Intelligent inspection method and system for compliance of fishing vessels leaving port

Abstract

The present invention provides an intelligent inspection method and system for the compliance of fishing vessels leaving the port, aiming to improve the efficiency and accuracy of fishing vessel supervision through automation technology. The method uses a pan-tilt camera and a thermal imaging dome camera to collect comprehensive video data of fishing vessels, and accurately identifies fishing vessels and their ship numbers through image recognition technology. The system can feed back identification information to the supervision system in real time, and trigger an early warning mechanism when the identification fails. In addition, the system will automatically check the status of the fishing vessel's communication and navigation equipment and the status of the port entry and exit report review to ensure that the fishing vessel's activities comply with regulatory requirements. Especially for the special situation of going out to sea during the fishing moratorium, the system can perform detailed compliance inspections. Through intelligent monitoring and early warning, the method effectively improves the safety and compliance of fishing vessel departure management, helps protect marine resources and reduce illegal fishing.

Description

Technical Field

[0001] The invention relates to an intelligent inspection method for compliance of fishing boats leaving a port realized by a camera and a port leaving data acquisition system and a system thereof, belonging to the field of fishing boat safety supervision. Background Art

[0002] At present, the fishing vessel safety supervision system mainly focuses on three links: port area, navigation process and operation site. Among them, the compliance inspection before the fishing vessel leaves the port is regarded as a key node in the entire safety supervision chain. It is the primary measure to prevent safety accidents and ensure navigation safety from the root. At present, the safety supervision of leaving the port mainly relies on the positioning device of the fishing vessel. The review of the port entry and exit reporting information and on-site law enforcement are used to evaluate the compliance of the fishing vessel leaving the port. However, this method has limitations and it is difficult to capture certain illegal operations, such as fishing vessels leaving the port without authorization when the positioning equipment is turned off, illegal sea voyage during the fishing moratorium, and illegal departure of "three no" (no ship name and number, no ship certificate, no port of registry) fishing vessels, departure in bad weather, etc. These problems are often difficult to be detected in time in the daily supervision of law enforcement personnel due to the large number of fishing vessels leaving the port, especially in the record verification one by one. There are huge challenges. Summary of the invention

[0003] In order to solve the current problem, the present invention provides an intelligent inspection method for compliance of fishing vessels leaving the port. The technical solution of the present invention is:

[0004] An intelligent inspection method for compliance of fishing vessels leaving the port includes the following steps: (1) Information collection stage: using a pan-tilt camera and a thermal imaging dome camera to collect all-round video data of dynamic fishing vessels about to go out to sea; (2) Ship and ship number identification and feedback: based on the collected video stream data information, the image recognition technology is used to accurately identify and analyze the fishing vessel and the ship number, and then the successfully identified ship number information is fed back to the fishing vessel supervision system in real time; (3) Early warning mechanism: if the system fails to successfully identify the fishing vessel number, the early warning mechanism is immediately triggered, the early warning information is generated, and the relevant staff is quickly notified so that timely response measures can be taken; (4) Compliance inspection and problem notification: if the ship number is successfully identified, the fishing vessel communication and navigation equipment corresponding to the ship number is automatically associated, and its self-inspection record is retrieved and checked. At the same time, the switch status of AIS (Automatic Identification System) and Beidou key navigation equipment is confirmed to ensure that the equipment is in normal working condition;

[0005] Through the fishing port entry and exit reporting module of the fishing vessel supervision system, check the status of fishing vessel entry and exit reporting, including whether it is within the fishing moratorium period, whether the relevant reporting information has been completed, etc. Pay special attention to special circumstances during the fishing moratorium period, such as the legal sea records of aquaculture vessels or rescue vessels, to ensure that all activities comply with regulations;

[0006] When any problems are found, such as equipment failure, unreported or illegal sailing, etc., notifications will be sent to staff immediately for prompt processing.

[0007] The step (2) is specifically as follows:

[0008] (2-1) Video data acquisition: The PTZ camera works in conjunction with the thermal imaging dome camera and AI system. The AI ​​system controls the camera to collect high-definition video data of fishing boats about to go out to sea. The PTZ camera can capture clear images of fishing boats from different angles with its flexible rotation capability. The thermal imaging dome camera uses thermal imaging technology to obtain heat source images of fishing boats at night or in bad weather conditions, ensuring the comprehensiveness and accuracy of the data.

[0009] (2-2) Fishing vessel identification: By extensively collecting representative fishing vessel image datasets, we ensure that all types of fishing vessels that need to be identified are covered, providing a solid foundation for model training. Subsequently, the collected images are preprocessed, including scaling, cropping, rotating, and grayscale conversion, to unify the format and optimize the image quality.

[0010] Then, SIFT, HOG or LBP feature extraction algorithms are used to accurately extract color, shape, texture and key features of shipboard equipment from the preprocessed images, and redundant information is removed through the feature selection process to retain the features that are critical to classification;

[0011] Subsequently, a classification model is constructed and trained based on the selected features, so that the model can learn the relationship between the features and the types of fishing vessels. After the model training is completed, its classification performance is evaluated using the test data set, and the advantages and disadvantages of the model are analyzed by comparing the predicted results with the true labels. According to the test results, the model is optimized and the parameters are adjusted to improve the classification accuracy.

[0012] (2-3) Fishing vessel number recognition: Use image processing technology to analyze the characteristics of the target object and locate the possible area of ​​the ship number; this usually involves edge detection, image reconstruction and other algorithms to highlight the specific area where the ship number is located, reducing the complexity and error rate of subsequent recognition. When the ship number area is successfully located, it enters the ship number recognition stage; this step uses OCR pattern recognition or machine learning algorithms to conduct in-depth analysis of the ship number area and identify the specific ship number characters;

[0013] (2-4) Result feedback and records: After the vessel number is identified, the identification result will be fed back to the fishing vessel supervision system and corresponding records will be generated in the system; these records include not only the vessel number itself, but also additional information such as the identification time, location, and video screenshots, so that supervisors can conduct subsequent inquiries and comparisons; at the same time, it will also check whether the vessel number already exists in the database for compliance inspection and review.

