Infrared Target Recognition Based On Improved Convolution Neural Network

Abstract

Automatic target recognition is one of the key technologies for infrared imaging precision guided weapon systems, aiming at the problem of complex target feature modeling and low recognition rate in the traditional recognition algorithm, this paper proposed the convolution neural network method based on improved the Dropout layer. Firstly, this paper adjusted the number of convolution layers and pooled layers in combination with infrared target characteristics and improved the convolution neural network ZFNet model. Secondly, this paper analyzed the Dropout layer and the change of the discard rate by visualization during the process of training the model. Then this paper determined the selection principle of Dropout discard rate and analyzed the effect of the Dropout layer on the recognition results. The results show that the improved convolution neural network test accuracy rate is 92.08%, which outperforms the traditional algorithm. The method obviously improves the classification accuracy, and has good generalization ability and robustness, it can provide reference for the design of infrared imaging seeker target recognition algorithm.