Method and system for identifying abnormal behavior of a ship

Abstract

The application discloses a kind of identification method and system for ship abnormal behavior, identification method includes: address acquisition step: the URL video stream address of at least one video stream is obtained;Initial data acquisition step: video source data and video parameter information are obtained based on URL video stream address, video source data and the video parameter information are encapsulated to form initial data packet;Identification result acquisition step: based on URL video stream address and initial data packet, through the serial identification of pre-deployed ship body identification model, ship number identification model and abnormal behavior identification model, identification result is obtained.

Description

Technical Field

[0001] This invention relates to the field of ship identification, and more particularly to a method and system for identifying abnormal ship behavior. Background Technology

[0002] Many coastal cities have successively built a series of marine infrastructure and information systems, possessing a certain level of information technology foundation. However, traditional working methods are still used in some marine management and industry supervision services, and the integration of data resources is not high, indicating that the level of informatization needs further improvement. Furthermore, most intelligent fishing port platforms currently rely on BeiDou satellite positioning system data, AIS data, radar data, etc., as data sources for fishing vessel safety assurance and supervision, while the vast majority of video recording devices installed at many fishing port wharves are not being fully utilized.

[0003] Existing identification schemes, such as using servers and shipboard terminals on ships, include wireless communication modules, processors, RFID modules, and positioning modules (including AIS modules and Beidou / GPS dual-mode positioning modules). However, this scheme relies entirely on hardware devices and wireless communication to determine in real time whether a ship has entered a fishing port, and has no direct relationship with artificial intelligence algorithms, resulting in low identification efficiency.

[0004] For example, although existing identification schemes use artificial intelligence algorithms to detect the hull, ship number, and abnormal behavior, the various identification models in these schemes are isolated, that is, they adopt a parallel identification scheme. Therefore, their accuracy and inference speed are relatively low.

[0005] Therefore, through dedicated research, the inventors developed a method and system for identifying abnormal ship behavior that overcomes the aforementioned shortcomings. Summary of the Invention

[0006] To address the above problems, this invention provides a method for identifying abnormal ship behavior, comprising:

[0007] Address acquisition steps: Obtain the URL of at least one video stream;

[0008] Initial data acquisition steps: Obtain video source data and video parameter information based on the URL video stream address, and encapsulate the video source data and the video parameter information to form an initial data packet;

[0009] Steps for obtaining identification results: Based on the URL video stream address and the initial data packet, the identification results are obtained by serial identification through a pre-deployed ship hull identification model, ship number identification model and abnormal behavior identification model.

[0010] The above identification method, wherein the address acquisition step includes:

[0011] Obtain the RTSP format URL video stream address of multiple camera devices through an external platform.

[0012] The above-described identification method, wherein the initial data acquisition step includes:

[0013] The number of data acquisition processes is determined based on the number of video streams.

[0014] The data acquisition process obtains the video source data corresponding to the URL video stream address based on the URL video stream address;

[0015] The URL video stream address is parsed to obtain multiple video parameter information;

[0016] The video parameter information is assembled into the initial data packet.

[0017] The above-described identification method, wherein the step of obtaining the identification result includes:

[0018] Ship hull identification steps: The video source data obtained based on the URL video stream address and the initial data packet are used to obtain a first identification result through the ship hull identification model;

[0019] Ship number identification steps: Based on the first identification result, a second identification result is obtained through the ship number identification model.

[0020] The above-described identification method, wherein the step of obtaining the identification result further includes:

[0021] Abnormal behavior identification steps: Based on the first identification result or the second identification result, the final identification result is obtained through the abnormal behavior identification model.

[0022] The above-described identification method, wherein the hull identification step includes:

[0023] If the hull recognition model identifies a valid hull target, the hull type and coordinates of the valid hull target are obtained.

[0024] The effective hull target is marked on the original image of the image data frame of the initial data packet using a first bounding rectangle based on the coordinates of the effective hull target.

