Research On Ship Draught And Ship License Recognition Based On Deep Learning

Waterway transportation is an indispensable part of human integrated transportation network.With the continuous increase in the number of ships,the frequency of illegal activities such as ship overloading is also increasing,which brings huge potential safety hazards.Therefore,how to judge the ship over draught and identify the overloaded ship's identity is of great value.The license plate recognition technology based on deep learning has made a lot of important progress,but the application of related technologies in the field of waterway shipping is still lacking.This dissertation has conducted a certain research on the ship over draft and identification characters recognition based on deep learning.The main contributions of this dissertation are concluded as follows:This paper firstly builds a network of target detection algorithms based on YOLO v3,aiming to identify the ship's over draught status by water line on the image.In order to improve the accuracy of judgment,the method of re-identifying the overloaded ship is used by calculating the pixel height of the ship's side plate above the water to achieve another discrimination.Then,a simple target tracking method based on YOLO v3 algorithm was proposed,which realized the real-time tracking on the video surveillance.In order to warn and correct the overloaded ship,it is necessary to identify the ship's characters,including positioning,tilt correction,character segmentation and recognition.In this paper,a nameplate positioning extraction method based on YOLO v3 is proposed.The full image is divided into three equal parts by width and inputed into the network for detection.This method effectively solves the problem that the ship nameplate occupies a relatively small image.Aiming at the problem of license's character tilting,the characters are tilted in the horizontal direction.The single character on the license is divided by the shadow segmentation algorithm,and the YOLO v3 network model is trained to recognize the character.Finally,a limited Chinese database of ship names is established to match the misrecognized ship names according to the keyword search method.As a result,a high recognition accuracy rate has been achieved.Experiments show that the optimized ship over draught recognition system achieves a mean average precision(mAP)of 91.525%on the test set.The accuracy rate of the ship nameplate recognition system reaches 91.37%,and the detection speed reached 33 times per second,which meets the requirements of accuracy and real-time.The relevant research of this paper have been designed and implemented in the maritime monitoring system of Soochow Hexi Bridge's flight segment,having a good detection performance.