Research On Target Tracking Algorithm Based On Convolutional Neural Network

As the future development trend of unattended substations,the key is to use intelligent means to achieve unattended operation while ensuring the safety of substation operation.At present,unmanned substations mainly use remote monitoring systems for patrol inspections and rely on manual methods to find abnormal conditions.However,for a large number of live videos,this method will be inefficient and untimely.In addition,there will be misjudgments caused by the decline of staffs attention.In order to solve these problems,this paper proposes to use a deep convolutional neural network model to track targets in an unattended substation,which not only reduces manual reliance,but also improves the tracking accuracy while improving work efficiency.The main work of this article includes the following two aspects:After studying the Siamese Network and SiamFC(Full-convolutional Siamese Network)target tracking algorithms based on convolutional neural networks,this paper proposes an improved SiamFC algorithm to realize the target tracking of unattended substation.The SiamFC algorithm is an end-to-end target tracking algorithm,but the prediction frame size of this algorithm is fixed,so it cannot track the target with large scale change.The regional recommendation network RPN can adjust the size and position of the prediction frame through the candidate frame mechanism and the return of the boundary frame,so that the prediction frame can match the size of the tracking target.Therefore,this paper combines the RPN of the regional recommendation network with the siamfc algorithm,proposes an improved target tracking model,and enhances the adaptability of the model to different scales.In this paper,the experimental results of the improved algorithm and the original algorithm are compared.Firstly,the experimental environment is built,then the OTB data set and VOT data set which are commonly used in the field of target tracking are used to expand the experimental data set.Finally,the model is trained and tested by the preprocessed data set.The experimental results show that the improved SiamFC algorithm is superior to the original algorithm in accuracy,robustness and real-time.It can adapt to the change of target scale,realize the intelligent tracking of unattended substation target,and provide reliable technical support for the safe and stable operation of the substation.