基于Gabor滤波器组与BP神经网络的帘子布疵点检测研究与实现

Method of Fabric Defects Detection Based on Gabor Filters and BP Neural Network

采用Gabor滤波器组对帘子布疵点图像纹理进行滤波,对滤波后的模值图像使用最大熵阈值分割,提取疵点轮廓的长、宽、长宽比、面积等特征值。将上述特征值归一化后分为两类:一类作为训练样本输入BP神经网络,对网络进行训练学习,网络计算结果收敛后结束训练;另一类作为测试样本对训练好的网络进行疵点识别。实验证明,该方法可以快速地检测疵点,利用训练的BP神经网络实现疵点分类,识别率达94%。

In this paper, Gabor filters on cord fabric defects texture filtering and take an amplitude as out- put image, uses maximum entropy to segment the amplitude image, extract length, width, aspect ratio, size as a characteristics, then normalized the characteristics and divided into two categories, one category as the train- ing sample input training and learning of BP neural network- when the network converged end of the training, the other as a test sample to the trained network for defect detection. Experiments show that this method can quickly detect defects, the trained BP neural network to achieve a good defect classification, recognition rate of 94%.