基于ResNet-18的磁瓦缺陷检测

Magnetic Tile Defect Detection Based on ResNet-18

由于磁瓦在实际生产中会产生多种类型的缺陷,针对磁瓦生产的表面检测高质量要求,提出一种基于ResNet-18的磁瓦缺陷检测方法。首先通过构造磁瓦缺陷的不同类型数据集,再将数据集通过图像增强等方式扩充数据集进行预处理;引用ResNet-18作为训练所用的网络,设置好网络模型参数;最后通过网络训练完成对磁瓦缺陷的分类及标注。实验结果表明:扩充后数据集对比于原始数据集和其他方法,采用的方法中模型测试准确率更高,搭建的ResNet-18网络具有更好的鲁棒性和泛化性,证明了可以满足磁瓦缺陷检测中的可能性。

Due to the various types of defects in the actual production of magnetic tiles,a method of magnetic tile defect detection based on ResNet-18 was proposed to meet the requirements of high quality surface detection in the production of magnetic tiles.Firstly,different types of data sets of magnetic tile defects are constructed,and then the data sets are augmented by image enhancement and other methods for preprocessing.ResNet-18 was used as the training network,and the parameters of the network model were set.Finally,the classification and labeling of magnetic tile defects are completed by network training.The experimental results show that compared with the original data set and other methods,the model testing accuracy of the method adopted in this paper is higher,and the ResNet-18 network built has better robustness and generalization,which proves that it can meet the possibility of magnetic tile defect detection.