| With the adjustment of marine strategies,many countries pay attention to the management of marine resources and information.The target detection is widely used in various fields with the development of deep learning,the intelligent monitoring and management of Marine has become a trend.However,there is no unified platform to manage the information of sea.The functions of each platform are relatively single,and the module division is not clear.At the same time,the target detection algorithms are mostly applied on land,they have problems such as large illumination impact and single background on the sea,which makes the detection effect of the target unsatisfactory,the target detection algorithm requires high hardware equipment,and the model of the algorithm is too huge to deploy to the general equipment.Based on the above problems,a lightweight target detection algorithm is designed and integrated into data management system of intelligent terminal on board,which can which can manage the sea monitoring information and monitor the surrounding ships.It also integrates the target detection algorithm to monitor the ships in the surrounding sea area,improve the information management efficiency and supervision intensity,and ensure the navigation safety.The main work of this dissertation is as follows.(1)Based on the YOLOv5 algorithm and specific application scenarios,the dissertation designs and improves the target detection network.It adds a ghost module and a deep separable convolution,which reduces the parameters of the network and compresses the network model.At the same time,Bi FPN is used for feature extraction in the neck layer,and attention mechanism is added after the convolution layer,so that the amount of calculation is reduced without loss of accuracy.In terms of loss function,CIOU＿Loss is used to optimize the final calculation result.The improved algorithm is compared with the YOLOv3,YOLOv4 and YOLOv5,and the experimental results are analyzed.(2)Since there is no public data set of ships,this study has searched and labeled the data set.The data set mainly includes two parts: one part is the marine video taken by the laboratory personnel on the sea and the other part is the marine video is taken by the shore camera.We extract the video frame by frame to obtain images as a part of the ship data set,and the other part is to obtain ship images on the internet through crawlers.We filter the two parts of ship images to obtain the final data set which can be used for ship target detection.The data set is divided into five categories: official ship,passenger ship,cargo ship,fishing ship,and sailing ship according to the types of common ships,and labeled by the label Img tool.(3)The data management system of intelligent terminal on board is designed and implemented.The equipment for information monitoring and collection on the ship can have a unified information platform for management,and can monitor the ships in the surrounding sea area.This system mainly uses the flask framework,the front-end uses Java Script,CSS and HTML to write,and the back-end uses Python language to write.At the same time,we design the whole system architecture,module functions and database. |
