Research On Ceramic Tile Surface Defect Detection Algorithm Based On Deep Learning

Ceramic tile surface defect detection is an important part of ceramic tile quality management.At present,it mainly relies on manual work,but manual detection has the problems of high cost and low efficiency.The use of deep learning methods for ceramic tile surface defect detection has great potential,but there are not many related studies at present.The detection performance of existing object detection algorithms for ceramic tile surface defects still needs to be improved.At the same time,the data volume of the ceramic tile surface defects detection datasets used is small and not balanced.These problems all affect the detection performance.In order to solve the above problems,this thesis applies the deep learning method to the detection of ceramic tile surface defects.First,the improved algorithm of ceramic tile surface defect data augmentation based on Deep Convolution Generative Adversarial Networks(DCGAN)is studied.Then,according to the difficulties in the task of ceramic tile surface defect detection,a new ceramic tile surface defect detection algorithm is improved.First,in view of the limited object of defects marked in ceramic tile images,the amount of data is small and the categories are not balanced,this thesis improves a ceramic tile surface defect data augmentation algorithm based on the DCGAN algorithm.After analyzing and preprocessing the dataset,this thesis incorporates the Convolutional Block Attention Module(CBAM)attention mechanism into the generator of DCGAN to improve the generation quality of ceramic tile surface defect images.Afterwards,two commonly used object detection algorithms,Faster R-CNN and Cascade R-CNN,were trained on the augmented dataset of the improved algorithm to detect ceramic tile surface defects.The Mean Average Precision(m AP)of the two algorithms increased by 5.1 and 4.4% on the augmented dataset.Second,in view of the problem that the existing object detection algorithm has poor detection effect on some defects with extremely small size and inconspicuous detail features,some improvements based on the Cascade R-CNN algorithm is proposed by this thesis in terms of backbone network,feature fusion strategy,loss function,etc.First,the Residual Network(Res Net)backbone network is selected to enhance the feature extraction ability of the algorithm,and the recursive feature pyramid structure is used to fuse multi-scale feature information to enhance the detection effect of the algorithm on small ceramic tile defects.After that,in order to make the algorithm focus on difficult samples during training,the Focal Loss function is used.Finally,in the training strategy,the pre-training model and multi-scale training method are used to further improve the performance of the algorithm.Experiments confirm that the improved detection algorithm in this thesis has an 8.6% improvement in m AP compared to the original Cascade R-CNN algorithm.Finally,based on the improved detection algorithm in this thesis,a detection system for ceramic tile surface defects is designed and implemented.The system integrates data preprocessing and improved algorithm end-to-end,mainly realizes the functions of ceramic tile surface defect detection and detection result statistics,simplifies the user’s operation and shows the detection process of the algorithm.