[0014] The step (2-2) is specifically:

[0015] (22-1) Data collection: Build a comprehensive and representative fishing vessel image dataset to ensure that all fishing vessel types to be identified and their diversity are covered, including but not limited to different sizes, colors, hull structures, cabin layouts, and onboard equipment configurations;

[0016] (22-2) Data preprocessing: The collected fishing boat images are preprocessed, including scaling to a uniform size, cropping to remove useless areas, rotating to correct the image direction, and grayscale processing (or keeping the color to retain the color information), aiming to improve the image quality, reduce the computational complexity, and convert the image into a unified format suitable for model processing;

[0017] (22-3) Feature extraction: Use advanced feature extraction algorithms, such as SIFT (Scale Invariant Feature Transform), HOG (Histogram of Oriented Gradients), or LBP (Local Binary Patterns) to efficiently extract key features of color, shape, texture, and shipboard equipment from the preprocessed images; these features must be able to fully reflect the appearance, structure, and detailed information of the fishing boat, providing a strong basis for subsequent classification;

[0018] (22-4) Feature selection: After feature extraction, a feature selection process is performed to eliminate redundant and irrelevant features and retain the most critical feature subset for classification; these features include the appearance shape, size, color, hull structure, cabin layout, surface texture and pattern of the fishing boat, as well as the information of the ship-borne radar and navigation lights, which together constitute an effective identification for distinguishing different types of fishing boats;

[0019] (22-5) Model construction and training: Select a suitable classification algorithm or classifier, such as a deep learning network (e.g., CNN), a machine learning model, etc., and build a classification model based on the selected feature set. During the training process, the preprocessed image and its corresponding features are used as input, and the mapping relationship between the features and the fishing vessel type is learned through the model. The training goal is to enable the model to accurately predict the fishing vessel type of the new input image;

[0020] (22-6) Model optimization: Analyze the samples in the test set that were misclassified by the model, and adjust the model parameters based on the results of the error analysis.

[0021] The steps (2-3) are specifically:

[0022] (23-1) Image preprocessing: The acquired images are preprocessed to improve the image quality and prepare for analysis. This includes grayscale (if a color image is being processed), denoising (reducing random noise in the image), smoothing (eliminating small details and textures in the image), and binarization (converting the image to a black and white binary image). These steps help reduce the amount of computation and highlight key features in the image;

[0023] (23-2) Feature extraction: involves extracting information from the image that is useful for subsequent analysis. This process includes the extraction of edge, texture, and shape features. Edge features are extracted through edge detection using the Sobel operator. This method uses two 3x3 convolution kernels to calculate the gradient strength and direction of each pixel in the image to identify the boundary lines in the image; texture features are extracted through the grayscale co-occurrence matrix; this matrix can analyze the spatial relationship between pixels in the image and quantify the complexity and directionality of the texture. Shape features represent the outline of the object in the image. First, edge information is obtained through an edge detection algorithm, and then the complete outline of the object is extracted. Then, bounding box extraction techniques, such as minimum rectangle or ellipse fitting, are used to accurately describe the shape characteristics of the object. These feature extraction methods together form the basis of feature analysis in image processing;

[0024] (23-3) Image enhancement and transformation: In image processing, enhancing the contrast, brightness, and color of an image is an important step to improve the visual effect. In order to increase the contrast, histogram equalization technology is often used for grayscale images to optimize the image contrast by stretching the distribution range of pixel values, making the details in the image more distinct.

[0025] For color images, it is necessary to perform histogram equalization on each color channel (such as red, green, and blue) to achieve a similar effect;

[0026] Directly adjusting pixel values ​​via a linear transformation is also a way to improve contrast.

[0027] For brightness adjustment, for grayscale images, multiply the image by a preset coefficient to increase the overall brightness; for color images, first convert the RGB image to the HSI (hue, saturation, brightness) color space, because the HSI space is more suitable for color adjustment; in the HSI space, the brightness of the image can be changed by adjusting the brightness component without affecting the color information;

[0028] For color enhancement, the overall hue of the image is changed by adjusting the hue component, or the hue of each color channel is fine-tuned separately; at the same time, the saturation component is adjusted to enhance or weaken the color vividness of the image and improve the visual appeal. Finally, the adjusted HSI image is converted back to the RGB color space for display on a display device or subsequent storage processing;

[0029] (23-4) Text detection and recognition: Using the convolutional neural network architecture, the system is trained by collecting and annotating image data to learn the characteristics of text areas in the image. The knowledge gained from the training is used for reasoning and analysis to accurately locate all areas containing text in the image and output the coordinates of these areas.

[0030] A text recognition model based on a convolutional recurrent neural network is used. During the recognition process, the model performs a detailed character sequence analysis on each detected text area, and finally presents the recognition result in the form of text, and outputs it together with its corresponding precise location information, thus achieving comprehensive capture and understanding of the text content in the image.

[0031] The OCR algorithm is used to perform in-depth recognition of the detected text area, and the intelligent segmentation technology is used to separate the characters one by one, laying the foundation for subsequent single character recognition. For each segmented character, the OCR algorithm is used to recognize it one by one. After being trained with a large number of character samples, the algorithm can accurately recognize characters of various fonts, sizes, directions and slight deformations;

[0032] (23-5) Result output and display: After the processing is completed, the target feature area will be obtained and the location will be marked with a rectangular box. The processed image result will be output in a format that can be saved or displayed. This may include saving the image as a file (such as JPEG, PNG, etc.) and displaying it on the screen.

[0033] The step (23-2) is specifically as follows:

[0034] (232-1) Preprocessing: First, the input image is denoised to reduce the interference of noise on subsequent feature extraction; commonly used denoising methods include Gaussian filtering, median filtering, etc.

[0035] If the input image is in color, it needs to be converted to grayscale, because algorithms such as edge detection usually operate more efficiently on grayscale images;

[0036] (232-2) Edge feature extraction: Use two 3x3 Sobel convolution kernels (one for detecting horizontal edges and one for detecting vertical edges) to perform convolution operations on the grayscale image; the gradient strength and direction of each pixel point are calculated from these two convolution results (usually using gradient magnitude and gradient direction angle).

[0037] A threshold is set and pixels with gradient strength greater than the threshold are considered edge points, thereby generating an edge image; non-maximum suppression and edge connection are used to refine the edge detection results;

[0038] (232-3) Texture feature extraction: Select appropriate pixel spacing (d) and direction (such as 0°, 45°, 90°, 135°) to calculate GLCM. Each element of GLCM represents the number of times a pair of pixel grayscale values ​​appear simultaneously under the specified direction and spacing. Various texture features can be extracted from GLCM, such as contrast, energy, homogeneity, correlation, etc. These features quantify the complexity and directionality of the texture.