[0025] The original image of the labeled image data frame is sent to a file server, and the URL address returned by the file server is received.

[0026] The first cropped image is obtained by cropping the original image of the marked image data frame based on the first boundary rectangle, and the first cropped image is sent to the file server and the URL address returned by the file server is received.

[0027] The hull type, coordinates, and URL address are written into the initial data packet to form the first identification result.

[0028] The above-described identification method, wherein the ship number identification step includes:

[0029] The captured image is identified using the ship number recognition model.

[0030] If the vessel number recognition model identifies a valid vessel number target, it obtains the vessel license plate number area and vessel number of the valid vessel number target;

[0031] Mark the vessel license plate number area and vessel number of the valid vessel number target in the first captured image to obtain the second captured image;

[0032] The marked second cropped image is sent to the file server, and the URL address returned by the file server is received;

[0033] The ship number and URL address are written into the first identification result to form the second identification result.

[0034] The above-described identification method, wherein the abnormal behavior identification step includes:

[0035] The abnormal behavior recognition model is used to identify the first or second cropped image.

[0036] If the abnormal behavior recognition model identifies a valid abnormal behavior target, the abnormal behavior type of the valid abnormal behavior target is obtained;

[0037] The effective abnormal behavior target is marked on the first or second cropped image using a second bounding rectangle to obtain a third cropped image;

[0038] The marked third cropped image is sent to the file server, and the URL address returned by the file server is received;

[0039] The abnormal behavior type and URL address of the valid abnormal behavior target are written into the second identification result to obtain the final identification result.

[0040] The above-described identification method, wherein the abnormal behavior identification step further includes: training the abnormal behavior identification model before identification, including:

[0041] The datasets with different abnormal behavior types obtained in advance are divided into initial training set, initial validation set and initial test set according to the abnormal behavior type;

[0042] The initial training set, initial validation set, and initial test set are fused together to obtain the final training set, final validation set, and final test set.

[0043] The abnormal behavior recognition model is trained using the final training set, the final validation set, and the final test set.

[0044] This invention provides a system for identifying abnormal ship behavior, wherein the identification method described in any one of the above-mentioned methods is applied, and the identification system includes:

[0045] The address acquisition unit retrieves the URL video stream address of at least one video stream;

[0046] The initial data acquisition unit acquires video source data and video parameter information based on the URL video stream address, and encapsulates the video source data and the video parameter information to form an initial data packet.

[0047] The identification result acquisition unit obtains the identification result by performing serial identification based on the URL video stream address and the initial data packet through a pre-deployed ship hull identification model, ship number identification model and abnormal behavior identification model.

[0048] The advantages of this invention over existing technologies are as follows:

[0049] This solution fully utilizes multi-channel surveillance video data to intelligently identify and process the real-time dynamics of fishing vessels entering and leaving the port using various artificial intelligence neural network algorithms. Its main functions include hull identification and vessel number identification for vessels entering and leaving the port, as well as identification of illegal electric welding and ice-boarding during the fishing moratorium. It fully leverages the existing video recording equipment to provide law enforcement officers with real-time information and evidence for investigation of illegal fishing vessels such as those going to sea illegally, unloading cargo at port, and using fake vessel names, greatly improving law enforcement efficiency.

[0050] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures pointed out in the description and the drawings. Attached Figure Description

[0051] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0052] Figure 1 This is a flowchart of the identification method of the present invention;

[0053] Figure 2 for Figure 1 The step-by-step flowchart of step S3;

[0054] Figure 3 for Figure 1 Application flowchart of step S3;

[0055] Figure 4 This is a schematic diagram of the identification system of the present invention;

[0056] Figure 5 This is a schematic diagram illustrating the application of the identification system of the present invention. Detailed Implementation

[0057] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0058] The illustrative embodiments and descriptions of the present invention are used to explain the invention, but are not intended to limit the invention. Furthermore, elements / components using the same or similar reference numerals in the drawings and embodiments are used to represent the same or similar parts.

