Fishing boat operator safety monitoring method and system based on artificial intelligence

By integrating weighted training of subnets with safety area, environment and behavior monitoring submodels, visual and neural network technology is improved, and the problems of incomplete analysis and low computing efficiency in safety monitoring of fishing boat operators are solved, and the safety of fishing boat operators is improved Comprehensive and intelligent monitoring.

CN118942043AActive Publication Date: 2024-11-12DITAI (ZHEJIANG) COMM TECH CO LTD
View PDF 12 Cites 0 Cited by

Patent Information

Application Number
CN202411419078.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-10-12
Publication Date
2024-11-12
Estimated Expiration
2044-10-12

AI Technical Summary

Technical Problem

The existing safety monitoring technology for fishing boat operators has problems such as insufficient analysis, low computing efficiency, poor identification of small targets, and ignoring the safety of fishing boat navigation process.

Method used

The subnet integrated weighted training scheme combining the safe area monitoring submodel, the fishing boat environmental monitoring submodel and the personnel behavior monitoring submodel are adopted to improve the visual detection model and the convolutional long short-term memory neural network, and combine fuzzy reasoning and deep attention network for multi-level Feature processing and trajectory feature description improve the comprehensiveness and intelligence of safety monitoring of fishing boat operators.

Benefits of technology

It improves the comprehensiveness, intelligence and reliability of safety monitoring of fishing boat operators, enhances the small target recognition ability and computing efficiency, captures potential behavioral logic, and provides support for complex environmental monitoring.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118942043A_ABST
    Figure CN118942043A_ABST
Patent Text Reader

Abstract

The invention discloses a fishing boat operating personnel safety monitoring method and system based on artificial intelligence. The method comprises the steps of basic data acquisition, safety area monitoring, fishing boat environment monitoring, personnel behavior monitoring and comprehensive safety monitoring. The invention relates to the technical field of fishing boat safety monitoring, in particular to a fishing boat operator safety monitoring method and system based on artificial intelligence, and adopts an integrated safety analysis scheme that a safety area monitoring sub-model, a fishing boat environment monitoring sub-model and a personnel behavior monitoring sub-model are combined to carry out sub-net integrated weighted training; a method of combining an improved visual detection model with multi-level feature processing is adopted to perform safety area monitoring, and the small target recognition capability and the overall calculation efficiency are enhanced by improving the model structure; performing fishing boat environment monitoring by adopting a convolutional long-short-term memory neural network optimized by combining fishing boat track feature description; and performing personnel behavior monitoring by adopting a deep attention network combining fuzzy reasoning and pre-training visual extraction.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of fishing vessel safety monitoring, and in particular to an artificial intelligence-based fishing vessel operator safety monitoring method and system. Background Art

[0002] The artificial intelligence-based safety monitoring method for fishing vessel operators refers to the use of AI technology to monitor and analyze the activities of fishing vessel operators in real time to identify and prevent potential safety risks, thereby ensuring the life safety of fishing vessel operators and improving the overall safety of fishery operations.

[0003] However, in the existing safety monitoring of fishing vessel operators, there is a technical problem that traditional intelligent technical solutions often only perform a single image analysis or a single numerical analysis, which leads to the technical problem that the intelligent data support for the safety monitoring of fishing vessel operators is not comprehensive enough and has low reference value; in the existing safety area monitoring methods, there is a technical problem that the existing standard image detection model has low computational efficiency and poor recognition of smaller targets in dealing with complex fishing vessel safety area monitoring tasks, which is likely to lead to poor safety area monitoring results and thus affect the safety risks of fishing vessel operators; in the existing fishing vessel environment monitoring methods, there is a technical problem that traditional environmental detection only performs a general intelligent analysis of the external environment, while ignoring the possible safety problems in the fishing vessel navigation process itself; in the existing personnel behavior monitoring methods, there is a technical problem that personnel behavior monitoring involves intuitive visual features, and may also involve other potential features that cannot be directly observed through camera data. Summary of the invention

[0004] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides a drug-target interaction prediction method and system based on artificial intelligence. In view of the fact that in the existing safety monitoring of fishing vessel operators, traditional intelligent technical solutions often only perform a single image analysis or a single numerical analysis, which leads to the technical problems that the intelligent data support for the safety monitoring of fishing vessel operators is not comprehensive enough and has low reference value, this solution creatively adopts an integrated safety analysis solution that combines the safety area monitoring sub-model, the fishing vessel environment monitoring sub-model and the personnel behavior monitoring sub-model to perform sub-network integrated weighted training. The combination of trajectory environment and personnel behavior analysis improves the comprehensiveness of safety monitoring of fishing vessel operators. It also improves the intelligence and reliability of safety monitoring of fishing vessel operators through weighted processing. At the same time, the design of sub-item sub-models is also conducive to the separate operation and maintenance of sub-tasks, which improves the overall stability and scalability of the system. In the existing safety area monitoring methods, there is a problem that the existing standard image detection model has low computational efficiency and poor recognition of smaller targets in dealing with complex fishing vessel safety area monitoring tasks, which is likely to lead to poor results in safety area monitoring and thus affect the safety risks of fishing vessel operators. The scheme creatively adopts the method of improving visual detection model combined with multi-level feature processing to monitor the safe area. By improving the model structure, the small target recognition ability and overall computing efficiency are enhanced, and the overall usability of safe area monitoring is improved through multi-level feature fusion. In view of the technical problem that in the existing fishing vessel environmental monitoring methods, traditional environmental detection only conducts a general intelligent analysis of the external environment, while ignoring the possible safety problems in the fishing vessel navigation process itself, this scheme creatively adopts the convolutional long short-term memory neural network optimized by combining the fishing vessel trajectory feature description to monitor the fishing vessel environment. By designing four fishing vessel navigation trajectory environmental characteristics and combining the convolutional long short-term memory neural network, the fishing vessel environment is monitored. The deep attention network combined with fuzzy reasoning and pre-trained visual extraction is creatively used for personnel behavior monitoring. It combines the results of fishing vessel environmental monitoring and the improvement of the model of safe area monitoring, and improves the potential logical capture ability of personnel behavior analysis, thereby improving the overall practicality of safety monitoring of fishing vessel operators.

[0005] The technical solution adopted by the present invention is as follows: The present invention provides a method for safety monitoring of fishing vessel operators based on artificial intelligence, the method comprising the following steps:

[0006] Step S1: basic data collection;

[0007] Step S2: safe area monitoring;

[0008] Step S3: Fishing vessel environment monitoring;

[0009] Step S4: personnel behavior monitoring;

[0010] Step S5: Comprehensive safety monitoring.

[0011] Further, in step S1, the basic data collection is used to collect the original data set required for the safety monitoring of fishing vessel operators, specifically, the fishing vessel safety monitoring data is collected through the camera device and the sensor to obtain the original data of the safety monitoring of fishing vessel operators;

[0012] The camera device is specifically placed in the cockpit, deck and stern of the fishing boat for monitoring safety areas and personnel behavior;

[0013] The sensors specifically include temperature sensors, humidity sensors, gas sensors, acceleration sensors and distance recognition communication sensors, which are used to monitor environmental conditions and hull status to assist in environmental monitoring of fishing vessels;

[0014] The raw data of safety monitoring of fishing vessel operators specifically includes raw video data and raw sensor data;

[0015] The original camera data specifically includes area division data and personnel behavior data;

[0016] The sensor raw data specifically includes temperature data, humidity data, gas data, fishing boat motion state data and distance recognition data.

