Abstract: As a standard transmission part, aluminum tube surface defect detection is necessary to ensure product quality and operation safety. The surface defect detection method based on machine vision for aluminum tubes has replaced the manual detection method and become one of the mainstream detection methods due to its metrics such as high detection accuracy and fast detection speed. However, the imbalanced distribution between defective and background samples leads to the shift of the classifier decision boundary, and a decrease in detection accuracy, which limits its application range. To solve this problem, a surface defect detection method of aluminum tube based on ensemble adaptive undersampling is proposed. Firstly, the support vector data description method is used to identify the overlapping areas in the data distribution. Secondly, the local density relationship is utilized to determine which and how many majority samples are to be removed. Finally, the random space generation technology is used to optimize the sample distribution space and feature space simultaneously. The experimental results show that the Precision of the proposed method is 98.52%, which is superior to other advanced detection methods.

Key words: surface defect detection, support vector data description, adaptive undersampling, random space generation