Abstract: A kind of image analysis algorithm combining on-line analysis and off-line analysis was proposed: SVM was adopted to detect conjunctions images, and saliency map was used to check image of damaged belt during on-line analysis phase; Image processing methods such as thinning and regional analysis were used to detect joints twitching and classify a variety of injuries during off-line analysis phase. The conveyer belt detection system could analyze X image of conveyor belt in real time and detect damage accurately via applying these algorithms. The experiments and field deployments verify effectiveness of the proposed image analysis algorithm.