[0017] Further, in step S2, the safety area monitoring is used to screen and pre-classify the safety area for fishing vessel operations, specifically, based on the original camera data in the original data of safety monitoring of fishing vessel operators, a multi-level feature fusion improved visual monitoring model combined with small target optimization is used to perform safety area monitoring, and obtain safety area monitoring reference data and safety area monitoring sub-model;

[0018] The multi-level feature fusion improved visual monitoring model combined with small target optimization specifically includes an improved visual monitoring model, a spatial depth convolution conversion sub-block, and a multi-level feature fusion sub-block;

[0019] The step of using a multi-level feature fusion improved visual monitoring model combined with small target optimization to monitor the safe area and obtain safe area monitoring reference data and a safe area monitoring sub-model includes:

[0020] Step S21: basic data optimization, specifically, performing data enhancement on the original camera data through basic image enhancement and filtering denoising to obtain optimized safe area monitoring data;

[0021] The basic image enhancement specifically refers to random rotation, scaling and flipping data enhancement operations;

[0022] The filtering and denoising specifically refers to a Gaussian filtering operation;

[0023] Step S22: improving the visual monitoring model, specifically selecting a pre-trained standard YOLOv8 model as a basic visual monitoring model, and improving the hierarchical structure of the basic visual monitoring model to obtain an improved visual monitoring model;

[0024] The hierarchical structure improvement specifically refers to deleting the seventh and eighth downsampling layers of the standard YOLOv8 model backbone network to optimize computational efficiency, and adding a regional monitoring optimization feature processing sub-block to the first convolutional layer of the standard YOLOv8 model backbone network to enhance the improved visual monitoring model's ability to recognize smaller targets in safe area monitoring;

[0025] The regional monitoring optimization feature processing sub-block specifically includes a first two-dimensional convolution layer, a second two-dimensional convolution layer, a third two-dimensional convolution layer and a branch processing layer, the first two-dimensional convolution layer and the second two-dimensional convolution perform integrated feature extraction, and perform matrix multiplication, scale adjustment and classification operations on the extracted results in sequence to obtain first regional feature information, the third two-dimensional convolution layer performs feature extraction, and performs a second matrix multiplication on the original feature information extracted by the third two-dimensional convolution layer and the first regional feature information, performs random branch processing on the predicted features obtained by the second matrix multiplication, constructs a first feature branch and a second feature branch, applies a bounding box prediction loss to optimize the feature data in the first feature branch, applies a classification loss to optimize the feature data in the second feature score, obtains a first optimized feature and a second optimized feature, performs weighted fusion on the first optimized feature and the second optimized feature, obtains regional monitoring optimization feature data, and uses the regional monitoring optimization feature data to improve feature downsampling of the visual monitoring model;

[0026] Step S23: constructing a spatial depth convolution conversion sub-block, specifically using the spatial depth convolution conversion sub-block to replace the traditional pooling downsampling method, performing feature downsampling, and obtaining spatial depth feature conversion data;

[0027] Step S24: construct a multi-level feature fusion sub-block, specifically, decomposing the feature map data in the spatial depth feature conversion data into three feature maps, and sequentially performing 1×1 convolution and two 3×3 convolution operations to obtain refined feature map data, and performing feature connection on the refined feature map data to obtain safe area monitoring feature data, and classifying safe area monitoring by constructing a fully connected layer and a classifier based on the safe area monitoring feature data;

[0028] Step S25: training the safety area monitoring sub-model, specifically, training the safety area monitoring sub-model through the data basic optimization, the improved visual monitoring model, the spatial depth convolution conversion sub-block and the multi-level feature fusion sub-block to obtain the safety area monitoring sub-model Model SA ;

[0029] Step S26: Safe area monitoring, specifically using the safe area monitoring sub-model Model SA , based on the original camera data in the original data of safety monitoring of fishing vessel operators, safety area monitoring is performed to obtain safety area monitoring reference data;

[0030] The safety area monitoring reference data specifically includes the safety area monitoring category and the safety valuation of the safety area monitoring reference data;

[0031] The safety area monitoring categories include high risk, medium risk and low risk;

[0032] The safety valuation of the safety area monitoring reference data ranges from 0 to 100%.

[0033] Further, in step S3, the fishing vessel environment monitoring is used to perform safety assessment on the navigation track of the fishing vessel, specifically, based on the sensor raw data in the fishing vessel operator safety monitoring raw data, a convolutional long short-term memory neural network optimized in combination with the fishing vessel trajectory feature description is used to perform fishing vessel environment monitoring, and obtain fishing vessel environment monitoring reference data and fishing vessel environment monitoring sub-model;

[0034] The clustering classification model combining the optimization of fishing vessel trajectory feature description and convolutional long short-term memory encoding specifically includes a trajectory parameter optimization sub-block, a trajectory feature optimization sub-block, a convolutional long short-term memory subnet, a clustering subnet and a classification subnet;

[0035] The step of using the long short-term memory neural network optimized in combination with the fishing vessel trajectory feature description to monitor the fishing vessel environment and obtain the fishing vessel environment monitoring reference data and the fishing vessel environment monitoring sub-model includes:

[0036] Step S31: trajectory parameter optimization, specifically, performing basic data cleaning and fishing boat trajectory construction based on the sensor raw data in the fishing boat operator safety monitoring raw data to obtain optimized fishing boat trajectory data;

[0037] The basic data cleaning specifically refers to the setting of time interval thresholds for fishing vessel tracks and cleaning of abnormal data;

[0038] The fishing vessel track time interval threshold is set to maintain the continuity and non-overlap of the fishing vessel track data;

[0039] The abnormal data cleaning is specifically performed by constructing fishing vessel trajectories and screening abnormal data to obtain optimized fishing vessel trajectory data;

[0040] The fishing boat trajectory construction is specifically to regard the fishing boat as a motion trajectory particle, and convert the latitude and longitude coordinates of the motion trajectory particle into northeast direction coordinates to construct the fishing boat trajectory;

[0041] The abnormal data screening specifically refers to calculating the motion parameters of the trajectory points according to the timestamps, and deleting the calculated abnormal speed data points;

[0042] Step S32: trajectory feature optimization, specifically, constructing a trajectory feature optimization data set based on the optimized fishing boat trajectory data through artificial feature engineering design, specifically including fishing boat displacement ratio feature, navigation stability feature, trajectory profile feature and minimum moment of inertia feature;

[0043] Step S33: convolutional long short-term memory encoding, specifically constructing a single-layer convolutional long short-term memory neural subnet, performing feature encoding and feature decoding, and obtaining convolutional long short-term feature decoding data;

[0044] Step S34: constructing a clustering subnet, specifically constructing a standard deep clustering neural network based on K-means clustering as a clustering subnet, and performing cluster analysis based on the convolutional long-term and short-term feature decoding data to obtain clustering feature data;

[0045] Step S35: constructing a classification subnet, specifically constructing a standard random forest model as a classifier, and classifying the fishing vessel environment monitoring based on the clustering feature data to obtain classification data on the safety of the fishing vessel trajectory;

[0046] Step S36: Fishing vessel environment monitoring sub-model training, specifically, through the trajectory parameter optimization, the trajectory feature optimization, the long short-term memory encoding, the clustering sub-network and the classification sub-network, the fishing vessel environment monitoring sub-model training is performed to obtain the fishing vessel environment monitoring sub-model Model ED ;

[0047] Step S37: Fishing vessel environment monitoring, specifically using the fishing vessel environment monitoring sub-model ModelED , based on the sensor raw data in the fishing vessel operator safety monitoring raw data, the fishing vessel environment monitoring is performed to obtain the fishing vessel environment monitoring reference data;

[0048] The fishing vessel environmental monitoring reference data specifically include the fishing vessel environmental monitoring category and the safety valuation of the fishing vessel environmental monitoring reference data;

[0049] The above fishing vessel environmental monitoring categories include high risk, medium risk and low risk;

[0050] The safety estimate of the fishing vessel environmental monitoring reference data ranges from 0 to 100%.

