A multi-source integrated ship identification method

Abstract

This invention discloses a multi-element fusion method for ship identification, including video input, ship target detection, ship image location acquisition, conversion to actual location, AIS data acquisition, AIS data parsing, ship AIS location acquisition, AIS ship name acquisition, ship image acquisition, ship name location recognition, acquisition of image ship name, and recognition of both AIS and image ship names. This invention combines the advantages of AIS and video image recognition technologies, enabling ship identification even when AIS is not enabled, thus improving the accuracy of ship identification and providing a foundation for maritime law enforcement and supervision. Furthermore, the cross-verification and comparison between the AIS ship name and the image-recognized ship name enhances the reliability of ship identification. This solves the problem of unidentifiable ships without AIS enabled and allows for the use of AIS as a supplementary identifier for ships with non-standard names, further improving the accuracy of ship identification and demonstrating promising application prospects.

Description

Technical Field

[0001] This invention relates to the technical field of ship supervision, specifically a multi-dimensional integrated ship identification method. Background Technology

[0002] Vessel identification methods are information processing methods used in maritime law enforcement, maritime supervision, and vessel monitoring. Currently, there are three main methods: First, AIS identification. AIS is the Automatic Identification System, allowing vessels equipped with AIS to actively exchange important information such as position, speed, course, name, and call sign. However, this method is passive, and situations such as vessels not equipped with AIS, or AIS being manually turned off or tampered with, are common. Second, RFID identification. Vessels on inland waterways are equipped with marine RFID electronic tags. These tags are used to identify the vessel and are installed on... RFID readers on the riverbank can identify RFID tags within 300 meters to achieve the purpose of identification. However, once a ship approaches the identification range, it can be identified, which may result in multiple ships being identified at the same time, and it is impossible to distinguish the identities of multiple ships. Finally, there is the video image recognition method, which aims to identify ships by collecting photos and recognizing the ship name characters. However, many inland waterway ships have non-standard names, with problems such as unclear markings, obscured names, and blurred names. Relying solely on image recognition of ship name characters is difficult, has low accuracy, and low reliability. Therefore, a multi-dimensional and integrated ship identification method needs to be designed. Summary of the Invention

[0003] This invention provides a multi-source integrated ship identification method that can solve the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides the following technical solution:

[0005] A multi-source integrated ship identification method, the ship identification method comprising the following steps:

[0006] Step 1: Detect the ship's position coordinates in the image using a background modeling method based on the video.

[0007] Step 2: Obtain a high-resolution image of the ship based on its position coordinates in the image, and use deep learning methods to locate and identify the ship name region to obtain the image-recognized ship name;

[0008] Step 3: Use the AIS receiving terminal to obtain the ship's AIS information and parse the ship's AIS information to obtain the parsed information, including: AIS ship name and AIS ship position;

[0009] Step 4: Transform the ship's position coordinates in the image into the ship's position coordinates in the world coordinate system using the camera imaging principle. Match the ship's position coordinates in the world coordinate system with the AIS ship position obtained in Step 3. If the matching conditions are met, it is considered a match and proceeds to Step 5. Otherwise, the ship name obtained from the image recognition in Step 2 is used as the final ship identity.

[0010] Step 5: When a ship has both an AIS ship name and an image recognition ship name, determine whether the two are equal. If they are equal, the image recognition ship name is used as the final ship identity. If they are not equal, determine whether the confidence level of the image recognition ship name is greater than the second preset threshold. If it is, the image recognition ship name is used as the final ship identity; otherwise, the AIS ship name is used as the final ship identity.

[0011] Preferably, in step two, a pre-trained YOLOv5 detection model is used to locate and identify the ship name region. The training process of the YOLOv5 detection model is as follows: images containing ship names in different scenarios, including daytime, nighttime, rainy days, and foggy days, are collected as a dataset, and the positions of the ship names in the dataset are manually labeled; the manually labeled dataset is enriched using a mosaic enhancement mechanism, that is, four images are randomly selected as a combination, randomly cropped and scaled, and then randomly arranged and spliced ​​to form a new image. The above operation is repeated until there are no new combinations. The YOLOv5 detection model is trained using the enriched dataset to obtain a trained YOLOv5 detection model.

