporal behavioral characteristics of fishing vessels, providing an accurate data foundation and support for subsequent fishing vessel behavior identification.

[0049] In some embodiments of the present invention, the fishing vessel behavior recognition model can be trained in the following manner: Label smoothing is performed on the fishing vessel behavior category labels in the training samples to obtain soft labels for fishing vessel behavior categories; The fishing vessel behavior identification model makes predictions based on the trajectory features of the training samples, and trains the model according to the predicted category label and the soft label of the fishing vessel behavior category until the model converges.

[0050] Specifically, in the model initialization phase, the predicted values ​​are first set to the mean of the target values, where the target values ​​are soft labels after label smoothing. The label smoothing formula is:

[0051] In the formula, Indicates the first k Soft tags for fishing vessel behavior categories Indicates the first k Individual fishing vessel behavior category tags This represents the smoothing coefficient (typically 0.1). K This indicates the number of categories of fishing vessel behavior.

[0052] In each iteration, the model calculates the residual between the current predicted value and the smoothed target value. The formula for calculating the residual is:

[0053] In the formula, It is the target value after smoothing.

[0054] Then, the model is trained with the residuals as the optimization objective until the model converges.

[0055] This embodiment can effectively alleviate the overfitting problem caused by annotation noise or category ambiguity during the training process of fishing vessel behavior recognition model, and improve the model's generalization ability to complex trajectory features in real-world scenarios.

[0056] In some embodiments of the present invention, the fishing vessel behavior identification model is a gradient boosting decision tree model based on the LightGBM framework. The step of training the fishing vessel behavior identification model according to the predicted category label and the soft label of the fishing vessel behavior category until the model converges includes: The symmetric cross-entropy loss is calculated based on the predicted category label and the soft label of the fishing vessel behavior category, and the parameters of the gradient boosting decision tree model are updated based on the symmetric cross-entropy loss until the model converges.

[0057] Symmetric cross-entropy loss is a loss function formed by linearly combining standard cross-entropy loss and inverse cross-entropy loss. This loss function can alleviate the class imbalance problem in training data and is robust to label noise. The LightGBM framework is an efficient machine learning framework based on gradient boosting decision trees. Specifically, it can be implemented using histogram algorithms and gradient-based one-sided sampling techniques. It can quickly process high-dimensional structured data and reduce memory consumption, making it suitable for classification tasks based on fishing boat trajectory features.

[0058] In this embodiment, the gradient can be calculated using the symmetric cross-entropy loss function:

[0059] In the formula, Represents cross-entropy. and These are hyperparameters that adjust the weights. Based on these gradients, the model calculates the optimal split point for the features using a histogram algorithm. Simultaneously, a leaf node-based growth strategy is employed to select the optimal split node. The formula for calculating the split gain is:

[0060] In the formula, It is a first-order gradient. It is the second-order gradient, and λ is the regularization system. γ For the minimum gain required for splitting, L and R These represent the left child node and the right child node, respectively.

[0061] After the tree is constructed, the model needs to update the predicted values. The update formula is:

[0062] In the formula, It's the learning rate. These are newly added trees.

[0063] The residual calculation and updated predictions will be repeated until a predetermined number of trees are reached or other stopping conditions are met.

[0064] The gradient boosting decision tree model in this embodiment, also known as the LabelSmoothing-Noise-Robust LightGBM model, retains the efficient characteristics of the original LightGBM through its leaf growth strategy. At the same time, it significantly enhances the model's adaptability to label noise and class imbalance in fishing vessel behavior recognition tasks through label smoothing and symmetric cross-entropy loss.

[0065] In some embodiments of the present invention, the intelligent identification method for fishing vessel behavior further includes: SHAP values ​​for trajectory features are generated using the kernel SHAP algorithm and the tree SHAP algorithm. Visualize the SHAP value.

[0066] Specifically, this step aims to utilize the SHAP (SHapley Additive exPlanations) algorithm for in-depth interpretation and analysis of the model, particularly in gradient boosting decision tree models based on LightGBM. The implementation process of this step includes the following key aspects: The SHAP algorithm is based on the Shapley value concept in cooperative game theory, aiming to fairly distribute the contribution of each feature in the model output. The Shapley value is defined as the average marginal contribution of each participant (feature) to the cooperative payoff. Its core objective is to quantify the specific impact of different feature combinations on the classification results during model interpretation.

[0067] Traditional methods for calculating SHAP values ​​require traversing all possible feature combinations; however, this becomes infeasible with high-dimensional data due to the combinatorial explosion problem. Therefore, this implementation uses two main algorithms to efficiently calculate SHAP values: kernel SHAP and tree SHAP.

[0068] The kernel SHAP algorithm constructs a simplified linear model as a proxy, using high-dimensional kernel techniques to map the feature space. The core expression of this linear model is:

[0069] Kernel SHAP estimates the SHAP value by minimizing the following loss function:

[0070] By optimizing this loss function, kernel SHAP can estimate the SHAP value of each feature, revealing the average contribution of the feature to the model output.