[0051] Further, in step S4, the personnel behavior monitoring is used to analyze the behavior of fishing boat operators in non-safe areas, specifically, based on the original camera data and the fishing boat environment monitoring reference data in the original data of the fishing boat operator safety monitoring, a deep attention network combining fuzzy reasoning and pre-trained visual extraction is used to perform personnel behavior monitoring, and obtain personnel behavior monitoring reference data and personnel behavior monitoring sub-model;

[0052] The deep attention network combining fuzzy reasoning and pre-trained visual extraction specifically includes a fuzzy layer, a visual feature extraction layer, a squeeze-excitation feature extraction layer, and a classification output layer;

[0053] The step of using a deep attention network combining fuzzy reasoning and pre-trained visual extraction to monitor personnel behavior and obtain personnel behavior monitoring reference data and a personnel behavior monitoring sub-model includes:

[0054] Step S41: constructing a fuzzy layer, specifically constructing a Gaussian fuzzy membership function, performing fuzziness calculation, and suppressing irrelevant features by applying an aggregation operator or operation to obtain fuzzy layer output dimension data for optimizing the recognition capability of uncertainty data;

[0055] Step S42: constructing a visual feature extraction layer, specifically using the improved visual monitoring model in step S22 as a pre-trained visual feature extraction layer to extract personnel behavior analysis feature data;

[0056] Step S43: constructing a squeeze-excitation feature extraction layer, specifically, performing average pooling, squeeze operations and excitation operations on the personnel behavior analysis feature data in sequence, and applying shortcut connections and nonlinear transformations to enhance the feature data to obtain optimized personnel behavior analysis features;

[0057] Step S44: constructing a classification output layer, specifically constructing a standard softmax classifier as a classification output layer, and classifying the personnel behavior monitoring data according to the optimized personnel behavior analysis features;

[0058] Step S45: training the sub-model for monitoring human behavior, specifically, training the sub-model for monitoring human behavior through the fuzzy layer, the visual feature extraction layer, the squeeze excitation feature extraction layer and the classification output layer, to obtain the sub-model for monitoring human behavior Model BD ;

[0059] Step S46: Personnel behavior monitoring, specifically using the personnel behavior monitoring sub-model Model BD , based on the original camera data and the fishing vessel environment monitoring reference data in the original data of the fishing vessel operator safety monitoring, personnel behavior monitoring is performed to obtain personnel behavior monitoring reference data;

[0060] The personnel behavior monitoring reference data specifically includes the personnel behavior monitoring category and the safety valuation of the personnel behavior monitoring reference data;

[0061] The categories of personnel behavior monitoring include high risk, medium risk and low risk;

[0062] The safety valuation of the personnel behavior monitoring reference data ranges from 0 to 100%.

[0063] Further, in step S5, the comprehensive safety monitoring is used to conduct weighted comprehensive safety monitoring by integrating the three fishing vessel operation safety attributes, specifically combining the safety area monitoring sub-model, the fishing vessel environment monitoring sub-model and the personnel behavior monitoring sub-model, and performing sub-network integrated weighted training to obtain a comprehensive safety monitoring model Model L , and by using the comprehensive safety monitoring model Model L , based on the original data of safety monitoring of fishing vessel operators, real-time safety monitoring of fishing vessel operators is performed to obtain a comprehensive safety situation estimation, and safety situation feedback and alarm of fishing vessel operators are performed based on the comprehensive safety situation estimation;

[0064] The calculation formula for performing subnet integrated weighted training is:

[0065] ;

[0066] Where Y L is the comprehensive security situation estimation, which is used to represent the comprehensive security monitoring output after subnet integration and weighting. is the safety area monitoring weight, Y sa is the safety estimate in the safety area monitoring reference data, is the weight of fishing vessel environmental monitoring, Y ed is the safety estimate in the reference data of fishing vessel environmental monitoring, is the weight of personnel behavior monitoring, Y bdis the safety estimate in the personnel behavior monitoring reference data, where .

[0067] The invention provides a fishing vessel operator safety monitoring system based on artificial intelligence, comprising a basic data acquisition module, a safety area monitoring module, a fishing vessel environment monitoring module, a personnel behavior monitoring module and a comprehensive safety monitoring module;

[0068] The basic data acquisition module is used for basic data acquisition, and obtains the original data of safety monitoring of fishing vessel operators through basic data acquisition, and sends the original data of safety monitoring of fishing vessel operators to the safety area monitoring module, the fishing vessel environment monitoring module and the personnel behavior monitoring module;

[0069] The safety area monitoring module is used for safety area monitoring, obtains safety area monitoring reference data and safety area monitoring sub-model through safety area monitoring, and sends the safety area monitoring sub-model to the comprehensive safety monitoring module;

[0070] The fishing vessel environment monitoring module is used for fishing vessel environment monitoring. Through fishing vessel environment monitoring, fishing vessel environment monitoring reference data and fishing vessel environment monitoring sub-model are obtained, and the fishing vessel environment monitoring reference data is sent to the personnel behavior monitoring module, and the fishing vessel environment monitoring sub-model is sent to the comprehensive safety monitoring module;

[0071] The personnel behavior monitoring module is used for personnel behavior monitoring, obtains personnel behavior monitoring reference data and personnel behavior monitoring sub-model through personnel behavior monitoring, and sends the personnel behavior monitoring sub-model to the comprehensive safety monitoring module;

[0072] The comprehensive safety monitoring module is used for comprehensive safety monitoring. Through comprehensive safety monitoring, a comprehensive safety situation estimation is obtained, and safety situation feedback and alarms are provided to fishing vessel operators based on the comprehensive safety situation estimation.

[0073] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0074] (1) In the existing safety monitoring of fishing vessel operators, traditional intelligent technical solutions often only perform single image analysis or single numerical analysis, which leads to the technical problems that the intelligent data support for the safety monitoring of fishing vessel operators is not comprehensive enough and has low reference value. This solution creatively adopts an integrated safety analysis solution that combines the safety area monitoring sub-model, the fishing vessel environment monitoring sub-model and the personnel behavior monitoring sub-model to perform subnet integrated weighted training. By combining the three aspects of fishing vessel safety area, fishing vessel driving trajectory environment and personnel behavior analysis, the comprehensiveness of the safety monitoring of fishing vessel operators is improved. By weighted processing, the intelligence and reliability of the safety monitoring of fishing vessel operators are also improved. At the same time, the design of the sub-item sub-model is also conducive to the separate operation and maintenance of sub-tasks, which improves the overall stability and scalability of the system.

[0075] (2) In view of the technical problem that the existing standard image detection model has low computational efficiency and poor recognition of smaller targets in dealing with complex fishing vessel safety area monitoring tasks, which is likely to lead to poor safety area monitoring results and thus affect the safety risks of fishing vessel operators, this solution creatively adopts an improved visual detection model combined with a multi-level feature processing method to perform safety area monitoring. By improving the model structure, the small target recognition capability and overall computational efficiency are enhanced, and the overall usability of safety area monitoring is improved through multi-level feature fusion.

