Longitudinal tear detection of conveyor belt under uneven light based on Haar-AdaBoost and Cascade algorithm

Abstract As the light of the belt conveyor changes drastically and unevenly, it will cause the longitudinal tear detection rate of the traditional methods drop sharply. To improve the situation, a novel method, which replaces the traditional geometric features with the Haar features, is proposed. First, the Haar feature was employed to train weak classifiers. Then, the AdaBoost algorithm is utilized to upgrade the weak classifiers to strong classifiers. Finally, the Cascade algorithm is introduced to combine strong classifiers in series into a cascade classifier that can reduce the training and processing time. The experiment following the above proposed method has shown that the recall, accuracy, and precision of tear detection under uneven light almost approaches the level under the uniform light (less than 3% difference), which indicates that our method is more accurate and robust than the existing methods in real-time belt tear detection.