Classification for Time Series of Aerospace Targets with Deep Convolutional Neural Networks

Aerospace target classification is very important for strategic early warning and defense, and radar cross section (RCS) contains a wealth of information about aerospace targets. Based on RCS, conventional methods for aerospace target classification need to construct artificial features, and there is little explanation for the classification results. In this paper, an intelligent and partly interpretable method is proposed for classifying a swarm of aerospace targets, including cone targets, false targets, engines and so on. The fully convolutional neural network is adopted to extract the features of time series and to achieve the classification of different targets. Moreover, the class activation map is introduced to highlight the subsequences that contribute the most to a certain classification, and Gramian Angular Summation Field is used to visualize the time series and the weights in the neural network. Simulation results verify the effectiveness of the proposed method for aerospace target classification and analysis.