[0071] The Tree SHAP algorithm is specifically designed for tree-based models (such as LightGBM). It directly analyzes the influence path of each feature within the tree structure, eliminating the need to build a surrogate model. Tree SHAP calculates the marginal contribution of a feature on different paths by parsing the classification paths in the tree. The formula for calculating the SHAP value for a given classification path is:

[0072] This formula shows that the feature The SHAP value is evaluated by analyzing the role of the weights and features of each node on the path in the split, which significantly improves computational efficiency.

[0073] By calculating SHAP values, we can generate an analysis of the contribution of each feature to the final decision, forming a comprehensive interpretation of the model. This process can display the importance of features through visualization tools (such as SHAP value plots) and understand the impact of each feature on a specific prediction through local interpretations (such as the SHAP value of a single prediction case).

[0074] Reference Figure 2 This paper illustrates a framework diagram of an intelligent identification method for fishing vessel behavior provided by the present invention. The method includes establishing fishing vessel behavior features, behavior identification based on spatiotemporal semantic features, and analysis of the decision-making process of interpretable AI. Establishing fishing vessel behavior features involves performing global and local statistics on the trajectory representation of the fishing vessel to obtain global statistical features and local differential features. The trajectory point sequence is then input into a semantic model to obtain global semantic features, local semantic features, and their interaction features. Behavior identification based on spatiotemporal semantic features involves inputting the trajectory features of the fishing vessel into a label-smoothed noise robust LightGBM model to obtain the behavior identification result. Analysis of the decision-making process of interpretable AI involves analyzing the input features of the label-smoothed noise robust LightGBM model using the SHAP algorithm.

[0075] Reference Figure 3 The diagram shows a structural schematic of an embodiment of the intelligent identification device for fishing vessel behavior provided by the present invention. The device 300 includes: The feature extraction module 301 is used to extract features from the fishing vessel trajectory data of the target fishing vessel to obtain trajectory features. The trajectory features include global statistical features of trajectory representation, local differential features corresponding to each grid region, global semantic features, local semantic features, statistical features of local semantic features, and interaction features between local semantic features and global semantic features. The fishing vessel behavior recognition module 302 is used to input trajectory features into the trained fishing vessel behavior recognition model to obtain the behavior information of the target fishing vessel.

[0076] It should be noted that the implementation principles or processes of the above modules can be referred to the aforementioned implementation examples of the intelligent identification method for fishing vessel behavior, and will not be elaborated here.

[0077] Reference Figure 4 The present invention illustrates an electronic device 400. The electronic device 400 includes a processor 401, a memory 402, and a display 403. Figure 4 Only some components of the electronic device 400 are shown, but it should be understood that it is not required to implement all the components shown, and more or fewer components may be implemented instead.

[0078] In some embodiments, processor 401 may be a central processing unit (CPU), microprocessor or other data processing chip, used to run program code stored in memory 402 or process data, such as the intelligent identification method for fishing vessel behavior in this invention.

[0079] In some embodiments, processor 401 may be a single server or a group of servers. The server group may be centralized or distributed. In some embodiments, processor 401 may be local or remote. In some embodiments, processor 401 may be implemented on a cloud platform. In one embodiment, the cloud platform may include a private cloud, public cloud, hybrid cloud, community cloud, distributed cloud, internal cloud, multi-cloud, etc., or any combination thereof.

[0080] In some embodiments, memory 402 may be an internal storage unit of electronic device 400, such as a hard disk or memory of electronic device 400. In other embodiments, memory 402 may also be an external storage device of electronic device 400, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc. equipped on electronic device 400.

[0081] Furthermore, the memory 402 may include both internal storage units of the electronic device 400 and external storage devices. The memory 402 is used to store application software and various types of data installed on the electronic device 400.

[0082] In some embodiments, display 403 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. Display 403 is used to display information from electronic device 400 and to display a visual user interface. Components 401-403 of electronic device 400 communicate with each other via a system bus.

[0083] In one embodiment, when the processor 401 executes the fishing vessel behavior intelligent recognition program in the memory 402, the following steps can be implemented: Feature extraction is performed on the trajectory data of the target fishing vessel to obtain trajectory features. The trajectory features include global statistical features of trajectory representation, local differential features corresponding to each grid region, global semantic features, local semantic features, statistical features of local semantic features, and interaction features between local semantic features and global semantic features. The trajectory features are input into the trained fishing vessel behavior recognition model to obtain the behavior information of the target fishing vessel.

[0084] It should be understood that when the processor 401 executes the fishing vessel behavior intelligent identification program in the memory 402, in addition to the functions mentioned above, it can also perform other functions, as can be found in the description of the corresponding method embodiments above.