[0059] The terms "first," "second," "S1," "S2," etc., used in this document do not specifically refer to any order or sequence, nor are they intended to limit the invention. They are merely used to distinguish elements or operations described using the same technical terms.

[0060] The directional terms used in this article, such as up, down, left, right, front, or back, are for reference only when referring to the accompanying drawings. Therefore, the use of directional terms is for illustrative purposes and not to limit this work.

[0061] The terms “include,” “including,” “have,” “contain,” etc., used in this article are all open-ended terms, meaning that they include but are not limited to.

[0062] The term "and / or" as used herein includes any or all of the things mentioned.

[0063] The term "multiple" in this article includes "two" and "more than two"; the term "multiple groups" in this article includes "two groups" and "more than two groups".

[0064] The terms "approximately," "about," etc., used herein are intended to modify any quantity or error that may vary slightly, but these slight variations or errors do not change the essence of the quantity or error. Generally, the range of slight variations or errors modified by such terms may be 20% in some embodiments, 10% in others, 5% in still others, or other values. Those skilled in the art should understand that the aforementioned values ​​can be adjusted according to actual needs and are not limited thereto.

[0065] Certain terms used to describe this application will be discussed below or elsewhere in this specification to provide additional guidance to those skilled in the art in describing the application.

[0066] Please refer to Figure 1 , Figure 1 This is a flowchart of the identification method of the present invention. Figure 1 As shown, a method for identifying abnormal ship behavior according to the present invention includes:

[0067] Address acquisition step S1: Obtain the URL video stream address of at least one video stream. In this step, the URL video stream address of the multiple camera devices in RTSP format is obtained through an external platform.

[0068] Initial data acquisition step S2: Obtain video source data and video parameter information based on the URL video stream address, and encapsulate the video source data and the video parameter information to form an initial data packet;

[0069] Step S3 for obtaining identification results: Based on the URL video stream address and the initial data packet, the identification results are obtained by serial identification through a pre-deployed ship hull identification model, ship number identification model and abnormal behavior identification model.

[0070] Based on this, the present invention directly reads image data frame by frame from each RTSP video input stream, and sequentially obtains the recognition results obtained by the ship hull recognition model, ship number recognition model and violation behavior recognition model according to the initial data packet to obtain the final recognition result.

[0071] In this embodiment, using JSON data packets for both the initial data packet and the final identification result is a preferred implementation method.

[0072] The initial data acquisition step S2 includes:

[0073] The number of data acquisition processes is determined based on the number of video streams.

[0074] The data acquisition process obtains the video source data corresponding to the URL video stream address based on the URL video stream address;

[0075] The URL video stream address is parsed to obtain multiple video parameter information, including: port name, port code, ID of the organization to which the video belongs, video device code, video channel code, video longitude, video latitude and other basic information;

[0076] The video parameter information is assembled into the initial data packet.

[0077] Please refer to Figures 2-3 , Figure 2 for Figure 1 The step-by-step flowchart of step S3; Figure 3 for Figure 1 The application flowchart for step S3. (See attached flowchart.) Figures 2-3 As shown, step S3, which involves obtaining the recognition result, includes:

[0078] Ship hull identification step S31: The video source data obtained based on the URL video stream address and the initial data packet are used to obtain a first identification result through the ship hull identification model;

[0079] Ship number identification step S32: Based on the first identification result, obtain the second identification result through the ship number identification model;

[0080] Abnormal behavior identification step S33: Based on the first identification result or the second identification result, obtain the final identification result through the abnormal behavior identification model.

[0081] Furthermore, the hull identification step S31 includes:

[0082] If the hull recognition model identifies a valid hull target, it obtains the hull type and coordinates of the valid hull target.

[0083] The effective hull target is marked on the original image of the image data frame of the initial data packet using a first bounding rectangle based on the coordinates of the effective hull target.

[0084] The original image of the labeled image data frame is sent to a file server, and the URL address returned by the file server is received.

