Underwater target tracking method based on convolutional neural network

Abstract

In order to solve the problems of low accuracy of underwater target tracking, poor real-time performance and large amount of calculation required, an underwater target tracking method based on the improved SiamRPN++ algorithm is adopted. By selecting the inverted residual bottleneck block to construct a new backbone network SmallMobileNet, instead of the backbone network ResNet-50 of the SiamRPN++algorithm, the use of deep separable convolution to reduce the amount of calculation, while ensuring accuracy and real-time performance, adjust The number of channels, layers, parameters of the network and the complexity of each segment of the network are used to reduce the computational cost and hardware requirements, so that the algorithm can be transplanted to the underwater tracking platform. Through experiments, compared with the original algorithm, the accuracy of the SiamRPN++ algorithm with Small-MobileNet as the backbone network is improved, the amount of network parameters and calculations are reduced, and the tracking speed is improved, which verifies the effectiveness of the method.