Woven Fabric Defect Detection Based on the Cascade Classifier

In this paper, the cascade classifier obtained by combining the adaboost algorithm and the haar-like feature is used to detect the defects of five types. The average F1 score is above 0.90, and the average false positive rate is 3%. This method can quickly remove the background area of the image and focus on the object area, and the average detection time is about 0.07s, which achieves good real-time performance. After histogram equalization and repeated training on difficult samples, the detection efficiency is improved, the F1 score reaches 0.96, and the false positive rate drops to 2%. The cascade classifier has a good adaptability to different kinds of textiles which can be used in fabric defect detection.