[0039] (232-4) Shape feature extraction: Based on the edge detection results, the edge contours of objects in the image are extracted through contour finding algorithms (such as contour finding after Canny edge detection); a bounding box (such as a minimum enclosing rectangle) is used to roughly describe the position and size of the object; for situations where a more precise shape description is required, ellipse fitting, polygonal approximation and other methods can be used to more accurately describe the contours of the object. Shape invariant features such as Hu moments and Zernike moments can also be calculated. These features are invariant to transformations such as rotation and scaling, which helps with subsequent recognition and classification tasks;

[0040] (232-5) Feature selection and fusion: According to the specific needs of ship number recognition, the edge, texture and shape features that can most effectively characterize the characteristics of the ship number are selected, and different types of features are fused to form a comprehensive feature vector to improve the accuracy and robustness of recognition;

[0041] (232-6) Subsequent processing: The extracted features will be used in subsequent recognition algorithms, such as machine learning-based classifiers (SVM, neural networks, etc.) to identify ship numbers. The recognition results may also require post-processing, such as removing misidentifications and optimizing recognition results.

[0042] The steps of step (232-2) are specifically as follows:

[0043] (2322-1) Cell segmentation: In image processing, a gridding method is used to segment an image into several regions of uniform and equal size, which are called cells. Each cell covers a group of continuous pixels in the image. Its size is an important hyperparameter that needs to be flexibly selected and adjusted according to the needs of specific application scenarios and image characteristics.

[0044] (2322-2) Direction histogram calculation: In the process of constructing the direction histogram, we first need to determine the number of histogram groups and their corresponding angle ranges. A common practice is to evenly divide the gradient direction from 0° to 180° into several intervals, such as 8 intervals, each of which covers 22.5° (180° / 8=22.5°). For each cell in the image, initialize an array that matches the number of direction intervals as the direction histogram of the cell to count the distribution of the gradient direction of the pixel points in the area. Each element of this array corresponds to a direction interval, and the initial value is set to 0;

[0045] Next, traverse each pixel in the cell and determine the direction interval to which it belongs based on its gradient direction; for example, if the gradient direction of a pixel is 30°, then the pixel is classified into the first direction interval based on its range (0°-22.5° is the first interval). Then, the gradient amplitude of the pixel is used as the contribution value and added to the statistical value of the corresponding direction interval. If the gradient amplitude of the pixel is 5, then 5 is added to the statistical value of the first direction interval;

[0046] Repeat the above steps until the gradient contributions of all pixels in the cell are counted; at this point, the cumulative value in each direction interval reflects the overall intensity of the gradient direction in that interval; in order to make the direction histograms of different cells comparable, the direction histogram of each cell is normalized; this is achieved by dividing the gradient contribution value of each direction interval by the sum of the gradient values ​​of all pixels in the entire cell, thereby ensuring that the histograms of all cells are represented based on the same scale;

[0047] (2322-3) Construct a histogram: First, you need to determine the size and step size of the block; a block is a rectangular area composed of several cells, and its size is often set according to specific application requirements, such as 2×2, 3×3 or larger cell combinations. The step size defines the distance the block slides on the image. Setting it to the width or height of 1 cell means that the block will cover the entire image cell by cell as it moves; traverse all cells in the block, and for each cell in the block, obtain the calculated direction histogram (the histogram represents the distribution of the gradient direction of the pixel points in the cell). Then, add up the direction histogram values ​​of these cells to get the sum of the direction histograms of all cells in the block;

[0048] In order to make the directional histograms between different blocks comparable, the directional histograms within the block are normalized. The commonly used method is L2 norm normalization, also known as Euclidean normalization. The specific operation is: divide the directional histogram of each cell in the block (regarded as a vector) by its L2 norm (that is, the modulus of the vector). This step ensures that regardless of the gradient strength of the cells in the block, its directional histogram is scaled to a uniform scale. Repeat the above accumulation and normalization process for all blocks in the image until the normalized directional histogram corresponding to each block is obtained;

[0049] (2322-4) Feature vector construction: An overlapping strategy is used to enhance the richness and robustness of features. First, the image is divided into several overlapping blocks, which are slid on the image according to a preset step size (usually smaller than the block size) to ensure that there is an overlapping area between adjacent blocks.

[0050] Next, these blocks are arranged in a preset order (such as a raster scan order from left to right and from top to bottom); for each block, its internal normalized direction histogram has been calculated and obtained, which reflects the distribution of the gradient direction of the pixel points in the block;

[0051] Connect the normalized histograms in all blocks one by one in the order in which they are arranged in the image to form a long vector whose length is equal to the length of the histogram in a single block multiplied by the total number of blocks; for example, if the normalized histogram in each block contains statistical values ​​in 8 directions (i.e., the histogram has 8 bins) and the image is divided into 4 overlapping blocks, then the final feature vector will be a vector of length 8×4=32. This long vector contains the statistical information of all selected blocks in the image in the gradient direction, providing a comprehensive feature description for subsequent image analysis, recognition or classification tasks;

[0052] The specific method of step (4) is:

[0053] (4-1) Communication and navigation equipment status monitoring: If the vessel number is successfully identified, the operation status of the communication and navigation equipment and other information are obtained through the fishing vessel information acquisition module. If there is any abnormality, feedback will be given directly;

[0054] (4-2) Fishing vessel reporting review and fishing moratorium supervision: Next, determine whether to report and review in the fishing vessel reporting module. It should be noted that according to the fishing moratorium period, special attention should be paid to whether there is a record of passing the reporting and review during the fishing moratorium period. If there is an abnormality, feedback will be given directly;

[0055] (4-3) Problem feedback: When the above problems are found, provide problem feedback and select communication equipment and links for feedback.

[0056] An inspection system based on an intelligent inspection method for compliance of fishing vessels leaving a port, comprising:

[0057] The information collection module is used to collect all-round video data of the dynamic fishing boats about to go out to sea using the PTZ camera and the thermal imaging dome camera;

[0058] The ship and ship number identification and feedback module is used to accurately identify and analyze fishing vessels and ship numbers based on the collected video stream data information, and then feed back the successfully identified ship number information to the fishing vessel supervision system in real time;

[0059] The early warning mechanism module is used to immediately trigger the early warning mechanism if the system fails to successfully identify the fishing vessel number, generate early warning information, and quickly notify relevant personnel so that timely response measures can be taken;

[0060] The compliance inspection and problem notification module is used to automatically associate the fishing vessel navigation equipment corresponding to the ship number if the ship number is successfully identified, retrieve and check its self-inspection record. At the same time, the switch status of AIS and Beidou key navigation equipment is confirmed to ensure that the equipment is in normal working condition.

