Research And Implementation Of Container Number Recognition Technology Based On Deep Learning

With the development of China's economy,more and more international trade and cargo transportation depend on containers.Container reduces the cost of goods transportation and accelerates the development of the world economy.However,its wide application leads to the acceleration of the throughput growth of the port,which easily leads to the congestion and chaos of the major ports,and causes great pressure on the management of the port.In order to better control the container,the container number identification system came into being.Traditional container number recognition technology is vulnerable to complex environment,including lighting,character skew,distortion,damage and so on.In recent years,deep learning has become an important research field of pattern recognition because of its fast and accurate advantages.Therefore,this paper proposes a container number recognition algorithm based on deep learning.The algorithm is mainly divided into three parts: container number region location,the segmentation of container number character and the recognition of container number character.The main work of this paper is as follows:(1)Location of container number areaTraditional container number positioning method generally adopts mathematical morphology,character edge,character structure positioning method,which is easy to be affected by complex environment,such as stains,rust,door lock lever,corrugated and so on.To solve the above problems,this paper proposes a container number detection algorithm based on yolov4.In this paper,the captured image is used for data enhancement,and the container image data set is generated.Combined with the characteristics of the container number,the image is annotated.The number features are extracted by CSP packet network training,and the model suitable for detecting the container number is generated to locate the container number area.Experiments prove that the positioning method proposed in this article is fast and accurate.(2)Character segmentation of container numberIn order to solve the problems of container number skew,interference and damage,this paper adopts the container number segmentation method based on connected domain.This method traverses the 8 neighborhood of the seed point by depth,so as to determine the cutting position,screen and segment it.However,due to other interference around the characters,the segmented single image is filtered according to the prior knowledge of the numbered characters,so as to ensure the correct segmentation of the numbered characters and lay the foundation for character recognition.(3)Character recognition of container numberAiming at the similar character interference and character breakage of container number,this paper adopts a template matching algorithm with high recognition efficiency and accuracy.So as to enhance the precision of template matching recognition,according to the characteristics of the container number,the letters and numbers are recognized separately,which avoids the recognition error caused by the similarity of some letters and numbers,shortens the matching time of a single character,thereby enhancing the speed of the whole recognition.The experimental results show that the recognition method improves the recognition accuracy.