Ship Plate Text Detection And Recognition In Complex Scene Based On Deep Learning

Ship detection and identification is an important part of intelligent port management.It aims to grasp the information of ships entering and leaving the port,to provide support for navigation evaluation,to combat illegal berthing,illegal departure and other irregularities,and to ensure port traffic safety.In the scene of port camera surveillance,license plate detection and recognition technology based on image analysis has become an important means to realize ship identification.Although image analysis technology is relatively mature in the field of license plate detection and recognition,the technology of license plate text detection and recognition based on image analysis is facing severe challenges because of the complex background of license plate,the complexity and diversity of license plate text,the significant difference of text characters and the inconsistency of font types.In this context,this paper proposes a ship license plate text detection and recognition system SP-TDC(Ship Plate Text Detection and Recognition)based on in-depth learning in complex background.In order to avoid the complex background in the image affecting the accuracy of license plate text detection and recognition,an improved YOLO(You Only Look Once)detection algorithm is used to detect ship targets and narrow the detection range.In view of the challenges brought by the complexity and diversity of license plate text detection,a text detection algorithm TDNet(Text Detection Network)based on full convolution neural network is proposed.To deal with the challenges brought to license plate text recognition by large character differences and different font types,an online adaptive network structure TCNet(Text Classification Network)is proposed to realize license plate text classification and recognition.To speed up the detection efficiency,TDNet and TCNet are combined into an end-to-end neural network structure TDCNet(Text Detection and Classification Network)through feature sharing strategy.Network).In SP-TDC system,surveillance video stream is captured by camera,then decoded to generate continuous image frames,and ship target detection is carried out to intercept ship area.In ship area,ship license plate text is detected and recognized by TDCNet.Experiments show that the ship license plate text detection and recognition system SPTDC proposed in this paper can continuously process video data under the port surveillance camera,and has good adaptability,high accuracy and processing speed.