[0061] The advantages of the present invention are: providing a method and system for intelligent inspection of the compliance of fishing vessels leaving the port. When a fishing vessel leaves the port, firstly, the thermal imaging dome camera and the pan-tilt camera of the bayonet are used to analyze the video stream data information, obtain the ship number, and judge the status of the fishing vessel leaving the port. At the same time, it is judged whether the fishing vessel has turned on the communication and navigation equipment and whether it has reported. Finally, under the analysis of the algorithm, it is judged whether the fishing vessel is compliant with the port departure. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 It is a system structure block diagram of the present invention. DETAILED DESCRIPTION

[0063] An intelligent inspection method for compliance of fishing vessels leaving the port includes the following steps: (1) Information collection stage: using a pan-tilt camera and a thermal imaging dome camera to collect all-round video data of dynamic fishing vessels about to go out to sea; (2) Ship and ship number identification and feedback: based on the collected video stream data information, the image recognition technology is used to accurately identify and analyze the fishing vessel and the ship number, and then the successfully identified ship number information is fed back to the fishing vessel supervision system in real time; (3) Early warning mechanism: if the system fails to successfully identify the fishing vessel number, the early warning mechanism is immediately triggered, the early warning information is generated, and the relevant staff is quickly notified so that timely response measures can be taken; (4) Compliance inspection and problem notification: if the ship number is successfully identified, the fishing vessel communication and navigation equipment corresponding to the ship number is automatically associated, and its self-inspection record is retrieved and checked. At the same time, the switch status of AIS (Automatic Identification System) and Beidou key navigation equipment is confirmed to ensure that the equipment is in normal working condition;

[0064] Through the fishing port entry and exit reporting module of the fishing vessel supervision system, check the status of fishing vessel entry and exit reporting, including whether it is within the fishing moratorium period, whether the relevant reporting information has been completed, etc. Pay special attention to special circumstances during the fishing moratorium period, such as the legal sea records of aquaculture vessels or rescue vessels, to ensure that all activities comply with regulations;

[0065] When any problems are found, such as equipment failure, unreported or illegal sailing, etc., notifications will be sent to staff immediately for prompt processing.

[0066] The step (2) is specifically as follows:

[0067] (2-1) Video data acquisition: The PTZ camera works in conjunction with the thermal imaging dome camera and AI system. The AI ​​system controls the camera to collect high-definition video data of fishing boats about to go out to sea. The PTZ camera can capture clear images of fishing boats from different angles with its flexible rotation capability. The thermal imaging dome camera uses thermal imaging technology to obtain heat source images of fishing boats at night or in bad weather conditions, ensuring the comprehensiveness and accuracy of the data.

[0068] (2-2) Fishing vessel identification: By extensively collecting representative fishing vessel image datasets, we ensure that all types of fishing vessels that need to be identified are covered, providing a solid foundation for model training. Subsequently, the collected images are preprocessed, including scaling, cropping, rotating, and grayscale conversion, to unify the format and optimize the image quality.

[0069] Then, SIFT, HOG or LBP feature extraction algorithms are used to accurately extract color, shape, texture and key features of shipboard equipment from the preprocessed images, and redundant information is removed through the feature selection process to retain the features that are critical to classification;

[0070] Subsequently, based on the selected features, a classification model is constructed and trained so that the model can learn the relationship between the features and the types of fishing vessels. After the model training is completed, the classification performance is evaluated using a test data set, and the strengths and weaknesses of the model are analyzed by comparing the predicted results with the true labels. Based on the test results, the model is optimized and the parameters are adjusted to improve the classification accuracy. (2-3) Fishing vessel number recognition: Image processing technology is used to analyze the characteristics of the target object and locate the area with the possible ship number. This usually involves algorithms such as edge detection and image reconstruction to highlight the specific area where the ship number is located, thereby reducing the complexity and error rate of subsequent recognition. When the vessel number area is successfully located, the vessel number recognition stage begins. This step uses OCR pattern recognition or machine learning algorithms to conduct in-depth analysis of the vessel number area and identify the specific vessel number characters. (2-4) Result feedback and record: After the vessel number is recognized, the recognition result is fed back to the fishing vessel supervision system and corresponding records are generated in the system. These records include not only the vessel number itself, but also the recognition time, location, and video screenshots for additional information to facilitate subsequent inquiries and comparisons by supervisors. At the same time, it will also check whether the vessel number already exists in the database for compliance checks and reviews.

[0071] The step (2-2) is specifically:

[0072] (22-1) Data collection: Build a comprehensive and representative fishing vessel image dataset to ensure that all fishing vessel types to be identified and their diversity are covered, including but not limited to different sizes, colors, hull structures, cabin layouts, and onboard equipment configurations;

[0073] (22-2) Data preprocessing: The collected fishing boat images are preprocessed, including scaling to a uniform size, cropping to remove useless areas, rotating to correct the image direction, and grayscale processing (or keeping the color to retain the color information), aiming to improve the image quality, reduce the computational complexity, and convert the image into a unified format suitable for model processing;

[0074] (22-3) Feature extraction: Use advanced feature extraction algorithms, such as SIFT (Scale Invariant Feature Transform), HOG (Histogram of Oriented Gradients), or LBP (Local Binary Patterns) to efficiently extract key features of color, shape, texture, and shipboard equipment from the preprocessed images; these features must be able to fully reflect the appearance, structure, and detailed information of the fishing boat, providing a strong basis for subsequent classification;

[0075] (22-4) Feature selection: After feature extraction, a feature selection process is performed to eliminate redundant and irrelevant features and retain the most critical feature subset for classification; these features include the appearance shape, size, color, hull structure, cabin layout, surface texture and pattern of the fishing boat, as well as the information of the ship-borne radar and navigation lights, which together constitute an effective identification for distinguishing different types of fishing boats;

[0076] (22-5) Model construction and training: Select a suitable classification algorithm or classifier, such as a deep learning network (e.g., CNN), a machine learning model, etc., and build a classification model based on the selected feature set. During the training process, the preprocessed image and its corresponding features are used as input, and the mapping relationship between the features and the fishing vessel type is learned through the model. The training goal is to enable the model to accurately predict the fishing vessel type of the new input image;

[0077] (22-6) Model optimization: Analyze the samples in the test set that were misclassified by the model, and adjust the model parameters based on the results of the error analysis.