[0085] The first cropped image is obtained by cropping the original image of the marked image data frame based on the first boundary rectangle, and the first cropped image is sent to the file server and the URL address returned by the file server is received.

[0086] The hull type, coordinates, and URL address are written into the initial data packet to form the first identification result.

[0087] Specifically, in this step, the URL video stream address and the initial data packet serve as input to the ship hull recognition model. This model is trained using the PPYOLOE+ object detection algorithm. If a valid ship hull target is identified, for example, if the threshold > 0.5, several bounding boxes (bboxes) are generated on the original image data frame. Each bbox corresponds to the coordinates of a identified ship hull position. The wholeUrls field in the ship hull recognition result boat_hull_json corresponds to the URL address returned after the result image with multiple bboxes marked on the original image is uploaded to the file server. The partUrls field corresponds to the URL address returned after a portion of the image captured from each bbox is uploaded to the file server. At this point, because the ship number recognition model has not yet been entered, the boatNum field is temporarily empty. The first recognition result (boat_hull_json) is:

[0088]

[0089]

[0090] Furthermore, the ship number identification step S32 includes:

[0091] The captured image is identified using the ship number recognition model.

[0092] If the vessel number recognition model identifies a valid vessel number target, it obtains the vessel license plate number area and vessel number of the valid vessel number target;

[0093] Mark the vessel license plate number area and vessel number of the valid vessel number target in the first captured image to obtain the second captured image;

[0094] The marked second cropped image is sent to the file server, and the URL address returned by the file server is received;

[0095] The ship number and URL address are written into the first identification result to form the second identification result.

[0096] Specifically, in this step, the list of bounding boxes generated by the hull recognition model is iterated through, and a screenshot of each hull section is taken, i.e., the first screenshot, which serves as the input data frame for the ship number recognition model. After the ship number recognition model's inference process, if a valid ship number is detected and recognized, the boatNum field in the input hull recognition result boat_hull_json is updated, and the ship number area and text recognition result are highlighted on the original screenshot corresponding to each hull bounding box. The corresponding URL address returned after uploading this image to the file server is used as the updated partUrls field, and the corresponding AI recognition JSON data packet at this time becomes the ship number recognition result boat_num_json format, and then it is sent as a Kafka producer to send Kafka messages.

[0097] In this embodiment, the ship number recognition model is trained, inferred, and deployed based on the Paddle OCR development kit within the PaddlePaddle framework. Unlike vehicle license plates, which are fixed, ship license plates are scattered and highly irregular. We collected nearly 6,000 ship license plate photos and randomly divided them into a training set and a test set with a ratio of 8:2. Due to the relatively small dataset, to ensure better and faster model convergence, we selected the PP-OCRv3 model from Paddle OCR for text detection and recognition, and used the PP-OCRv3 model parameters as the pre-trained model.

[0098] The second recognition result (boat_num_json) is:

[0099]

[0100]

[0101]

[0102] It should be noted that if a valid boat number is not detected and identified, the boatNum field will not be updated, and a Kafka message will be sent directly in the boat_hull_json format as the boat identification result.

[0103] Furthermore, the abnormal behavior identification step S33 includes:

[0104] The abnormal behavior recognition model is used to identify the first or second cropped image.

[0105] If the abnormal behavior recognition model identifies a valid abnormal behavior target, the abnormal behavior type of the valid abnormal behavior target is obtained;

[0106] The effective abnormal behavior target is marked on the first or second cropped image using a second bounding rectangle to obtain a third cropped image;

[0107] The marked third cropped image is sent to the file server, and the URL address returned by the file server is received;

[0108] The abnormal behavior type and URL address of the valid abnormal behavior target are written into the second identification result to obtain the final identification result.

[0109] The abnormal behavior identification step further includes: training the abnormal behavior identification model before identification, including:

[0110] The datasets with different abnormal behavior types obtained in advance are divided into initial training set, initial validation set and initial test set according to the abnormal behavior type;

[0111] The initial training set, initial validation set, and initial test set are fused together to obtain the final training set, final validation set, and final test set.

