Rasearch On Maritime Ship Target Detection Method And FPGA Implementation Base On YOLOv5

The real-time observation of the movement of marine naval targets not only defends the sovereign integrity of our territorial waters against enemies,but also support various maritime tasks,such as the exploration of maritime resource,the rescue of ships,and so on.Compared with other means of ship image acquisition,the mean of visible light acquisition maintains a longer work duration,higher image resolution,and lower operation cost,which has become one of the research focuses recently.Currently,because of the capability of its own feature extraction,the convolutional neural network based on deep learning is widely used in the field of target detection.However,the complexity of the model and the number of computations complicate the algorithm of the target detection,which makes it difficult to explore the algorithm of the field of the ship detection terminal application and to process the field data simultaneously.The severe weather conditions at sea also lower the accuracy of ship detection in the actual processing.To solve the above problems,this paper provides an image pre-processing algorithm,an improved target detection algorithm,and a detection system of the maritime target ship,which implements FPGA as the computing device.The focus of the paper has been demonstrated into four major perspectives below.(1)Maritime ship surveillance is susceptible to severe weather conditions,such as haze,low light,and rainfall.In order to improve the accuracy of ship targets detection under adverse weather conditions,the method of image pre-processing has been implemented to optimize the data set: the images in haze conditions are synthesized by the atmospheric scattering model(ATSM);the images in low-light conditions are synthesized by the Retinex Theory;the images in rain conditions are synthesized by random noise and filters.(2)Most of the existing target detection algorithms have problems of high complexity and a large amount of computation.In order to reduce the complexity and improve the performance of the algorithm,this paper proposes the CO-YOLO Algorithm on the basis of YOLOv5.In terms of feature extraction capability,the Cross-stage One-shot Aggregation Module is designed.In terms of feature fusion structure,the Bi FPN structure is optimized.In terms of the loss function,the Alpha-Io U loss function is used to replace the C-Io U loss function.Finally,CO-YOLO Algorithm is compared with the classical target algorithm.(3)The paper designs an accelerated system based on FPGA to solve the problem that it is difficult to process the field data simultaneously in the terminal field.The hardware design of this accelerated system is based on actual resources.The deployment of the algorithm hinges on the structure of the target detection algorithm.Finally,the efficacy of the acceleration system is contrasted with the ones on different hardware platforms.(4)According to the previous three perspectives,a maritime ship target detection system has been designed to achieve visual monitoring,embracing the Py Qt framework.The Front-End adopts the QT to design the page,while the Back-End utilizes Python to develop the server.SQLite is adopted as the database.The design of the system is also explained specifically and its performance is measured by system testing.