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,逼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,暗伤检测合格率为100%,检测效率128个/min。 Aiming at the problem that the dynamic nearest neighbor clustering algorithm is not suitable when the center point is selected improperly and the hidden layer nodes are few, the improved dynamic nearest neighbor clustering algorithm is proposed to construct RBF neural network (IDARBF neural network) Output characteristics to effectively overcome the problems existing in the original algorithm. Experiments show that the IDARBF neural network has better nonlinear correction ability. After using the improved dynamic nearest-neighbor clustering algorithm, the sensor performance is greatly improved, and the accuracy is more High, dark injury detection pass rate of 100%, detection efficiency of 128 / min.