基于GRNN-PNN神经网络的印铁缺陷分类方法

Classification Method of Printed Iron Defects Based on GRNN-PNN Neural Network

目的针对在印铁过程中缺陷检测系统存在不同缺陷类型检测精度不高,对于产品整体质量无法实现智能判断的问题,基于GRNN-PNN神经网络,提出一种适用于印铁在线检测的分类算法。方法对平面印刷铁片进行小波变换提取低频信息,在低频信息中进行缺陷定位并对缺陷区域进行标记和分割。通过缺陷面积、周长等评价指数和缺陷形状构建GRNN神经网络,对缺陷进行分类。通过构建PNN神经网络智能化判别整体产品是否属于合格产品。结果GRNN-PNN平均耗时0.69s,达到了厂方对于缺陷在线检测的响应时间要求。GRNN-PNN多分类的准确率为86%,能够对印铁过程中产生的主要缺陷进行分类。二分类的灵敏度为96%,可以准确地判断产品整体的合格性。在5%的椒盐噪声干扰下,准确率为63%,具有良好的鲁棒性。结论该设计能够对印铁缺陷进行精确的分类和智能的判断,GRNN-PNN神经网络可以在印铁过程中进一步提高检测精度,GRNN-PNN神经网络可帮助质检员及时判断生产质量。

During the iron printing process,the defect detection system has a problem that the accuracy of defect detection is not high,and intelligent judgment cannot be achieved for the overall product quality.This paper aims to propose a classification algorithm suitable for on-line detection of printed iron based on GRNN-PNN neural network.Wavelet transform was performed on printed iron sheet to extract low-frequency information.The defects in low-frequency information were located,and the defective areas were marked and segmented.A GRNN neural network was constructed based on defect area,perimeter and other evaluation indexes to classify defects.A PNN neural network was constructed to intelligently determine whether the overall product was a qualified product.The average time of GRNN-PNN was 0.69 s,which met the factory’s requirement on the response time of online defect detection.The accuracy of GRNN-PNN multi-classification was 86%,which can classify the main defects generated during the iron printing process.The sensitivity of the two classifications was 96%,which can accurately judge the overall qualification of the product.Under 5%salt and pepper noise,the accuracy rate was 63%,and GRNN-PNN had good robustness.The design can accurately classify and intelligently judge the defects of printed iron.The GRNN-PNN neural network can further improve the detection accuracy during the iron printing process.The GRNN-PNN neural network can help quality inspectors to judge the production quality in time.