[0076] (3) In view of the technical problem that the traditional environmental detection only conducts a general intelligent analysis of the external environment and ignores the possible safety problems in the navigation process of the fishing vessel itself, this scheme creatively adopts the convolutional long short-term memory neural network optimized by combining the fishing vessel trajectory feature description to monitor the fishing vessel environment. By designing four types of fishing vessel navigation trajectory environmental features, combining the convolutional long short-term network and the clustering classification network, complex fishing vessel environment monitoring is realized, which provides strong support for subsequent comprehensive monitoring and personnel behavior monitoring.

[0077] (4) In view of the technical problem that existing personnel behavior monitoring methods involve both intuitive visual features and other potential features that cannot be directly observed through camera data, this scheme creatively adopts a deep attention network that combines fuzzy reasoning and pre-trained visual extraction to monitor personnel behavior. It combines the results of fishing vessel environmental monitoring and the model improvement of safe area monitoring, and improves the ability to capture the potential logic of personnel behavior analysis, thereby improving the overall practicality of safety monitoring of fishing vessel operators. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] Figure 1A schematic diagram of a flow chart of a method for safety monitoring of fishing vessel operators based on artificial intelligence provided by the present invention;

[0079] Figure 2 A schematic diagram of a fishing vessel operator safety monitoring system based on artificial intelligence provided by the present invention;

[0080] Figure 3 This is a schematic diagram of the process of safety area monitoring in step S2;

[0081] Figure 4 This is a schematic diagram of the process of fishing vessel environmental monitoring in step S3;

[0082] Figure 5 This is a flow chart of personnel behavior monitoring in step S4.

[0083] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0084] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0085] In the description of the present invention, it should be understood that terms such as “upper”, “lower”, “front”, “back”, “left”, “right”, “top”, “bottom”, “inside” and “outside” indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as limiting the present invention.

[0086] Example 1, see Figure 1 The present invention provides a method for safety monitoring of fishing vessel operators based on artificial intelligence, the method comprising the following steps:

[0087] Step S1: basic data collection;

[0088] Step S2: safe area monitoring;

[0089] Step S3: Fishing vessel environment monitoring;

[0090] Step S4: personnel behavior monitoring;

[0091] Step S5: Comprehensive safety monitoring.

[0092] Example 2, see Figure 1 , Figure 2 and Figure 3 In step S1, the basic data collection is used to collect the original data set required for the safety monitoring of fishing vessel operators, specifically, the fishing vessel safety monitoring data is collected through the camera device and the sensor to obtain the original data of the safety monitoring of fishing vessel operators;

[0093] The camera device is specifically placed in the cockpit, deck and stern of the fishing boat for monitoring safety areas and personnel behavior;

[0094] The sensors specifically include temperature sensors, humidity sensors, gas sensors, acceleration sensors and distance recognition communication sensors, which are used to monitor environmental conditions and hull status to assist in environmental monitoring of fishing vessels;

[0095] The raw data of safety monitoring of fishing vessel operators specifically includes raw video data and raw sensor data;

[0096] The original camera data specifically includes area division data and personnel behavior data;

[0097] The sensor raw data specifically includes temperature data, humidity data, gas data, fishing boat motion state data and distance recognition data.

[0098] By executing the above operations, in view of the fact that in the existing supply chain management methods, supply chain data generally comes from different special processing systems, which leads to the technical problem that supply chain management needs to face greater data complexity integration and more difficult decision-making, this solution creatively adopts the process design of demand forecasting, inventory management optimization and operation optimization in sequence, which can better handle complex integrated supply chain data and execute various specific management tasks, provide a basis for subsequent inventory management and operation management by predicting market demand, and improve the quality of supply chain management through inventory management and operation management. At the same time, the sub-item functional design is also helpful for the operation and maintenance of the system.

[0099] Example 3, see Figure 1 , Figure 2 and Figure 3 This embodiment is based on the above embodiment. In step S2, the safety area monitoring is used to screen and pre-classify the safety area for fishing vessel operations. Specifically, based on the original camera data in the original data of safety monitoring of fishing vessel operators, a multi-level feature fusion improved visual monitoring model combined with small target optimization is used to perform safety area monitoring, and obtain safety area monitoring reference data and safety area monitoring sub-model;

[0100] The multi-level feature fusion improved visual monitoring model combined with small target optimization specifically includes an improved visual monitoring model, a spatial depth convolution conversion sub-block, and a multi-level feature fusion sub-block;

[0101] The step of using a multi-level feature fusion improved visual monitoring model combined with small target optimization to monitor the safe area and obtain safe area monitoring reference data and a safe area monitoring sub-model includes:

[0102] Step S21: basic data optimization, specifically, performing data enhancement on the original camera data through basic image enhancement and filtering denoising to obtain optimized safe area monitoring data;

[0103] The basic image enhancement specifically refers to random rotation, scaling and flipping data enhancement operations;

[0104] The filtering and denoising specifically refers to a Gaussian filtering operation;

[0105] Step S22: improving the visual monitoring model, specifically selecting a pre-trained standard YOLOv8 model as a basic visual monitoring model, and improving the hierarchical structure of the basic visual monitoring model to obtain an improved visual monitoring model;

[0106] The hierarchical structure improvement specifically refers to deleting the seventh and eighth downsampling layers of the standard YOLOv8 model backbone network to optimize computational efficiency, and adding a regional monitoring optimization feature processing sub-block to the first convolutional layer of the standard YOLOv8 model backbone network to enhance the improved visual monitoring model's ability to recognize smaller targets in safe area monitoring;

[0107] The regional monitoring optimization feature processing sub-block specifically includes a first two-dimensional convolution layer, a second two-dimensional convolution layer, a third two-dimensional convolution layer and a branch processing layer, the first two-dimensional convolution layer and the second two-dimensional convolution perform integrated feature extraction, and perform matrix multiplication, scale adjustment and classification operations on the extracted results in sequence to obtain first regional feature information, the third two-dimensional convolution layer performs feature extraction, and performs a second matrix multiplication on the original feature information extracted by the third two-dimensional convolution layer and the first regional feature information, performs random branch processing on the predicted features obtained by the second matrix multiplication, constructs a first feature branch and a second feature branch, applies a bounding box prediction loss to optimize the feature data in the first feature branch, applies a classification loss to optimize the feature data in the second feature score, obtains a first optimized feature and a second optimized feature, performs weighted fusion on the first optimized feature and the second optimized feature to obtain regional monitoring optimization feature data, and uses the regional monitoring optimization feature data to improve feature downsampling of the visual monitoring model;

[0108] Step S23: construct a spatial depth convolution conversion sub-block, specifically, use the spatial depth convolution conversion sub-block to replace the traditional pooling downsampling method, perform feature downsampling, and obtain spatial depth feature conversion data. The calculation formula is:

[0109] ;

[0110] Where Y i,j,k is the feature output corresponding to the feature map position (i, j) and the kth feature channel. i,j,k Perform point-by-point convolution operations to obtain spatial depth feature conversion data, where i is the horizontal pixel index of the feature map, j is the vertical pixel index of the feature map, k is the feature channel index, and K H is the depth convolution kernel height, m ​​is the depth convolution kernel height index, K W is the depth convolution kernel width, n is the depth convolution kernel width index, W m,n,k is the convolution weight, X i+m,j+n+k is a spatial deep convolution input feature, used to represent the regional monitoring optimization feature data;

