基于改进PCNN的轴尖表面缺陷检测

Modified-PCNN based detection of gyroscope pivot surface defects

脉冲耦合神经网络(PCNN,Pulse Coupled Neural Network)与传统神经网络不同,不经过训练即可用于图像处理.针对PCNN模型中结构参数较多,且需要人工反复试验进行设置的困难,改进模型结构,简化了馈送输入和连接输入,减少了待定参数;根据邻域灰度动态地计算内部连接系数,由邻域的欧氏距离计算权值矩阵,再由图像的灰度特征计算动态阈值.将改进的PCNN用于陀螺轴尖表面缺陷图像的分割,用基于完整性与正确性指标的缓冲区匹配方法评价所提方法、最大熵法及Canny方法.针对不同缺陷图像的实验表明:所提算法的完整性与正确性都高于0.9,证明所提方法更有效.

Pulse coupled neural networks（PCNN） differs from traditional neural networks.PCNN can be applied to image processing without training.There are many structure parameters in PCNN model,and it is difficult to determine these parameters by manually trying.The model structure was improved by simplifying feedback input and connection input,and thus the number of the parameters was reduced.The inside connection coefficient was calculated dynamically based on neighborhood grayscale.The weight matrix was obtained by utilizing neighborhood Euclidean distance.The dynamic threshold was calculated from image grayscale character.The modified PCNN was used to segment several gyroscope pivot surface defects images.Based on the buffer region matching method,the completeness and correctness measures were used to compare the presented method,maximum entropy and Canny segmentation,and the results showed the two measures were not less than 0.9,which means that the proposed method is more effective.