Research on Fine-Grained LED Screen Defect Detection Method Based on Improved Mask R-CNN
In the production process of LED screens,there are often delicate defects such as surface scratches,dirt spots,and cracks.In the case of limited data sets,it is necessary to explore a detection method that can improve the accuracy and speed of detection in LED screens'surface defects while meeting the requirements of actual factory inspections.Based on this,a liquid crystal display defect detection method based on improved Mask R-CNN and transfer learning is proposed.This method solves the problem of traditional image defect detection methods being unable to effectively utilize feature information for high-precision detection and misdetection.The K-means algorithm is designed to cluster the label frames of defective LED screens to obtain suitable aspect ratios and numbers of anchor frames.The improved Mask R-CNN model is applied to extract features from the defective regions of the data set in this paper and trained for classification.To accelerate convergence of the model,the transfer learning approach is employed by pre-training the DAGM 2007 dataset on the improved model and trans-ferring the optimal weights to the liquid crystal display defect detection task in this paper.Comparative experiments with ma-jor object detection methods show that the proposed method which combines transfer learning and improved Mask R-CNN,can accurately identify complex background and minor defect flaws with limited samples.The defects recognition rate on the test set reaches 95.25％,achieving a 5％overall improvement compared to the previous method.
万方数据
维普
