Research On Ship Identification And Tracking Technology Based On Convolutional Neural Network

China is rich in waterway resources,with a long coastline and numerous inland waterways,ensuring the safety of ships and improving the efficiency of navigation has always been a research hotspot in the field of shipping,and various ship monitoring systems have very important application values.At present,ship monitoring equipment that has been maturely applied,such as AIS and radar,have played an important role in the field of ship supervision and monitoring.In recent years,with the research progress and achievements made by computer vision technology in target recognition tracking,the computer vision-based ship supervision system as a supplement to the above-mentioned supervision equipment in the field of ship supervision has very important application value.After years of research by a large number of scholars,some robust target detection algorithms based on artificial feature extraction and background modeling have been applied to ship target detection,but this moving target detection algorithm for fixed backgrounds cannot adapt to the changing background Ship detection and tracking tasks,the algorithm is susceptible to problems such as light,waves and occlusion.At present,the target detection and tracking algorithm based on convolutional neural network is becoming more and more mature.Since the convolutional neural network algorithm extracts the features in the image through convolution calculation,it can deal with complex water conditions well.However,the network structure of the algorithm is complex,and the large amount of calculation results in low detection efficiency of the algorithm.In order to solve the real-time problem in the process of ship target recognition and tracking,and improve the detection efficiency of the target detection algorithm based on convolutional neural network,this paper uses a lightweight network structure to improve the target detection algorithm based on convolutional neural network.Improve the algorithm’s target detection efficiency,and realize online recognition and tracking through convolutional neural network algorithm combined with filtering algorithm.The main research contents of this article are as follows:First complete the construction of the development environment of Tensorflow and OpenCV.Study the image preprocessing techniques such as graying,dilation and corrosion.The KNN background subtraction method and three-frame difference method are reproduced,and the target detection effect of the actual patrol ship video test algorithm is used.Secondly,Python crawler technology is used to collect ship image data,image enhancement technology is used to enrich the,data,and ship data is constructed for the training of algorithm models based on convolutional neural networks to achieve the purpose of classifying specific ships.Then the lightweight structure based on convolutional neural network is studied.Based on the InceptionV2 lightweight’ network structure,the feature extraction part of the Faster R-CNN algorithm is improved,the algorithm structure is optimized,and the algorithm detection efficiency is improved.Use MobilenetV2 lightweight network structure to improve the SSD algorithm,improve the algorithm backbone network,and improve the detection efficiency of the algorithm.Finally,Kalman filter algorithm combined with SSD_MobilenetV2 algorithm is used to realize online identification and tracking of ship targets.After the video is input,the target detection algorithm based on the convolutional neural network is used to detect the ship target in the video frame,and the Kalman filter is used as a tracker to predict and track the target position of the detected ship.Use the Hungarian algorithm to solve the problem of matching the detection frame and the tracker,determine the optimal correlation between the detection frame and the tracker,and achieve multi-target tracking.The identification and tracking results are displayed on the terminal through the streaming media server.The patrol ship video is used as the experimental object to verify the recognition and tracking effect of the algorithm.The results show that the SSD_MobilenetV2 algorithm combined with Kalman filtering can achieve higher recognition rate and better tracking effect in recognition and tracking.While achieving accurate detection,it can meet the real-time requirements of online identification and tracking,thereby meeting the identification and tracking requirements of the ship’s video surveillance system,and has application value.