[0078] The steps (2-3) are specifically:

[0079] (23-1) Image preprocessing: The acquired images are preprocessed to improve the image quality and prepare for analysis. This includes grayscale (if a color image is being processed), denoising (reducing random noise in the image), smoothing (eliminating small details and textures in the image), and binarization (converting the image to a black and white binary image). These steps help reduce the amount of computation and highlight key features in the image;

[0080] (23-2) Feature extraction: involves extracting information from the image that is useful for subsequent analysis. This process includes the extraction of edge, texture, and shape features. Edge features are extracted through edge detection using the Sobel operator. This method uses two 3x3 convolution kernels to calculate the gradient strength and direction of each pixel in the image to identify the boundary lines in the image; texture features are extracted through the grayscale co-occurrence matrix; this matrix can analyze the spatial relationship between pixels in the image and quantify the complexity and directionality of the texture. Shape features represent the outline of the object in the image. First, edge information is obtained through an edge detection algorithm, and then the complete outline of the object is extracted. Then, bounding box extraction techniques, such as minimum rectangle or ellipse fitting, are used to accurately describe the shape characteristics of the object. These feature extraction methods together form the basis of feature analysis in image processing;

[0081] (23-3) Image enhancement and transformation: In image processing, enhancing the contrast, brightness, and color of an image is an important step to improve the visual effect. In order to increase the contrast, histogram equalization technology is often used for grayscale images to optimize the image contrast by stretching the distribution range of pixel values, making the details in the image more distinct.

[0082] For color images, it is necessary to perform histogram equalization on each color channel (such as red, green, and blue) to achieve a similar effect;

[0083] Directly adjusting pixel values ​​via a linear transformation is also a way to improve contrast.

[0084] For brightness adjustment, for grayscale images, multiply the image by a preset coefficient to increase the overall brightness; for color images, first convert the RGB image to the HSI (hue, saturation, brightness) color space, because the HSI space is more suitable for color adjustment; in the HSI space, the brightness of the image can be changed by adjusting the brightness component without affecting the color information; for color enhancement, the overall hue of the image is changed by adjusting the hue component, or the hue of each color channel is fine-tuned separately; at the same time, by adjusting the saturation component, the color vividness of the image is enhanced or weakened to enhance visual appeal. Finally, the adjusted HSI image is converted back to the RGB color space for display on a display device or subsequent storage processing;

[0085] (23-4) Text detection and recognition: Using the convolutional neural network architecture, the system is trained by collecting and annotating image data to learn the characteristics of text areas in the image. The knowledge gained from the training is used for reasoning and analysis to accurately locate all areas containing text in the image and output the coordinates of these areas.

[0086] A text recognition model based on a convolutional recurrent neural network is used. During the recognition process, the model performs a detailed character sequence analysis on each detected text area, and finally presents the recognition result in the form of text, and outputs it together with its corresponding precise location information, thus achieving comprehensive capture and understanding of the text content in the image.

[0087] The OCR algorithm is used to perform in-depth recognition of the detected text area, and the intelligent segmentation technology is used to separate the characters one by one, laying the foundation for subsequent single character recognition. For each segmented character, the OCR algorithm is used to recognize it one by one. After being trained with a large number of character samples, the algorithm can accurately recognize characters of various fonts, sizes, directions and slight deformations;

[0088] (23-5) Result output and display: After the processing is completed, the target feature area will be obtained and the location will be marked with a rectangular box. The processed image result will be output in a format that can be saved or displayed. This may include saving the image as a file (such as JPEG, PNG, etc.) and displaying it on the screen.

[0089] The step (23-2) is specifically as follows:

[0090] (232-1) Preprocessing: First, the input image is denoised to reduce the interference of noise on subsequent feature extraction; commonly used denoising methods include Gaussian filtering, median filtering, etc. If the input image is in color, it needs to be converted to a grayscale image, because algorithms such as edge detection usually operate more efficiently on grayscale images;

[0091] (232-2) Edge feature extraction: Use two 3x3 Sobel convolution kernels (one for detecting horizontal edges and one for detecting vertical edges) to perform convolution operations on the grayscale image; the gradient strength and direction of each pixel point are calculated from these two convolution results (usually using gradient magnitude and gradient direction angle).

[0092] A threshold is set and pixels with gradient strength greater than the threshold are considered edge points, thereby generating an edge image; non-maximum suppression and edge connection are used to refine the edge detection results;

[0093] (232-3) Texture feature extraction: Select appropriate pixel spacing (d) and direction (such as 0°, 45°, 90°, 135°) to calculate GLCM. Each element of GLCM represents the number of times a pair of pixel grayscale values ​​appear simultaneously under the specified direction and spacing. Various texture features can be extracted from GLCM, such as contrast, energy, homogeneity, correlation, etc. These features quantify the complexity and directionality of the texture.

[0094] (232-4) Shape feature extraction: Based on the edge detection results, the edge contours of objects in the image are extracted through contour finding algorithms (such as contour finding after Canny edge detection); a bounding box (such as a minimum enclosing rectangle) is used to roughly describe the position and size of the object; for situations where a more precise shape description is required, ellipse fitting, polygonal approximation and other methods can be used to more accurately describe the contours of the object. Shape invariant features such as Hu moments and Zernike moments can also be calculated. These features are invariant to transformations such as rotation and scaling, which helps with subsequent recognition and classification tasks;

[0095] (232-5) Feature selection and fusion: According to the specific needs of ship number recognition, the edge, texture and shape features that can most effectively characterize the characteristics of the ship number are selected, and different types of features are fused to form a comprehensive feature vector to improve the accuracy and robustness of recognition;

[0096] (232-6) Subsequent processing: The extracted features will be used in subsequent recognition algorithms, such as machine learning-based classifiers (SVM, neural networks, etc.) to identify ship numbers. The recognition results may also require post-processing, such as removing misidentifications and optimizing recognition results.

[0097] The steps of step (232-2) are specifically as follows:

[0098] (2322-1) Cell segmentation: In image processing, a gridding method is used to segment an image into several regions of uniform and equal size, which are called cells. Each cell covers a group of continuous pixels in the image. Its size is an important hyperparameter that needs to be flexibly selected and adjusted according to the needs of specific application scenarios and image characteristics.

