改进小波序贯极限学习机的光电经纬仪空间配准算法研究

Research on modified wavelet online sequential extreme learning machine in space registration for photoelectric theodolite

针对光电经纬仪数据融合系统中的空间配准问题,提出复合函数小波神经网络序贯极限学习机光电经纬仪空间配准算法。该算法将小波理论引入到极限学习机中,利用小波函数和任意分段连续非线性函数构造极限学习机隐层节点激励函数,小波函数的伸缩因子和平移因子根据输入数据范围进行初始化,并结合极限学习机在线学习方法进行训练。实验结果表明:改进小波序贯极限学习机的光电经纬仪空间配准算法可以使光电经纬仪的测量精度提高到3″以内,与标准极限学习机空间配准算法相比,该算法能够实现在线增量式快速学习,具有更好的泛化性能。

An algorithm using composite functions and wavelet neural networks( WNN) in online sequential extreme learning machines( OS-ELM) was proposed to solve the problem in the space registration of photoelectric theodolite data fusion system. The wavelet theory was introduced to extreme learning machines and the wavelet function and bounded non-constant piecewise continuous function were used to build an hidden-node excitation function for extreme learning machine. The contraction-expansion and shift factors of the wavelet function were initiated with the input data range and it was trained in combination with the online learning methods of extreme learning machine. Experimental results show that this algorithm can improve the measurement accuracy of photoelectric theodolite to within 3″ and has fast online learning speed and good generalization compared with standard space registration algorithms.