[0112] The abnormal behavior recognition model is trained using the final training set, the final validation set, and the final test set.

[0113] Specifically, if the ship number recognition model detects and identifies a valid ship number, in addition to sending a Kafka message in the format of boat_num_json, the data will then proceed to the violation behavior recognition model. This model merges the electric welding and ice-related behavior datasets, dividing them into training, validation, and test sets, and then performs corresponding data fusion. It uses the same PPYOLOE+ object detection algorithm model for joint training and inference. The ice-related behavior is judged based on the detection of the ice block (ice) feature, while the electric welding behavior is judged based on the simultaneous detection of both sparks (sparks) and masks (masks). Because both are object detection tasks, similar to the ship hull recognition model, if this violation behavior recognition model identifies a valid target (ice or (spark and mask), and the threshold > 0.5), the recognition result will generate several bounding boxes (bboxes) on the original screenshot corresponding to each ship hull bbox. Each bbox corresponds to one identified violation behavior target.

[0114] The wholeUrls field in the port_illegal_json violation identification result corresponds to the original screenshot of each ship's bbox, i.e., the second screenshot image. The result image with the violation target marked on it, i.e., the third screenshot image, is the corresponding URL address returned after being uploaded to the file server. The boatNum field is the valid ship number identified by the ship number recognition model. The illegalType field corresponds to 3 to indicate that the violation type is electric welding, and corresponds to 4 to indicate that the violation type is ice-related.

[0115] Considering that some violations may not occur on the hull itself but on the outside, logic has been added to identify violations even if the hull cannot be detected. The `boatNum` field is left empty by default. Finally, the violation identification result (`port_illegal_json`) is processed consistently when a Kafka producer sends a Kafka message. The format of the final identification result (`port_illegal_json`) is as follows:

[0116]

[0117]

[0118] The fields in the above identification results are explained as follows: boat name (boatNum), boat hull type (boatHullType, including the following categories: ore carrier, passenger ship, bulk cargo carrier, general cargo ship, container ship, fishing boat), enter / exit type (enterExitType, 0-unknown, 1-entering port, 2-exiting port), identification type (identType, 1-hull identification, 2-boat number identification), violation type (illegalType, 1-fishing moratorium, 2-suspected unregistered vessel, 3-illegal electric welding in port, 4-iceboarding during the fishing moratorium), identification time (identTime), confidence score (confidenceScore), whole screenshot attachment URL (wholeUrls, overall identification result of the image data frame), and part screenshot attachment URL (partUrls, partial identification result screenshot of the image data frame).

[0119] Please refer to Figures 4-5 , Figure 4 This is a schematic diagram of the identification system of the present invention; Figure 5 This is a schematic diagram illustrating the application of the identification system of the present invention. For example... Figures 4-5As shown, the present invention provides a system for identifying abnormal ship behavior, using the identification method described in any one of the above descriptions. The identification system 1 includes: an address acquisition unit 11, an initial data acquisition unit 12, and an identification result acquisition unit 13. The address acquisition unit 11 acquires the URL video stream address of at least one video stream through an external platform 2. The initial data acquisition unit 12 acquires video source data and video parameter information based on the URL video stream address, and encapsulates the video source data and the video parameter information to form an initial data packet. The identification result acquisition unit 13 obtains the identification result by performing serial identification based on the URL video stream address and the initial data packet through a pre-deployed ship hull identification model, ship number identification model, and abnormal behavior identification model.