[0099] (2322-2) Direction histogram calculation: In the process of constructing the direction histogram, we first need to determine the number of histogram groups and their corresponding angle ranges. A common practice is to evenly divide the gradient direction from 0° to 180° into several intervals, such as 8 intervals, each of which covers 22.5° (180° / 8=22.5°). For each cell in the image, initialize an array that matches the number of direction intervals as the direction histogram of the cell to count the distribution of the gradient direction of the pixel points in the area. Each element of this array corresponds to a direction interval, and the initial value is set to 0;

[0100] Next, traverse each pixel in the cell and determine the direction interval to which it belongs based on its gradient direction; for example, if the gradient direction of a pixel is 30°, then the pixel is classified into the first direction interval based on its range (0°-22.5° is the first interval). Then, the gradient amplitude of the pixel is used as the contribution value and added to the statistical value of the corresponding direction interval. If the gradient amplitude of the pixel is 5, then 5 is added to the statistical value of the first direction interval;

[0101] Repeat the above steps until the gradient contributions of all pixels in the cell are counted; at this point, the cumulative value in each direction interval reflects the overall intensity of the gradient direction in that interval; in order to make the direction histograms of different cells comparable, the direction histogram of each cell is normalized; this is achieved by dividing the gradient contribution value of each direction interval by the sum of the gradient values ​​of all pixels in the entire cell, thereby ensuring that the histograms of all cells are represented based on the same scale;

[0102] (2322-3) Construct a histogram: First, you need to determine the size and step size of the block; a block is a rectangular area composed of several cells, and its size is often set according to specific application requirements, such as 2×2, 3×3 or larger cell combinations. The step size defines the distance the block slides on the image. Setting it to the width or height of 1 cell means that the block will cover the entire image cell by cell as it moves; traverse all cells in the block, and for each cell in the block, obtain the calculated direction histogram (the histogram represents the distribution of the gradient direction of the pixel points in the cell). Then, add up the direction histogram values ​​of these cells to get the sum of the direction histograms of all cells in the block;

[0103] In order to make the directional histograms between different blocks comparable, the directional histograms within the block are normalized. The commonly used method is L2 norm normalization, also known as Euclidean normalization. The specific operation is: divide the directional histogram of each cell in the block (regarded as a vector) by its L2 norm (that is, the modulus of the vector). This step ensures that regardless of the gradient strength of the cells in the block, its directional histogram is scaled to a uniform scale. Repeat the above accumulation and normalization process for all blocks in the image until the normalized directional histogram corresponding to each block is obtained;

[0104] (2322-4) Feature vector construction: An overlapping strategy is used to enhance the richness and robustness of features. First, the image is divided into several overlapping blocks, which slide on the image according to a preset step size (usually smaller than the block size) to ensure that there are overlapping areas between adjacent blocks. Next, the blocks are arranged in a preset order (such as a raster scanning order from left to right and from top to bottom). For each block, its internal normalized directional histogram has been calculated and obtained, which reflects the distribution of the gradient direction of the pixel points in the block. The normalized histograms in all blocks are connected one by one according to their arrangement order in the image to form a long vector, the length of which is equal to the length of the histogram in a single block multiplied by the total number of blocks. For example, if the normalized histogram in each block contains statistical values ​​in 8 directions (that is, the histogram has 8 bins) and the image is divided into 4 overlapping blocks, then the final feature vector will be a vector with a length of 8×4=32. This long vector contains the statistical information of all selected blocks in the image in the gradient direction, providing a comprehensive feature description for subsequent image analysis, recognition or classification tasks;

[0105] The specific method of step (4) is:

[0106] (4-1) Communication and navigation equipment status monitoring: If the vessel number is successfully identified, the operation status of the communication and navigation equipment and other information are obtained through the fishing vessel information acquisition module. If there is any abnormality, feedback will be given directly;

[0107] (4-2) Fishing vessel reporting review and fishing moratorium supervision: Next, determine whether to report and review in the fishing vessel reporting module. It should be noted that according to the fishing moratorium period, special attention should be paid to whether there is a record of passing the reporting and review during the fishing moratorium period. If there is an abnormality, feedback will be given directly;

[0108] (4-3) Problem feedback: When the above problems are found, provide problem feedback and select communication equipment and links for feedback.

[0109] The present invention also relates to an inspection system based on an intelligent inspection method for compliance of fishing vessels leaving a port, comprising:

[0110] The information collection module 1 is used to collect all-round video data of the dynamic fishing boat about to go out to sea by using a pan-tilt camera and a thermal imaging dome camera;

[0111] The vessel and vessel number identification and feedback module 2 is used to accurately identify and analyze fishing vessels and vessel numbers based on the collected video stream data information, and then feed back the successfully identified vessel number information to the fishing vessel supervision system in real time;

[0112] Early warning mechanism module 3 is used to immediately trigger the early warning mechanism if the system fails to successfully identify the fishing vessel number, generate early warning information, and quickly notify relevant staff so that timely response measures can be taken;

[0113] The compliance inspection and problem notification module 4 is used to automatically associate the fishing vessel navigation equipment corresponding to the ship number if the ship number is successfully identified, retrieve and check its self-inspection record. At the same time, the switch status of AIS and Beidou key navigation equipment is confirmed to ensure that the equipment is in normal working condition.

[0114] The present invention realizes its function through the following key steps:

[0115] Information collection stage: PTZ cameras and thermal imaging dome cameras are used to collect all-round video data of the dynamic fishing boats about to go out to sea. The PTZ camera provides flexible viewing angle adjustment to capture clear images of the fishing boats; the thermal imaging dome camera ensures the comprehensiveness of the data through heat source images at night or in bad weather conditions.

[0116] Vessel and vessel number identification and feedback: Through the collected video stream data, image recognition technology is used to accurately identify fishing vessels and their vessel numbers. The identified vessel number information is fed back to the fishing vessel supervision system in real time, providing basic data for subsequent compliance inspections.

[0117] Early warning mechanism: If the system fails to successfully identify the fishing vessel number, the early warning mechanism will be triggered, an early warning message will be generated, and relevant staff will be promptly notified so that timely response measures can be taken.

[0118] Compliance inspection and problem notification: After successfully identifying the vessel number, the system automatically links the fishing vessel's navigation equipment to check its self-inspection records and the working status of key navigation equipment (such as AIS and Beidou system). At the same time, the system verifies the status of the fishing vessel's entry and exit report review to ensure that all activities comply with regulations. If any problems are found, such as equipment failure, failure to report or illegal sea voyage, the staff will be notified immediately to handle them.

[0119] The advantages of the present invention are:

[0120] Comprehensive data collection: Combining PTZ cameras and thermal imaging dome cameras ensures all-round monitoring of fishing vessels, whether in bad weather or at night, improving the comprehensiveness and accuracy of data collection.

[0121] Accurate vessel number identification: Through advanced image recognition technology, fishing vessels and their vessel numbers can be accurately identified, providing reliable data support for supervision.

[0122] Real-time feedback and early warning: The identification results are fed back to the supervision system in real time, and an early warning is triggered when the identification fails, which improves the timeliness and initiative of supervision.

[0123] Automated compliance checks: Automatically check the navigation equipment and reporting status of fishing vessels, reducing the workload of manual inspections and improving the efficiency and accuracy of inspections.

[0124] Strengthened supervision: Special attention is paid to the special circumstances during the fishing moratorium to ensure that fishing activities strictly comply with laws and regulations, thereby enhancing the protection and management of marine resources.

