基于双隐含层BP算法的激光主动成像识别系统  

Laser active imaging and recognition system based on double hidden layer BP algorithm

作　　者：王灿进 孙涛[1] 石宁宁[3] 王锐[1] 王挺峰[1] 王卫兵[1] 郭劲[1] 陈娟[1] 

机构地区：[1]中国科学院长春光学精密机械与物理研究所激光与物质相互作用国家重点实验室,吉林长春130033 [2]中国科学院大学,北京100049 [3]北京航天自动控制研究所,北京100039

出　　处：《光学精密工程》2014年第6期1639-1647,共9页Optics and Precision Engineering

基　　金：吉林省重大科技攻关专项资助项目(No.20120615)

摘　　要：在传统激光主动成像系统的基础上,结合目标识别技术搭建了一个激光主动成像识别系统实验平台,用于研究激光主动成像后的目标识别。介绍了实验平台的工作原理,基于Hu矩特征的双隐含层BP神经网络算法以及实验处理流程和实验结果。特征量由7个不变Hu矩构成,通过240张原始目标样本库对由136个权值系数构成的双隐含层BP神经网络算法进行了训练。利用训练好的双隐含层BP算法对黑夜条件下远处的运动目标--43式冲锋模具枪进行了实验研究,成功获得了清晰的红外激光主动成像效果。实验显示对450m处2 740帧和550m处2 420帧激光主动成像图像的统计识别率达到了68.87%和72.11%,其中旋转变换下的统计识别率可达80.05%和84%,好于仿射变换的识别效果。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 alter 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 librar- ies. 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 2 740 frames of images at 450 m and 72.11% for 2 420 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.

关 键 词：激光主动成像 图像识别系统 Hu矩特征量 双隐含层BP算法 目标识别率 

分 类 号：TN249[电子电信—物理电子学] TP391[自动化与计算机技术—计算机应用技术]