【作者】 徐宝昌； 陈哲；

【Author】 XU Bao-chang, CHEN Zhe School of Automation Science and Electrical Engineering, Beijing University of Aeronautics and Astronautics, Beijing 100074

【机构】 北京航空航天大学自动化科学与电气工程学院；

【摘要】 本文基于神经网络理论给出了一种二层前向神经网络递推匹配算法。神经网络的权值与匹配定位参数相对应,通过训练求出网络权值,从而确定匹配定位参数。由于网络的学习算法采用了递推计算,大大减小了权值计算的存储量,避免陷入局部极小点;同时其为二阶算法, 收敛速度较快,精度高。仿真结果表明其精度和匹配速度均优于最小二乘匹配算法。更多还原

【Abstract】 Based on the neural network theory, a two layers forward neural network based recursive scene matching algorithm is proposed in this paper. The weight values of the neural network are correspondence with the positioning parameters. The weight values of the network are calculated by treating. Then the positioning parameters are got. By using recursive calculation, the storage quantity of the learning algorithm is reduced greatly and it is avoided to trap into local minimum. The new matching algorithm is a second-order learning algorithm, so it converges rapidly and has high precision. The simulation result shows that the precision and the matching rate of the new matching algorithm are better than that of the least square matching algorithm.更多还原