[0125] Timely problem handling: Once a violation or abnormal situation is discovered, relevant personnel can be notified quickly for processing, effectively avoiding potential safety risks and waste of resources.

Claims

1. A method for intelligent inspection of compliance of fishing vessels leaving a port, comprising the following steps: (1) Information collection stage: Using a pan-tilt camera and a thermal imaging dome camera, we collect all-round video data of the dynamic fishing boat about to go out to sea; (2) Vessel and vessel number identification and feedback: Based on the collected video stream data information, fishing vessels and vessel numbers are accurately identified and analyzed, and then the successfully identified vessel number information is fed back to the fishing vessel supervision system in real time; (3) Early warning mechanism: If the system fails to successfully identify the fishing vessel number, the early warning mechanism will be triggered immediately, generating early warning information and promptly notifying relevant personnel so that timely response measures can be taken; (4) Compliance inspection and problem notification: If the vessel number is successfully identified, the navigation equipment of the fishing vessel corresponding to the vessel number will be automatically linked, and its self-inspection record will be retrieved and checked. At the same time, the switch status of AIS and Beidou key navigation equipment will be confirmed to ensure that the equipment is in normal working condition; Through the fishing port entry and exit reporting module of the fishing vessel supervision system, check the status of fishing vessel entry and exit reporting, including whether it is within the fishing moratorium period, whether the relevant reporting information has been completed, etc. Pay special attention to special circumstances during the fishing moratorium period, such as the legal sea records of aquaculture vessels or rescue vessels, to ensure that all activities comply with regulations; When any issues are discovered, notifications are sent immediately to staff so they can be handled quickly.

2. According to claim 1, a method for intelligent inspection of compliance of fishing vessels leaving port, characterized in that: The step (2) is specifically as follows: (2-1) Video data acquisition: The pan-tilt camera works in conjunction with the thermal imaging dome camera and the AI ​​system, and the AI ​​system controls the camera to collect high-definition video data of fishing boats about to go out to sea; (2-2) Fishing vessel identification: The fishing vessel image dataset ensures that all types of fishing vessels that need to be identified are covered, providing a solid foundation for model training; Subsequently, the acquired images were preprocessed, including scaling, cropping, rotation, and grayscale, to unify the format and optimize the image quality; Then, SIFT, HOG or LBP feature extraction algorithms are used to accurately extract color, shape, texture and key features of shipboard equipment from the preprocessed images, and redundant information is removed through the feature selection process to retain the features that are critical to classification; Subsequently, based on the selected features, a classification model is constructed and trained so that the model can learn the relationship between the features and the fishing vessel types; After the model training is completed, the classification performance is evaluated using the test data set. By comparing the prediction results with the true labels, the advantages and disadvantages of the model are analyzed. According to the test results, the model is optimized and the parameters are adjusted to improve the classification accuracy; (2-3) Fishing vessel number recognition: Use image processing technology to analyze the characteristics of the target object and locate the area of ​​the possible ship number. Once the ship number area is successfully located, the ship number recognition stage begins. This step uses OCR pattern recognition or machine learning algorithms to conduct in-depth analysis of the ship number area and identify the specific ship number characters. (2-4) Result feedback and records: After the vessel number is identified, the identification result will be fed back to the fishing vessel supervision system and corresponding records will be generated in the system; these records include not only the vessel number itself, but also additional information such as the identification time, location, and video screenshots, so that supervisors can conduct subsequent inquiries and comparisons; at the same time, it will also check whether the vessel number already exists in the database for compliance inspection and review.

3. According to claim 2, a method for intelligent inspection of compliance of fishing vessels leaving port is characterized in that: The step (2-2) is specifically: (22-1) Data collection: Build a fishing vessel image dataset to ensure that all types of fishing vessels to be identified and their diversity are covered, including but not limited to different sizes, colors, hull structures, cabin layouts, and onboard equipment configurations; (22-2) Data preprocessing: The collected fishing boat images are preprocessed, including scaling to a uniform size, cropping to remove useless areas, rotating to correct the image direction, and grayscale processing, and the images are converted into a uniform format suitable for model processing; (22-3) Feature extraction: Using feature extraction algorithm, color, shape, texture and key features of shipborne equipment are efficiently extracted from the preprocessed image; (22-4) Feature selection: After feature extraction, a feature selection process is performed to eliminate redundant and irrelevant features and retain the feature subset that is most critical for classification; These features include the appearance, size, color, hull structure, cabin layout, surface texture and pattern of the fishing vessel, as well as the information of the onboard radar and navigation lights, which together constitute an effective identification to distinguish different types of fishing vessels; (22-5) Model construction and training: Select a suitable classification algorithm or classifier, and build a classification model based on the selected feature set. During the training process, the preprocessed image and its corresponding features are used as input, and the mapping relationship between features and fishing vessel types is learned through the model; (22-6) Model optimization: Analyze the samples in the test set that are misclassified by the model, and adjust the model parameters based on the results of the error analysis.

4. According to claim 2, a method for intelligent inspection of compliance of fishing vessels leaving port is characterized in that: The steps (2-3) are specifically: (23-1) Image preprocessing: preprocess the acquired images to improve image quality and prepare for analysis; (23-2) Feature extraction: Edge features are extracted by performing edge detection using the Sobel operator. This method uses two 3x3 convolution kernels to calculate the gradient strength and direction of each pixel in the image, thereby identifying the boundary lines in the image. Texture features are extracted through the gray-level co-occurrence matrix. First, edge information is obtained through the edge detection algorithm, and then the complete outline of the object is extracted. Then, the bounding box extraction technology is used to describe the shape characteristics of the object. (23-3) Image enhancement and transformation: In image processing, enhancing the contrast, brightness, and color of an image is an important step to improve the visual effect. In order to increase the contrast, histogram equalization technology is often used for grayscale images to optimize the image contrast by stretching the distribution range of pixel values; For color images, it is necessary to perform histogram equalization on each color channel separately to achieve a similar effect; For brightness adjustment, for grayscale images, multiply the image by a preset coefficient to increase the overall brightness; for color images, first convert the RGB image to the HSI color space, because the HSI space is more suitable for color adjustment; in the HSI space, the brightness of the image can be changed by adjusting the brightness component without affecting the color information; For color enhancement, the overall hue of the image is changed by adjusting the hue component, or the hue of each color channel is fine-tuned separately; at the same time, the color vividness of the image is enhanced or weakened by adjusting the saturation component; finally, the adjusted HSI image is converted back to the RGB color space for display on a display device or subsequent storage processing; (23-4) Text detection and recognition: Using the convolutional neural network architecture, the system is trained by collecting and annotating image data to learn the features of text areas in the image. The knowledge gained from the training is used to locate all areas in the image that contain text and output the coordinates of these areas. A text recognition model based on a convolutional recurrent neural network is used. During the recognition process, the model performs a detailed character sequence analysis on each detected text area, and finally presents the recognition result in the form of text, and outputs it together with its corresponding precise location information, thus achieving comprehensive capture and understanding of the text content in the image. Use the OCR algorithm to deeply identify the detected text area, and use intelligent segmentation technology to separate the characters one by one. For each segmented character, use the OCR algorithm to identify each character one by one, and identify characters of various fonts, sizes, directions and slight deformations; (23-5) Result output and display: After the final processing is completed, the target feature area will be obtained and the location will be marked with a rectangular box. The processed image results will be output in a format that can be saved or displayed.