[0012] The trained YOLO v5 detection model is used to identify the ship name region, and the ship name is obtained from the image.

[0013] Preferably, in step two, a high-resolution image of the ship is acquired using a capture camera based on the ship's position coordinates in the image, and a deep learning method is used to locate and recognize the text on the ship to obtain the text recognition result, including the position, confidence level Conf, and recognized content information Txt. The text recognition result is further judged, and the judgment logic is as follows:

[0014]

[0015] in,( x min , y min () represents the coordinates of the top-left pixel of the text recognition result. x max , y max () represents the pixel coordinates of the bottom right corner of the text recognition result. nThe number of characters in the Txt file is the final ship name recognition result if the above judgment logic is satisfied.

[0016] Preferably, in step four, the ship's position coordinates in the image are transformed into the ship's position coordinates in the world coordinate system using the camera imaging principle. x p , y p}, the coordinates of that position { x p , y p} and the AIS ship position obtained from step three analysis { x w , y w} Perform a match if the following matching conditions are met:

[0017]

[0018] Then it is considered a match.

[0019] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention combines the advantages of AIS and video image recognition technologies, enabling the acquisition of vessel identity even when AIS is not enabled, thus improving the accuracy of vessel identification and providing a foundation for maritime law enforcement and supervision. Secondly, when comparing and verifying the vessel name in AIS with the image recognition, the reliability of vessel identity is improved. This not only solves the problem of unidentifiable vessel identities when AIS is not enabled, but also allows for the use of AIS as a supplementary identity for vessels with non-standard names, further improving the accuracy of vessel identification and demonstrating promising application prospects. Attached Figure Description

[0020] Figure 1 This is a flowchart of the algorithm for the multi-element fusion ship identification method of the present invention. Detailed Implementation

[0021] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Example

[0024] Please see Figure 1This invention provides a multi-source integrated ship identification method, comprising the following steps:

[0025] Step 1: Detect the ship's position coordinates in the image using a background modeling method based on the video.

[0026] Step 2: Obtain a high-resolution image of the ship based on its position coordinates in the image, and use deep learning methods to locate and identify the ship name region to obtain the image-recognized ship name;

[0027] Step 3: Use the AIS receiving terminal to obtain the ship's AIS information and parse the ship's AIS information to obtain the parsed information, including: AIS ship name and AIS ship position;

[0028] Step 4: Transform the ship's position coordinates in the image into the ship's position coordinates in the world coordinate system using the camera imaging principle. Match the ship's position coordinates in the world coordinate system with the AIS ship position obtained in Step 3. If the matching conditions are met, it is considered a match and proceeds to Step 5. Otherwise, the ship name obtained from the image recognition in Step 2 is used as the final ship identity.

[0029] Step 5: When a ship has both an AIS ship name and an image recognition ship name, determine whether the two are equal. If they are equal, the image recognition ship name is used as the final ship identity. If they are not equal, determine whether the confidence level of the image recognition ship name is greater than the second preset threshold. If it is, the image recognition ship name is used as the final ship identity; otherwise, the AIS ship name is used as the final ship identity.

[0030] In step one, a background modeling method is used to detect the set of changing foreground connected components in the video. S i | i =1, 2, 3… n The area of ​​the root connected region. n Aspect Ratio of the Circumscribed Rectangle of the Connected Component r Rectangularity q ,width w m ,high h m After further filtering of the parameters, the final area obtained is the ship area, which gives the image location of the ship.

[0031]

[0032] In step two, a high-resolution image of the ship is captured by a camera based on the image location coordinates. A deep learning algorithm is then used to locate and recognize text on the ship. Each text recognition result has a location... Confidence level ConfThe system uses the identified content information Txt to further evaluate all text recognition results. The evaluation logic is as follows, which yields the final ship name recognition result.

[0033]

[0034] In step three, the AIS receiving terminal is used to acquire the ship's AIS information and parse it. The main information parsed includes: the AIS ship name and the ship's position. n It represents the number of characters in the Txt file.

