Abstract: The surface of mine conveyor belt is easy to be damaged under very bad conditions. If the belt surface damage is not found and repaired in time,it would expand into a destructive rip which could lead to an accident. Currently,the belt surface on carrying side is usually inspected by worker with his eyes under no load conditions in frequent starts and stops. It is difficult to ensure the measurement precision and efficiency. The belt boundary is detected through a combination of polynomial fitting and edge-preserving denoising method,which is used to recognize surface breakage with projection method. According to the same column perceived consistency of light intensity in the image sampled by linear array camera,the segmentation model is established to suppress the uneven noise on the row orientation. This approach has better segmentation effect and less time-consuming,which is convenience to process and recognize the image. It presents an effective mean to other applications of linear array camera. The carrying side and noncarrying side of the conveyor belt are monitored by a computer vision system in real time. If there is a serious damage, the emergency alarm is sounded timely so as to repair belt promptly. It provides an effective technical way to improve the detection quality and efficiency,reduce mechanical wear,and increase production efficiency.