[0111] Step S24: construct a multi-level feature fusion sub-block, specifically, decomposing the feature map data in the spatial depth feature conversion data into three feature maps, and sequentially performing 1×1 convolution and two 3×3 convolution operations to obtain refined feature map data, and performing feature connection on the refined feature map data to obtain safe area monitoring feature data, and classifying safe area monitoring by constructing a fully connected layer and a classifier based on the safe area monitoring feature data;

[0112] Step S25: training the safety area monitoring sub-model, specifically, training the safety area monitoring sub-model through the data basic optimization, the improved visual monitoring model, the spatial depth convolution conversion sub-block and the multi-level feature fusion sub-block to obtain the safety area monitoring sub-model Model SA ;

[0113] Step S26: Safe area monitoring, specifically using the safe area monitoring sub-model Model SA , based on the original camera data in the original data of safety monitoring of fishing vessel operators, safety area monitoring is performed to obtain safety area monitoring reference data;

[0114] The safety area monitoring reference data specifically includes the safety area monitoring category and the safety valuation of the safety area monitoring reference data;

[0115] The safety area monitoring categories include high risk, medium risk and low risk;

[0116] The safety valuation of the safety area monitoring reference data ranges from 0 to 100%.

[0117] By performing the above operations, in order to address the technical problem that in the existing safety area monitoring methods, the existing standard image detection model has low computational efficiency and poor recognition of smaller targets in dealing with complex fishing vessel safety area monitoring tasks, which is likely to lead to poor safety area monitoring effects and thus affect the safety risks of fishing vessel operators, this scheme creatively adopts an improved visual detection model combined with a multi-level feature processing method to perform safety area monitoring, enhances the small target recognition capability and overall computational efficiency by improving the model structure, and improves the overall usability of safety area monitoring through multi-level feature fusion.

[0118] Example 4, see Figure 1 , Figure 2 and Figure 4 This embodiment is based on the above embodiment. In step S3, the fishing vessel environment monitoring is used to perform safety assessment on the navigation track of the fishing vessel. Specifically, the fishing vessel environment monitoring is performed based on the sensor raw data in the fishing vessel operator safety monitoring raw data, and a convolutional long short-term memory neural network optimized by combining the fishing vessel trajectory feature description is used to obtain the fishing vessel environment monitoring reference data and the fishing vessel environment monitoring sub-model;

[0119] The clustering classification model combining the optimization of fishing vessel trajectory feature description and convolutional long short-term memory encoding specifically includes a trajectory parameter optimization sub-block, a trajectory feature optimization sub-block, a convolutional long short-term memory subnet, a clustering subnet and a classification subnet;

[0120] The step of using the long short-term memory neural network optimized in combination with the fishing vessel trajectory feature description to monitor the fishing vessel environment and obtain the fishing vessel environment monitoring reference data and the fishing vessel environment monitoring sub-model includes:

[0121] Step S31: trajectory parameter optimization, specifically, performing basic data cleaning and fishing boat trajectory construction based on the sensor raw data in the fishing boat operator safety monitoring raw data to obtain optimized fishing boat trajectory data;

[0122] The basic data cleaning specifically refers to the setting of time interval thresholds for fishing vessel tracks and cleaning of abnormal data;

[0123] The fishing vessel track time interval threshold is set to maintain the continuity and non-overlap of the fishing vessel track data;

[0124] The abnormal data cleaning is specifically performed by constructing fishing vessel trajectories and screening abnormal data to obtain optimized fishing vessel trajectory data;

[0125] The fishing boat trajectory construction is specifically to regard the fishing boat as a motion trajectory particle, and convert the latitude and longitude coordinates of the motion trajectory particle into northeast direction coordinates to construct the fishing boat trajectory;

[0126] The abnormal data screening specifically refers to calculating the motion parameters of the trajectory points according to the timestamps, and deleting the calculated abnormal speed data points;

[0127] Step S32: trajectory feature optimization, specifically, constructing a trajectory feature optimization data set based on the optimized fishing boat trajectory data through artificial feature engineering design, specifically including fishing boat displacement ratio feature, navigation stability feature, trajectory profile feature and minimum moment of inertia feature;

[0128] The calculation formula of the displacement ratio characteristic of the fishing boat is:

[0129] ;

[0130] Where D R is the displacement ratio characteristic of the fishing boat, T disp is the vector length from the starting point to the ending point, T dist is the cumulative vector length of all trajectory segments;

[0131] The calculation formula of the navigation stability characteristic is:

[0132] ;

[0133] In the formula, C R is the navigation stability characteristic, T is the total number of track points, t is the track point index, cog t is the trajectory segment direction of the t-th trajectory point, I t is the trajectory point duration;

[0134] The calculation formula of the trajectory profile feature is:

[0135] ;

[0136] In the formula, O R is the trajectory profile feature, N O is the number of grids covered by the track segment on the fishing boat grid map, N ng is the number of repeated entries into the grid;

[0137] The calculation formula of the minimum moment of inertia characteristic is:

[0138] ;

[0139] Where M R is the minimum moment of inertia feature, T is the total number of trajectory points, t is the trajectory point index, V is the total amount of the rotational inertia axis, v is the rotational inertia axis index, dtv is the distance from the tth trajectory point to the vth axis of inertia;

[0140] The calculation formula for constructing the trajectory feature optimization data set is:

[0141] R R ={D R ;C R ;O R ;M R};

[0142] In the formula, R R is the trajectory feature optimization dataset, D R is the displacement ratio characteristic of the fishing boat, C R is the navigation stability characteristic, O R is the trajectory profile feature, M R is the minimum moment of inertia characteristic;

[0143] Step S33: convolutional long short-term memory encoding, specifically constructing a single-layer convolutional long short-term memory neural subnet, performing feature encoding and feature decoding, and obtaining convolutional long short-term feature decoding data;

[0144] Step S34: constructing a clustering subnet, specifically constructing a standard deep clustering neural network based on K-means clustering as a clustering subnet, and performing cluster analysis based on the convolutional long-term and short-term feature decoding data to obtain clustering feature data;

[0145] Step S35: constructing a classification subnet, specifically constructing a standard random forest model as a classifier, and classifying the fishing vessel environment monitoring based on the clustering feature data to obtain classification data on the safety of the fishing vessel trajectory;

[0146] Step S36: Fishing vessel environment monitoring sub-model training, specifically, through the trajectory parameter optimization, the trajectory feature optimization, the long short-term memory encoding, the clustering sub-network and the classification sub-network, the fishing vessel environment monitoring sub-model training is performed to obtain the fishing vessel environment monitoring sub-model Model ED ;

[0147] Step S37: Fishing vessel environment monitoring, specifically using the fishing vessel environment monitoring sub-model Model ED , based on the sensor raw data in the fishing vessel operator safety monitoring raw data, the fishing vessel environment monitoring is performed to obtain the fishing vessel environment monitoring reference data;

[0148] The fishing vessel environmental monitoring reference data specifically include the fishing vessel environmental monitoring category and the safety valuation of the fishing vessel environmental monitoring reference data;

[0149] The above fishing vessel environmental monitoring categories include high risk, medium risk and low risk;

[0150] The safety estimate of the fishing vessel environmental monitoring reference data ranges from 0 to 100%.

