改进的对向传播网络及其在多传感器目标识别中的应用

针对多传感器数据融合和目标识别的特点,提出了改进的对向传播网络(MCPN),并与Dempster-Shafer(D-S)证据推理相结合,实现了决策层数据融合目标识别.文中利用仿真数据对所提出的网络训练算法和融合结构进行了实验研究.结果表明:改进后的对向传播网络识别性能优于传统的对向传播网络(CPN),融合后的目标识别率较单传感器明显提高.最后,将该方法应用于前视红外(FLIR)和可见光摄像机目标跟踪系统对算法和融合结构进行验证,结果表明文中提出的方法是可行的.

Modified Counter Propagation Network and Its Application to Multi-Sensor Target Recognition

In view of the features of multi-sensor data fusion and target recognition, a modified counter propagation neural network (MCPN) is proposed. And incorporated with Dempster-Shafer (D-S) evidence reasoning, the data fusion at decision level is achieved for target recognition. The proposed algorithm for training network and the fusion architecture are studied using simulated data. The results show that the recognition performances of modified network outperform the general counter propagation network (CPN) and the correct recognition rate of the fusion system is significantly increased compared with single sensor. At the same time, the ability of system fault tolerance is improved and the uncertainty is decreased. Finally, to illustrate the effectiveness, MCPN algorithm and the fusion architecture for target recognition are applied to a target tracing system of FLIR and TV camera. The experimental results indicate that the approach in the paper is workable.