基于双隐含层BP算法的激光主动成像识别系统

Abstract

An experiment platform for laser active imaging and recognition was established based on the traditional laser active imaging system to investigate the target recognition after laser active imaging. The working mechanism of the platform was introduced and the Hu moment feature based BP neural network algorithm with double hidden layers and an experimental process were given. The target feature vector was consisted of seven invariant Hu moments. The BP neural network algorithm with double hidden layers including 136 weight coefficients was trained by 240 original sample libraries. The trained BP neural network algorithm was used to research a distance moving target in the dark condition, a model of 43 submachine gun, and a clear infrared laser active image was obtained. Experiment results show that statistical recognition probability is 68.87% for 2740 frames of images at 450 m and 72.11% for 2420 frames of images at 550 m. The corresponding recognition probabilities from rotation transformation are 80.05% and 84%, respectively, which is better than the results by affine transformation.