5. According to claim 4, a method for intelligent inspection of compliance of fishing vessels leaving port is characterized in that: The step (23-2) is specifically as follows: (232-1) Preprocessing: First, the input image is denoised to reduce the interference of noise on subsequent feature extraction; (232-2) Edge feature extraction: Use two 3x3 Sobel convolution kernels to perform convolution operations on the grayscale image; the gradient strength and direction of each pixel point are calculated through these two convolution results; Set a threshold and regard the pixels with gradient intensity greater than the threshold as edge points, thereby generating an edge image; use non-maximum suppression and edge connection to refine the edge detection results; (232-3) Texture feature extraction: select appropriate pixel spacing and direction to calculate GLCM. Each element of GLCM represents the number of times a pair of pixel grayscale values ​​appear simultaneously under the specified direction and spacing; (232-4) Shape feature extraction: Based on the edge detection results, the edge contours of the objects in the image are extracted through the contour finding algorithm; the bounding box is used to roughly describe the position and size of the object; (232-5) Feature selection and fusion: According to the specific needs of ship number recognition, select the edge, texture and shape features that can most effectively represent the characteristics of the ship number, and fuse different types of features to form a comprehensive feature vector; (232-6) Subsequent processing: The extracted features will be used in subsequent recognition algorithms.

6. According to claim 4, a method for intelligent inspection of compliance of fishing vessels leaving port is characterized in that: The steps of step (232-2) are specifically as follows: (2322-1) Cell segmentation: In image processing, a gridding method is used to divide an image into several uniformly sized and equal regions, which are called cells; each cell covers a group of continuous pixels in the image; (2322-2) Direction histogram calculation: In the process of constructing the direction histogram, we first need to determine the number of histogram groups and their corresponding angle ranges. For each cell in the image, initialize an array that matches the number of direction intervals as the direction histogram of the cell to count the distribution of the gradient direction of the pixel points in the area. Each element of this array corresponds to a direction interval, and the initial value is set to 0. Next, traverse each pixel in the cell and determine the direction interval to which it belongs based on its gradient direction; Repeat the above steps until the gradient contributions of all pixels in the cell are counted; at this point, the cumulative value in each direction interval reflects the overall intensity of the gradient direction in that interval; in order to make the direction histograms of different cells comparable, the direction histogram of each cell is normalized; this is achieved by dividing the gradient contribution value of each direction interval by the sum of the gradient values ​​of all pixels in the entire cell, thereby ensuring that the histograms of all cells are represented based on the same scale; (2322-3) Constructing the histogram: First, we need to determine the block size and step size; a block is a rectangular area consisting of a number of cells. Its size is often set according to specific application requirements. The step size defines the distance the block slides on the image. Setting it to the width or height of 1 cell means that the block will cover the entire image cell by cell when it moves; Traverse all cells in the block, obtain the calculated direction histogram for each cell in the block, accumulate the direction histogram values ​​of these cells, and get the sum of the direction histograms of all cells in the block; In order to make the direction histograms between different blocks comparable, the direction histograms within the block are normalized. The specific operation is as follows: the direction histogram of each cell in the block is divided by its L2 norm; the above accumulation and normalization process is repeated for all blocks in the image until the normalized direction histogram corresponding to each block is obtained; (2322-4) Feature vector construction: An overlapping strategy is used to enhance the richness and robustness of features. First, the image is divided into several overlapping blocks, which slide on the image according to a preset step size to ensure that there is an overlapping area between adjacent blocks. Next, these blocks are arranged in a preset order; for each block, the normalized direction histogram inside it has been calculated and obtained, which reflects the distribution of the gradient direction of the pixel points in the block; The normalized histograms within all blocks are connected one by one in the order of their arrangement in the image to form a long vector whose length is equal to the length of the histogram within a single block multiplied by the total number of blocks; this long vector contains the statistical information of all selected blocks in the image in the gradient direction, providing a comprehensive feature description for subsequent image analysis, recognition or classification tasks.

7. The intelligent inspection method for compliance of fishing vessels leaving port according to claim 4 is characterized in that: The specific method of step (4) is as follows: (4-1) Communication and navigation equipment status monitoring: If the vessel number is successfully identified, the operation status of the communication and navigation equipment and other information are obtained through the fishing vessel information acquisition module. If there is any abnormality, feedback will be given directly; (4-2) Fishing vessel reporting review and fishing moratorium supervision: Next, determine whether to report and review in the fishing vessel reporting module. It should be noted that according to the fishing moratorium period, special attention should be paid to whether there is a record of passing the reporting and review during the fishing moratorium period. If there is an abnormality, feedback will be given directly; (4-3) Problem feedback: When the above problems are found, provide problem feedback and select communication equipment and links for feedback.

8. An inspection system based on the intelligent inspection method for compliance of fishing vessels leaving a port according to any one of claims 1 to 7, comprising: The information collection module is used to collect all-round video data of the dynamic fishing boats about to go out to sea using the PTZ camera and the thermal imaging dome camera; The ship and ship number identification and feedback module is used to accurately identify and analyze fishing vessels and ship numbers based on the collected video stream data information, and then feed back the successfully identified ship number information to the fishing vessel supervision system in real time; The early warning mechanism module is used to immediately trigger the early warning mechanism if the system fails to successfully identify the fishing vessel number, generate early warning information, and quickly notify relevant personnel so that timely response measures can be taken; The compliance inspection and problem notification module is used to automatically associate the fishing vessel navigation equipment corresponding to the ship number if the ship number is successfully identified, retrieve and check its self-inspection record. At the same time, the switch status of AIS and Beidou key navigation equipment is confirmed to ensure that the equipment is in normal working condition.

Images