Considering the unsatisfying approximation result of dynamic nearest neighbor clustering algorithm when center is selected improper and hidden layer nodes are fewer, an approach that improved dynamic nearest neighbor clustering algorithm is used to construct RBF neural network (IDARBF neural networks) is raised, which calibrates the sensors output characteristic and effectively overcomes the problems in the original algorithm. An experimental result shows that IDARBF neural network has a better ability of non-linear correction, and the performance of sensor significantly is improved after the improved dynamic nearest neighbor clustering algorithm is used. Invisible injury rate is 100% tested and the detection efficiency of 128 / minute.%针对动态最近邻聚类算法因中心点选取不当以及隐含层节点较少时,通近效果不理想的问题,提出运用改进的动态最近邻聚类算法构造RBF神经网络( IDARBF神经网络),对传感器输出特性进行校正,有效地克服了原算法存在的问题;实验表明,IDARBF神经网络具有更好的非线性校正能力,运用改进的动态最近邻聚类算法处理后,传感器性能大幅度改善,精度更高,暗伤检侧合格率为100,检测效率128个/min.
展开▼