[0151] By executing the above operations, in order to address the technical problem that in the existing fishing vessel environment monitoring methods, traditional environmental detection only conducts a general intelligent analysis of the external environment, while ignoring the possible safety issues in the fishing vessel navigation process itself, this scheme creatively adopts a convolutional long short-term memory neural network optimized with the fishing vessel trajectory feature description to monitor the fishing vessel environment. By designing four types of fishing vessel navigation trajectory environmental characteristics, combining the convolutional long short-term network and the clustering classification network, complex fishing vessel environment monitoring is realized, which provides strong support for subsequent comprehensive monitoring and personnel behavior monitoring.

[0152] Example 5, see Figure 1 , Figure 2 and Figure 5 This embodiment is based on the above embodiment. In step S4, the personnel behavior monitoring is used to analyze the behavior of fishing boat operators in non-safe areas. Specifically, based on the original camera data and the fishing boat environment monitoring reference data in the original data of the fishing boat operator safety monitoring, a deep attention network combining fuzzy reasoning and pre-trained visual extraction is used to perform personnel behavior monitoring to obtain personnel behavior monitoring reference data and a personnel behavior monitoring sub-model;

[0153] The deep attention network combining fuzzy reasoning and pre-trained visual extraction specifically includes a fuzzy layer, a visual feature extraction layer, a squeeze-excitation feature extraction layer, and a classification output layer;

[0154] The step of using a deep attention network combining fuzzy reasoning and pre-trained visual extraction to monitor personnel behavior and obtain personnel behavior monitoring reference data and a personnel behavior monitoring sub-model includes:

[0155] Step S41: constructing a fuzzy layer, specifically constructing a Gaussian fuzzy membership function, performing fuzziness calculation, and suppressing irrelevant features by applying an aggregation operator or operation to obtain fuzzy layer output dimension data for optimizing the recognition capability of uncertainty data;

[0156] The Gaussian blur membership function is constructed and the calculation formula for fuzziness calculation is:

[0157] ;

[0158] In the formula, is the output value of the Gaussian fuzzy membership function, which is used to apply the aggregation operator or operation and obtain the fuzzy layer output dimension data, where X is the fuzzy layer input data, which is used to represent the original camera data and the fishing vessel environment monitoring reference data after artificial feature processing and connection operation. is the mean of the Gaussian function, is the standard deviation of the Gaussian function, exp(·) is the natural base function, is the input data index, is the Gaussian fuzzy set index;

[0159] Step S42: constructing a visual feature extraction layer, specifically using the improved visual monitoring model in step S22 as a pre-trained visual feature extraction layer to extract personnel behavior analysis feature data;

[0160] Step S43: constructing a squeeze-excitation feature extraction layer, specifically, performing average pooling, squeeze operations and excitation operations on the personnel behavior analysis feature data in sequence, and applying shortcut connections and nonlinear transformations to enhance the feature data to obtain optimized personnel behavior analysis features;

[0161] Step S44: constructing a classification output layer, specifically constructing a standard softmax classifier as a classification output layer, and classifying the personnel behavior monitoring data according to the optimized personnel behavior analysis features;

[0162] Step S45: training the sub-model for monitoring human behavior, specifically, training the sub-model for monitoring human behavior through the fuzzy layer, the visual feature extraction layer, the squeeze excitation feature extraction layer and the classification output layer, to obtain the sub-model for monitoring human behavior Model BD ;

[0163] Step S46: Personnel behavior monitoring, specifically using the personnel behavior monitoring sub-model Model BD , based on the original camera data and the fishing vessel environment monitoring reference data in the original data of the fishing vessel operator safety monitoring, personnel behavior monitoring is performed to obtain personnel behavior monitoring reference data;

[0164] The personnel behavior monitoring reference data specifically includes the personnel behavior monitoring category and the safety valuation of the personnel behavior monitoring reference data;

[0165] The categories of personnel behavior monitoring include high risk, medium risk and low risk;

[0166] The safety valuation of the personnel behavior monitoring reference data ranges from 0 to 100%.

[0167] By performing the above operations, in order to address the technical problem that in existing personnel behavior monitoring methods, personnel behavior monitoring involves both intuitive visual features and other potential features that cannot be directly observed through camera data, this solution creatively adopts a deep attention network that combines fuzzy reasoning and pre-trained visual extraction to perform personnel behavior monitoring, combines the results of fishing vessel environmental monitoring and the model improvement of safe area monitoring, and improves the ability to capture the potential logic of personnel behavior analysis, thereby improving the overall practicality of safety monitoring of fishing vessel operators.

[0168] Example 6, see Figure 1 and Figure 2 This embodiment is based on the above embodiment. In step S5, the comprehensive safety monitoring is used to comprehensively analyze the three fishing vessel operation safety attributes for weighted comprehensive safety monitoring. Specifically, the safety area monitoring sub-model, the fishing vessel environment monitoring sub-model and the personnel behavior monitoring sub-model are combined to perform sub-network integrated weighted training to obtain a comprehensive safety monitoring model Model L , and by using the comprehensive safety monitoring model Model L , based on the original data of safety monitoring of fishing vessel operators, real-time safety monitoring of fishing vessel operators is performed to obtain a comprehensive safety situation estimation, and safety situation feedback and alarm of fishing vessel operators are performed based on the comprehensive safety situation estimation;

[0169] The calculation formula for performing subnet integrated weighted training is:

[0170] ;

[0171] Where Y L is the comprehensive security situation estimation, which is used to represent the comprehensive security monitoring output after subnet integration and weighting. is the safety area monitoring weight, Y sa is the safety estimate in the safety area monitoring reference data, is the weight of fishing vessel environmental monitoring, Y ed is the safety estimate in the reference data of fishing vessel environmental monitoring, is the weight of personnel behavior monitoring, Y bd is the safety estimate in the personnel behavior monitoring reference data, where .

[0172] Table 1 is a reference data table for comprehensive safety situation valuation;

[0173] As shown in the table, the monitoring content is listed as the specific task content for comprehensive safety monitoring, the valuation range is listed as the specific value range of the safety valuation corresponding to the monitoring content, the remarks are listed as the specific description of the safety valuation corresponding to the monitoring content, and the alarm level is listed as the alarm level for safety situation feedback and alarm for the safety valuation corresponding to the monitoring content. When the alarm level is high, a safety alarm is issued, and when the alarm level is medium, safety situation feedback is issued.

[0174] Table 1 Comprehensive safety situation valuation reference data table

[0175]

[0176] Embodiment 7, see Figure 1 and Figure 2 , this embodiment is based on the above embodiment, and the present invention provides an artificial intelligence-based fishing vessel operator safety monitoring system, including a basic data acquisition module, a safety area monitoring module, a fishing vessel environment monitoring module, a personnel behavior monitoring module and a comprehensive safety monitoring module;

[0177] The basic data acquisition module is used for basic data acquisition, and obtains the original data of safety monitoring of fishing vessel operators through basic data acquisition, and sends the original data of safety monitoring of fishing vessel operators to the safety area monitoring module, the fishing vessel environment monitoring module and the personnel behavior monitoring module;

[0178] The safety area monitoring module is used for safety area monitoring, obtains safety area monitoring reference data and safety area monitoring sub-model through safety area monitoring, and sends the safety area monitoring sub-model to the comprehensive safety monitoring module;

[0179] The fishing vessel environment monitoring module is used for fishing vessel environment monitoring. Through fishing vessel environment monitoring, fishing vessel environment monitoring reference data and fishing vessel environment monitoring sub-model are obtained, and the fishing vessel environment monitoring reference data is sent to the personnel behavior monitoring module, and the fishing vessel environment monitoring sub-model is sent to the comprehensive safety monitoring module;

[0180] The personnel behavior monitoring module is used for personnel behavior monitoring, obtains personnel behavior monitoring reference data and personnel behavior monitoring sub-model through personnel behavior monitoring, and sends the personnel behavior monitoring sub-model to the comprehensive safety monitoring module;

[0181] The comprehensive safety monitoring module is used for comprehensive safety monitoring. Through comprehensive safety monitoring, a comprehensive safety situation estimation is obtained, and safety situation feedback and alarms are provided to fishing vessel operators based on the comprehensive safety situation estimation.