[0085] Furthermore, this embodiment of the invention does not specifically limit the type of electronic device 400 mentioned. Electronic device 400 can be a mobile phone, tablet computer, personal digital assistant (PDA), wearable device, laptop computer, or other portable electronic device. Exemplary embodiments of portable electronic devices include, but are not limited to, portable electronic devices running iOS, Android, Microsoft, or other operating systems. The aforementioned portable electronic device can also be other portable electronic devices, such as a laptop computer with a touch-sensitive surface (e.g., a touch panel). It should also be understood that in some other embodiments of the invention, electronic device 400 may not be a portable electronic device, but rather a desktop computer with a touch-sensitive surface (e.g., a touch panel).

[0086] In one embodiment, the present invention also provides a computer-readable storage medium storing a computer program that, when executed by the processor, implements the steps of any of the above-described intelligent identification methods for fishing vessel behavior.

[0087] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware, and the program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.

[0088] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for intelligent identification of fishing vessel behavior, characterized in that, include: Feature extraction is performed on the trajectory data of the target fishing vessel to obtain trajectory features; wherein, the trajectory features include global statistical features of trajectory representation, local differential features corresponding to each grid region, global semantic features, local semantic features, statistical features of the local semantic features, and interaction features between the local semantic features and the global semantic features; The trajectory features are input into the trained fishing vessel behavior recognition model to obtain the behavior information of the target fishing vessel.

2. The intelligent identification method for fishing vessel behavior according to claim 1, characterized in that, The local differential feature is obtained by the following formula: In the formula, For the set of trajectory segments corresponding to the grid region, For the local differential features corresponding to the fragment set, and For the first fragment in the collection k The and the first k+ One trajectory point, This is the time decay weighting coefficient. exp represents the exponential function. and The first k The and the first k+1 The time for each trajectory point and The first k The and the first k +1 longitude of the trajectory point, and The first k The and the first k +1 latitude of trajectory points, ⊕ is the feature splicing operation.

3. The intelligent identification method for fishing vessel behavior according to claim 1, characterized in that, The local semantic features are coupled embedding representations of the velocity and position of the trajectory points corresponding to the sliding window. The feature extraction of the target fishing vessel's trajectory data to obtain trajectory features includes: By using a word embedding model, a sliding window is used to process the trajectory point sequence, resulting in a coupled embedding representation of the velocity and position of the trajectory points corresponding to each sliding window.

4. The intelligent identification method for fishing vessel behavior according to claim 1, characterized in that, The fishing vessel behavior recognition model was trained in the following manner: Label smoothing is performed on the fishing vessel behavior category labels in the training samples to obtain soft labels for fishing vessel behavior categories; The fishing vessel behavior identification model makes predictions based on the sample trajectory features in the training samples, and trains the model according to the predicted category label and the soft label of the fishing vessel behavior category until the model converges.

5. The intelligent identification method for fishing vessel behavior according to claim 4, characterized in that, The step of smoothing the labels of fishing vessel behavior categories in the training samples to obtain soft labels for fishing vessel behavior categories includes: Label smoothing is performed using the following formula: In the formula, Indicates the first k Soft tags for fishing vessel behavior categories Indicates the first k Individual fishing vessel behavior category tags Represents the smoothing coefficient. K This indicates the number of categories of fishing vessel behavior.

6. The intelligent identification method for fishing vessel behavior according to claim 4, characterized in that, The fishing vessel behavior identification model is a gradient boosting decision tree model based on the LightGBM framework. The step of training the fishing vessel behavior identification model based on the predicted category label and the soft label of the fishing vessel behavior category until the model converges includes: The symmetric cross-entropy loss is calculated based on the predicted category label and the soft label of the fishing vessel behavior category, and the parameters of the gradient boosting decision tree model are updated based on the symmetric cross-entropy loss until the model converges.

7. The intelligent identification method for fishing vessel behavior according to claim 1, characterized in that, The method further includes: The SHAP values ​​of the trajectory features are generated using the kernel SHAP algorithm and the tree SHAP algorithm. The SHAP value is then visualized.

8. A smart identification device for fishing vessel behavior, characterized in that, include: The feature extraction module is used to extract features from the trajectory data of the target fishing vessel to obtain trajectory features; wherein, the trajectory features include global statistical features of trajectory representation, local differential features corresponding to each grid region, global semantic features, local semantic features, statistical features of the local semantic features, and interaction features between the local semantic features and the global semantic features. The fishing vessel behavior recognition module is used to input the trajectory features into the trained fishing vessel behavior recognition model to obtain the behavior information of the target fishing vessel.

9. An electronic device, characterized in that, Including memory and processor, among which, The memory is used to store programs; The processor, coupled to the memory, is used to execute the program stored in the memory to implement the steps in the intelligent identification method for fishing vessel behavior as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, Used to store computer-readable programs or instructions, which, when executed by a processor, can implement the steps in the intelligent identification method for fishing vessel behavior according to any one of claims 1 to 7.

Images