[0120] The present invention will be described with reference to a specific embodiment. After the address acquisition unit 11 of the present invention obtains the URL video stream addresses in RTSP format from the peripheral platform 2 for each camera device that needs to be accessed, the initial data acquisition unit 12 performs address parsing of basic information such as port name, port code, ID of the organization to which the video belongs, video device encoding, video channel encoding, video longitude, and video latitude. After the recognition result acquisition unit 13 generates recognition data in JSON format, the recognition result includes uploading each recognition result image to the file server 3 and returning the corresponding image URL address, and then packaging it into JSON format recognition data. Finally, this JSON format recognition data is pushed to the Kafka server cluster 4 as a Kafka producer message. Then, the Smart Ocean backend system 5, as a Kafka consumer, will pull the corresponding JSON recognition data packets from the Kafka server cluster 4 in real time.

[0121] Furthermore, the initial data acquisition unit 12 includes a multi-task scheduling module 121 and an RTSP address resolution module 122;

[0122] The multi-task scheduling module 121 has multiple data acquisition processes. This module is the first logic executed after the AI ​​backend service system starts. It mainly performs multi-task scheduling. According to the actual project requirements, this system needs to support intelligent processing of up to 10 concurrent real-time video stream addresses. The server used is a Galaxy Kylin operating system CPU machine with more than 10 kernels. Therefore, through Python's multi-process mechanism, up to 10 processes are started at the same time according to the actual video stream input. Each process processes one real-time video stream. That is, the number of processes is determined by the number of video streams. The process inputs the video stream address and obtains the video source data according to the address.

[0123] The RTSP address resolution module 122 can parse the corresponding port name (portName), port code (portCode), ID of the organization to which the video belongs (orgCode), video device code (deviceCode), video channel code (channelId), video longitude (gpsX), video latitude (gpsY), and other information based on this RTSP video stream address. At the same time, it assembles the information parsed from the address into a basic JSON data packet, i.e., the initial data packet.

[0124] Furthermore, the identification result acquisition unit 13 includes: an AI intelligent algorithm analysis module 131 and a Kafka data production module 132, specifically:

[0125] The AI ​​intelligent algorithm analysis module 131 is the core of this recognition system. It directly reads image data frame by frame from each RTSP video input stream, that is, after obtaining the video source data based on the URL video stream address, it sequentially passes the data through the ship hull recognition model, ship number recognition model, and violation behavior recognition model. Based on the recognition results, it processes the basic JSON data packet passed from the previous module into a new AI recognition JSON data packet, which is the final recognition result. Specifically, for ship hull, ship number, and violation behavior recognition, AI recognition JSON data packets with different formats are output respectively.

[0126] Kafka data production module 132. Kafka is a distributed, partitioned, multi-replica distributed messaging system coordinated by ZooKeeper. Its biggest feature is its ability to process large amounts of data in real time to meet various needs. This recognition system acts as a Kafka producer. After messages are placed into the message queue, the queue pushes the messages to consumers who have subscribed to that type of message, thus enabling real-time processing of large amounts of AI recognition data. This solution uses an asynchronous sending method with a callback function. When using the send method, a callback function is specified. The server calls this function when responding, and processes the result through this callback function. That is, if the message writing fails, a maximum of 5 retries will be performed.

[0127] In summary, this invention provides an identification method and system for automatic identification of vessels entering and leaving ports and intelligent analysis of illegal activities in fishing ports. Its main functions include vessel hull identification, vessel number identification, and identification of illegal electric welding and ice-related behaviors during the fishing moratorium. Conventional logic typically assigns one AI model to each function, meaning that illegal electric welding and illegal ice-related behaviors are trained using different target detection algorithm models, with targeted training based on the relevant features of electric welding and ice. This solution adopts a joint feature training approach, merging the electric welding and ice-related behavior datasets into a unified training, validation, and test set, and using the same PPYOLOE+ target detection algorithm model for joint training. This simplifies model inference and deployment, increases the processing speed of the identification system, lowers hardware and manpower maintenance costs, and improves the overall system efficiency.