[0182] By executing the above operations, in the existing safety monitoring of fishing vessel operators, there is a problem that traditional intelligent technical solutions often only perform a single image analysis or a single numerical analysis, which leads to the intelligent data support for the safety monitoring of fishing vessel operators, and inevitably has technical problems such as incomplete analysis and low referenceability. This solution creatively adopts an integrated safety analysis solution that combines the safety area monitoring sub-model, the fishing vessel environment monitoring sub-model and the personnel behavior monitoring sub-model to perform subnet integrated weighted training. By combining the three aspects of fishing vessel safety area, fishing vessel driving trajectory environment and personnel behavior analysis, the comprehensiveness of fishing vessel operator safety monitoring is improved. By weighted processing, the intelligence and reliability of fishing vessel operator safety monitoring are also improved. At the same time, the design of sub-item sub-models is also conducive to the separate operation and maintenance of sub-tasks, which improves the overall stability and scalability of the system.

[0183] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0184] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that many changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the invention.

[0185] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. A safety monitoring method for fishing vessel operators based on artificial intelligence, characterized in that: The method comprises the following steps: Step S1: basic data collection, obtaining the original data of safety monitoring of fishing vessel operators; Step S2: Safe area monitoring: using a multi-level feature fusion improved visual monitoring model combined with small target optimization to perform safe area monitoring, and obtain safe area monitoring reference data and safe area monitoring sub-model; Step S3: Fishing vessel environment monitoring, using a convolutional long short-term memory neural network optimized by combining the fishing vessel trajectory feature description to perform fishing vessel environment monitoring, and obtain fishing vessel environment monitoring reference data and fishing vessel environment monitoring sub-model; Step S4: Personnel behavior monitoring, using a deep attention network that combines fuzzy reasoning and pre-trained visual extraction to perform personnel behavior monitoring, and obtain personnel behavior monitoring reference data and personnel behavior monitoring sub-model; Step S5: Comprehensive safety monitoring: Combine the safety area monitoring sub-model, the fishing vessel environment monitoring sub-model and the personnel behavior monitoring sub-model to perform sub-network integrated weighted training to obtain a comprehensive safety monitoring model Model L , and by using the comprehensive safety monitoring model Model L Based on the original data of safety monitoring of fishing vessel operators, real-time safety monitoring of fishing vessel operators is carried out to obtain a comprehensive safety situation estimation, and safety situation feedback and alarm of fishing vessel operators are carried out based on the comprehensive safety situation estimation.

2. The artificial intelligence-based safety monitoring method for fishing vessel operators according to claim 1 is characterized in that: In step S1, the basic data collection is used to collect the original data set required for the safety monitoring of fishing vessel operators, specifically, the fishing vessel safety monitoring data is collected through a camera device and a sensor to obtain the original data of the safety monitoring of fishing vessel operators; The raw data of safety monitoring of fishing vessel operators specifically includes raw video data and raw sensor data.

3. The artificial intelligence-based safety monitoring method for fishing vessel operators according to claim 2 is characterized in that: In step S2, the safety area monitoring is used to screen and pre-classify the safety area for fishing boat operations, and the multi-level feature fusion improved visual monitoring model combined with small target optimization specifically includes an improved visual monitoring model, a spatial depth convolution conversion sub-block, and a multi-level feature fusion sub-block; The step of using a multi-level feature fusion improved visual monitoring model combined with small target optimization to monitor the safe area and obtain safe area monitoring reference data and a safe area monitoring sub-model includes: Step S21: basic data optimization, specifically, performing data enhancement on the original camera data through basic image enhancement and filtering denoising to obtain optimized safe area monitoring data; Step S22: improving the visual monitoring model, specifically selecting a pre-trained standard YOLOv8 model as a basic visual monitoring model, and improving the hierarchical structure of the basic visual monitoring model to obtain an improved visual monitoring model; The hierarchical structure improvement specifically refers to deleting the seventh and eighth downsampling layers of the standard YOLOv8 model backbone network to optimize computational efficiency, and adding a regional monitoring optimization feature processing sub-block to the first convolutional layer of the standard YOLOv8 model backbone network to enhance the improved visual monitoring model's ability to recognize smaller targets in safe area monitoring; Step S23: constructing a spatial depth convolution conversion sub-block, specifically using the spatial depth convolution conversion sub-block to replace the traditional pooling downsampling method, performing feature downsampling, and obtaining spatial depth feature conversion data; Step S24: construct a multi-level feature fusion sub-block, specifically, decomposing the feature map data in the spatial depth feature conversion data into three feature maps, and sequentially performing 1×1 convolution and two 3×3 convolution operations to obtain refined feature map data, and performing feature connection on the refined feature map data to obtain safe area monitoring feature data, and classifying safe area monitoring by constructing a fully connected layer and a classifier based on the safe area monitoring feature data; Step S25: training the safety area monitoring sub-model, specifically, training the safety area monitoring sub-model through the data basic optimization, the improved visual monitoring model, the spatial depth convolution conversion sub-block and the multi-level feature fusion sub-block to obtain the safety area monitoring sub-model Model SA ; Step S26: Safe area monitoring, specifically using the safe area monitoring sub-model Model SA , based on the original camera data in the original data of safety monitoring of fishing vessel operators, safety area monitoring is performed to obtain safety area monitoring reference data; The safety area monitoring reference data specifically includes a safety area monitoring category and a safety valuation of the safety area monitoring reference data.

4. The artificial intelligence-based safety monitoring method for fishing vessel operators according to claim 3 is characterized in that: In step S3, the fishing vessel environment monitoring is used to perform safety assessment on the navigation track of the fishing vessel, specifically, based on the sensor raw data in the fishing vessel operator safety monitoring raw data, a convolutional long short-term memory neural network optimized in combination with the fishing vessel trajectory feature description is used to perform fishing vessel environment monitoring, and obtain fishing vessel environment monitoring reference data and fishing vessel environment monitoring sub-model; The clustering classification model combining the optimization of fishing vessel trajectory feature description and convolutional long short-term memory encoding specifically includes a trajectory parameter optimization sub-block, a trajectory feature optimization sub-block, a convolutional long short-term memory subnet, a clustering subnet and a classification subnet; The step of using the long short-term memory neural network optimized in combination with the fishing vessel trajectory feature description to monitor the fishing vessel environment and obtain the fishing vessel environment monitoring reference data and the fishing vessel environment monitoring sub-model includes: Step S31: trajectory parameter optimization, specifically, performing basic data cleaning and fishing boat trajectory construction based on the sensor raw data in the fishing boat operator safety monitoring raw data to obtain optimized fishing boat trajectory data; Step S32: trajectory feature optimization, specifically, constructing a trajectory feature optimization data set based on the optimized fishing boat trajectory data through artificial feature engineering design, specifically including fishing boat displacement ratio feature, navigation stability feature, trajectory profile feature and minimum moment of inertia feature; Step S33: convolutional long short-term memory encoding, specifically constructing a single-layer convolutional long short-term memory neural subnet, performing feature encoding and feature decoding, and obtaining convolutional long short-term feature decoding data; Step S34: constructing a clustering subnet, specifically constructing a standard deep clustering neural network based on K-means clustering as a clustering subnet, and performing cluster analysis based on the convolutional long-term and short-term feature decoding data to obtain clustering feature data; Step S35: constructing a classification subnet, specifically constructing a standard random forest model as a classifier, and classifying the fishing vessel environment monitoring based on the clustering feature data to obtain classification data on the safety of the fishing vessel trajectory; Step S36: Fishing vessel environment monitoring sub-model training, specifically, through the trajectory parameter optimization, the trajectory feature optimization, the long short-term memory encoding, the clustering sub-network and the classification sub-network, the fishing vessel environment monitoring sub-model training is performed to obtain the fishing vessel environment monitoring sub-model Model ED ; Step S37: Fishing vessel environment monitoring, specifically using the fishing vessel environment monitoring sub-model Model ED , based on the sensor raw data in the fishing vessel operator safety monitoring raw data, the fishing vessel environment monitoring is performed to obtain the fishing vessel environment monitoring reference data; The fishing vessel environmental monitoring reference data specifically include fishing vessel environmental monitoring categories and safety valuations of the fishing vessel environmental monitoring reference data.

