Application of Mask R-CNN in defect detection of industrial filter cloth

Industrial filter cloth will produce stains and defects in the production process, which will have a certain impact on the sales of filter cloth. Traditional filter cloth manufacturers rely on manual classification of defect types. Due to human subjectivity and visual fatigue, classification errors and omissions will occur. To solve this problem, this paper uses Mask R-CNN convolutional neural network model to classify and identify filter cloth defects. In this paper, 3600 industrial filter cloth images are collected and preprocessed. The images are expanded to 12000 images. The network is trained by combining the improved feature pyramid network (FPN) algorithm. The results show that the combination of resnet101 model as the backbone network has the best application effect.