[0128] Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for identifying abnormal behavior of ships, characterized in that, include: Address acquisition steps: Obtain the URL of at least one video stream; Initial data acquisition steps: Obtain video source data and video parameter information based on the URL video stream address, and encapsulate the video source data and the video parameter information to form an initial data packet; Steps for obtaining identification results: Based on the URL video stream address and the initial data packet, the identification results are obtained by serial identification through a pre-deployed ship hull identification model, ship number identification model and abnormal behavior identification model. The initial data acquisition step includes: The number of data acquisition processes is determined based on the number of video streams. The data acquisition process obtains the video source data corresponding to the URL video stream address based on the URL video stream address; The URL video stream address is parsed to obtain multiple video parameter information; The video parameter information is assembled into the initial data packet; The steps for obtaining the recognition results include: Ship hull identification steps: The video source data obtained based on the URL video stream address and the initial data packet are used to obtain a first identification result through the ship hull identification model; Ship number identification steps: Based on the first identification result, obtain the second identification result through the ship number identification model; The hull identification step includes: If the hull recognition model identifies a valid hull target, it obtains the hull type and coordinates of the valid hull target. The effective hull target is marked on the original image of the image data frame of the initial data packet using a first bounding rectangle based on the coordinates of the effective hull target. The original image of the labeled image data frame is sent to a file server, and the URL address returned by the file server is received. The first cropped image is obtained by cropping the original image of the marked image data frame based on the first boundary rectangle, and the first cropped image is sent to the file server and the URL address returned by the file server is received. The first recognition result is formed by writing the ship type, coordinates, and URL addresses corresponding to the original image and the first captured image of the image data frame into the initial data packet.

2. The identification method as described in claim 1, characterized in that, The address acquisition step includes: Obtain the RTSP format URL video stream address of multiple camera devices through an external platform.

3. The identification method as described in claim 1, characterized in that, The step of obtaining the recognition result also includes: Abnormal behavior identification steps: Based on the first identification result or the second identification result, the final identification result is obtained through the abnormal behavior identification model.

4. The identification method as described in claim 3, characterized in that, The ship number identification step includes: The captured image is identified using the ship number recognition model. If the vessel number recognition model identifies a valid vessel number target, it obtains the vessel license plate number area and vessel number of the valid vessel number target; Mark the vessel license plate number area and vessel number of the valid vessel number target in the first captured image to obtain the second captured image; The marked second cropped image is sent to the file server, and the URL address returned by the file server is received; The ship number and the URL address corresponding to the second captured image are written into the first recognition result to form the second recognition result.

5. The identification method as described in claim 4, characterized in that, The abnormal behavior identification steps include: The abnormal behavior recognition model is used to identify the first or second cropped image. If the abnormal behavior recognition model identifies a valid abnormal behavior target, the abnormal behavior type of the valid abnormal behavior target is obtained; The effective abnormal behavior target is marked on the first or second cropped image using a second bounding rectangle to obtain a third cropped image; The marked third cropped image is sent to the file server, and the URL address returned by the file server is received; The abnormal behavior type of the valid abnormal behavior target and the URL address corresponding to the third captured image are written into the second recognition result to obtain the final recognition result.

6. The identification method as described in claim 5, characterized in that, The abnormal behavior identification step further includes: training the abnormal behavior identification model before identification, including: The datasets with different abnormal behavior types obtained in advance are divided into initial training set, initial validation set and initial test set according to the abnormal behavior type; The initial training set, initial validation set, and initial test set are fused together to obtain the final training set, final validation set, and final test set. The abnormal behavior recognition model is trained using the final training set, the final validation set, and the final test set.

7. A system for identifying abnormal ship behavior, characterized in that, The identification system, employing the identification method described in any one of claims 1-6, comprises: The address acquisition unit retrieves the URL video stream address of at least one video stream; The initial data acquisition unit acquires video source data and video parameter information based on the URL video stream address, and encapsulates the video source data and the video parameter information to form an initial data packet. The identification result acquisition unit obtains the identification result by performing serial identification based on the URL video stream address and the initial data packet through a pre-deployed ship hull identification model, ship number identification model and abnormal behavior identification model.

Images