5. The artificial intelligence-based safety monitoring method for fishing vessel operators according to claim 4 is characterized in that: In step S4, the personnel behavior monitoring is used to analyze the behavior of fishing boat operators in non-safe areas, specifically, based on the original camera data and fishing boat environment monitoring reference data in the original data of fishing boat operator safety monitoring, a deep attention network combining fuzzy reasoning and pre-trained visual extraction is used to perform personnel behavior monitoring, and obtain personnel behavior monitoring reference data and personnel behavior monitoring sub-model; The deep attention network combining fuzzy reasoning and pre-trained visual extraction specifically includes a fuzzy layer, a visual feature extraction layer, a squeeze-excitation feature extraction layer, and a classification output layer; The step of using a deep attention network combining fuzzy reasoning and pre-trained visual extraction to monitor personnel behavior and obtain personnel behavior monitoring reference data and a personnel behavior monitoring sub-model includes: Step S41: constructing a fuzzy layer, specifically constructing a Gaussian fuzzy membership function, performing fuzziness calculation, and suppressing irrelevant features by applying an aggregation operator or operation to obtain fuzzy layer output dimension data for optimizing the recognition capability of uncertainty data; Step S42: constructing a visual feature extraction layer, specifically using the improved visual monitoring model in step S22 as a pre-trained visual feature extraction layer to extract personnel behavior analysis feature data; Step S43: constructing a squeeze-excitation feature extraction layer, specifically, performing average pooling, squeeze operations and excitation operations on the personnel behavior analysis feature data in sequence, and applying shortcut connections and nonlinear transformations to enhance the feature data to obtain optimized personnel behavior analysis features; Step S44: constructing a classification output layer, specifically constructing a standard softmax classifier as a classification output layer, and classifying the personnel behavior monitoring data according to the optimized personnel behavior analysis features; Step S45: training the sub-model for monitoring human behavior, specifically, training the sub-model for monitoring human behavior through the fuzzy layer, the visual feature extraction layer, the squeeze excitation feature extraction layer and the classification output layer, to obtain the sub-model for monitoring human behavior Model BD ; Step S46: Personnel behavior monitoring, specifically using the personnel behavior monitoring sub-model Model BD , based on the original camera data and the fishing vessel environment monitoring reference data in the original data of the fishing vessel operator safety monitoring, personnel behavior monitoring is performed to obtain personnel behavior monitoring reference data; The personnel behavior monitoring reference data specifically includes a personnel behavior monitoring category and a safety valuation of the personnel behavior monitoring reference data.

6. The artificial intelligence-based safety monitoring method for fishing vessel operators according to claim 5 is characterized in that: In step S5, the comprehensive safety monitoring is used to conduct weighted comprehensive safety monitoring by integrating the three fishing vessel operation safety attributes. Specifically, the subnet integration weighted training is performed by combining the safety area monitoring submodel, the fishing vessel environment monitoring submodel and the personnel behavior monitoring submodel to obtain a comprehensive safety monitoring model Model L , and by using the comprehensive safety monitoring model Model L Based on the original data of safety monitoring of fishing vessel operators, real-time safety monitoring of fishing vessel operators is carried out to obtain a comprehensive safety situation estimation, and safety situation feedback and alarm of fishing vessel operators are carried out based on the comprehensive safety situation estimation.

7. The artificial intelligence-based safety monitoring method for fishing vessel operators according to claim 6 is characterized in that: In step S5, the calculation formula for performing subnet integrated weighted training is: ; Where Y L is the comprehensive security situation estimation, which is used to represent the comprehensive security monitoring output after subnet integration and weighting. is the safety area monitoring weight, Y sa is the safety estimate in the safety area monitoring reference data, is the weight of fishing vessel environmental monitoring, Y ed is the safety estimate in the reference data of fishing vessel environmental monitoring, is the weight of personnel behavior monitoring, Y bd It is the safety estimate in the personnel behavior monitoring reference data.

8. An artificial intelligence-based fishing vessel operator safety monitoring system, used to implement an artificial intelligence-based fishing vessel operator safety monitoring method as described in any one of claims 1 to 7, characterized in that: It includes basic data acquisition module, safety area monitoring module, fishing vessel environment monitoring module, personnel behavior monitoring module and comprehensive safety monitoring module.

9. The artificial intelligence-based fishing vessel operator safety monitoring system according to claim 8, characterized in that: The basic data acquisition module is used for basic data acquisition, and obtains the original data of safety monitoring of fishing vessel operators through basic data acquisition, and sends the original data of safety monitoring of fishing vessel operators to the safety area monitoring module, the fishing vessel environment monitoring module and the personnel behavior monitoring module; The safety area monitoring module is used for safety area monitoring, obtains safety area monitoring reference data and safety area monitoring sub-model through safety area monitoring, and sends the safety area monitoring sub-model to the comprehensive safety monitoring module; The fishing vessel environment monitoring module is used for fishing vessel environment monitoring. Through fishing vessel environment monitoring, fishing vessel environment monitoring reference data and fishing vessel environment monitoring sub-model are obtained, and the fishing vessel environment monitoring reference data is sent to the personnel behavior monitoring module, and the fishing vessel environment monitoring sub-model is sent to the comprehensive safety monitoring module; The personnel behavior monitoring module is used for personnel behavior monitoring, obtains personnel behavior monitoring reference data and personnel behavior monitoring sub-model through personnel behavior monitoring, and sends the personnel behavior monitoring sub-model to the comprehensive safety monitoring module; The comprehensive safety monitoring module is used for comprehensive safety monitoring. Through comprehensive safety monitoring, a comprehensive safety situation estimation is obtained, and safety situation feedback and alarms are provided to fishing vessel operators based on the comprehensive safety situation estimation.

Citation Information

Patent Citations

  • Sales risk point detection method, system and device based on semi-supervised learning

    CN109978358A

  • Fishing boat safety monitoring method based on deep learning

    CN112380914A

  • Intelligent ship safety monitoring method and system based on deep learning

    CN115661766A

  • YOLOv8-based corn leaf insect pest area detection method

    CN117475299A

  • Marine gearbox intelligent monitoring system driven by twin data

    CN118009019A