[0035] In step four, the ship's position in the image is transformed into its position in the world coordinate system using the camera imaging principle. This location corresponds to the ship's location in AIS. The system performs a matching process. If the following matching conditions are met, the system is considered to have matched and the AIS ship name is obtained. If no match is found, the ship name is determined by image recognition and becomes the final ship identity.

[0036]

[0037] In step five, when a vessel has both an AIS name and an image recognition name, it is first determined whether the two are equal. If they are equal, the final identity is the image recognition name. If they are not equal, it is determined whether the confidence level of the image recognition name is greater than the threshold. If it is, the image recognition name is the final vessel identity; otherwise, the final identity is the AIS name.

[0038] This invention combines the advantages of AIS and video image recognition technologies, enabling the acquisition of vessel identity even when AIS is not activated, thus improving the accuracy of vessel identification and providing a foundation for maritime law enforcement and supervision. Secondly, it enhances the reliability of vessel identity by cross-checking and comparing the AIS vessel name with the image-recognized vessel name. This not only solves the problem of unidentifiable vessel identities when AIS is not activated, but also allows AIS to be used as a supplementary identifier for vessels with non-standard names, further improving the accuracy of vessel identification.

[0039] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A multi-element fusion ship identity recognition method, characterized in that: The ship identity recognition method comprises the following steps: Step one, based on video, the position coordinates of the ship in the image are detected by using a background modeling method; Step two, according to the position coordinates of the ship in the image, a high-definition image of the ship is obtained, and a deep learning method is used to position and identify the ship name area to obtain an image-recognized ship name; Step three, the AIS information of the ship is obtained by using an AIS receiving terminal, and the AIS information of the ship is analyzed to obtain analysis information, including the AIS ship name and the AIS ship position; Step four, the position coordinates of the ship in the image are transformed into the position coordinates of the ship in the world coordinate system through the imaging principle of the camera, and the position coordinates of the ship in the world coordinate system are matched with the AIS ship position obtained in step three, if the matching condition is met, it is considered that the matching is successful, and step five is entered, otherwise, the image-recognized ship name obtained in step two is taken as the final ship identity; Step five, when the ship has both an AIS ship name and an image-recognized ship name, it is judged whether the two are equal, if they are equal, the image-recognized ship name is taken as the final ship identity, if they are not equal, it is judged whether the confidence of the image-recognized ship name is greater than a second preset threshold, if it is, the image-recognized ship name is taken as the final ship identity, otherwise, the AIS ship name is taken as the final ship identity; In step two, a pre-trained yolo v5 detection model is used to position and identify the ship name area, wherein the training process of the yolo v5 detection model is as follows: images containing ship names in different scenes including daytime, nighttime, rainy day and foggy day are collected as a data set, and the ship name positions in the data set are manually labeled; the manually labeled data set is enriched by using a mosaic enhancement mechanism, that is, four images are randomly selected as a combination, and after random cropping and scaling, a new image is formed by random arrangement and splicing, the above operation is repeated until there is no new combination, the enriched data set is used to train the yolo v5 detection model, and a trained yolo v5 detection model is obtained; The trained yolo v5 detection model is used to identify the ship name area to obtain an image-recognized ship name; In step two, according to the position coordinates of the ship in the image, a high-definition image of the ship is obtained by using a snapshot camera, and a deep learning method is used to position and identify the text on the ship to obtain a text recognition result, including position, confidence Conf and identification content information Txt, the text recognition result is further judged, and the judgment logic is as follows: ； wherein, (x, y) is the top-left pixel coordinate of the text recognition result, x min , y min (x, y) is the top-left pixel coordinate of the text recognition result, x max , y max (x, y) is the bottom-right pixel coordinate of the text recognition result, n is the number of characters in the text, and if the above judgment logic is met, the final ship name recognition result is obtained. In the fourth step, the position coordinates of the ship in the image are transformed into the position coordinates of the ship in the world coordinate system through the camera imaging principle x p , y p} are matched with the AIS ship position { x p , y p} obtained in step three, and if the following matching conditions are met: x w , y w} are met: ； It is considered that the matching is successful.

Images