Design And Implementation Of Unattended Platform For Ship Target Image Recognitionn

Ship targets are the main body of marine activities.Intelligent supervision of ship targets is one of the key links in building a maritime power,and has important applicable value.In the military field,it can be used for military reconnaissance,military intelligence collection,etc;In the civilian field,it can be used for maritime traffic control,marine economic zone protection,and illegal fishing.At present,most ship target recognition methods are manual recognition,which shows the characteristics of low intelligence and cannot meet the needs of intelligent supervision.Therefore,an unmanned platform for ship target image recognition is developed to solve the problems of limited power resources,limited computing resources,limited communication resources,difficult installation,deployment,maintenance,and low intelligence in the supervision process in the marine environment in this thesis.The main content of this thesis is as follows:1.An unattended platform for ship target image recognition is designed.In this platform,(a)The problem of platform power limitation is solved by adopting solar energy autonomous power supply method.(b)The problem of platform real-time communication and transmission limitation is solved by using ship target intelligence data composed of text and key frame images.This method greatly reduces the amount of communication data,compared with transmitting real-time video.(c)The platform real-time recognition processing problem is solved by optimizing the accuracy and realtime performance of the existing deep learning network.Results compared with experiment indicate that the platform can realize reliable,real-time and accurate identification of ship targets on the sea,and shows the characteristics of intelligence,compactness,easy deployment,and maintenance-free.2.Aiming at the demand of real-time and accurate recognition of ship target by unmanned platform for ship target image recognition,a ship target recognition method based on improved YOLOV3 is proposed.Firstly,the accuracy of ship target recognition is optimized by using the prediction optimization strategy based on pyramid pooling mechanism,the enhanced feature expression strategy based on hybrid attention mechanism and the anchor frame optimization strategy based on prior knowledge.Secondly,the channel clipping strategy based on error distribution is used to optimize the real-time performance of ship target recognition.The self-built ship target dataset is tested and verified,the results show that under the condition that the model size is15.8% of the original YOLOV3,the processing frame rate is 43 FPS,the ship target recognition accuracy rate is 88.43%,and the ship target recognition miss rate is 5.9%,which meets the real-time and accurate identification requirements of the platform.3.Based on the embedded platform,an unmanned platform for ship target image recognition was developed by comprehensively using development tools such as Python,Pytorch,and PyQt.The platform includes functional modules such as solar power supply,video collection,intelligent identification,intelligence generation,intelligent alarm,and data transmission.Then the function and performance test of the platform is carried out to verify the feasibility of the platform design and implementation scheme.