Abstract: To deal with the tearing or rupture problem occurred during the movement process of the conveyor belt in the power plant, a detection device based on machine vision and image recognition was designed according to the working environment and testing requirements. The grid line laser, used as structural light, was irradiated under the conveyor belt surface, and the real-time images of the belt were acquired through the high resolution industrial camera. The belt tearing was identified by feature of gird line in images, which cost only 40 ms, the recognition accuracy rate is as high as 96%, and the false alarm rate is only 2%. Compared to traditional detection methods, the whole time is shortened greatly, and the conveyer belt does not need stop, which ensures the safe and efficient production of the conveyor belt, reduces the accident rate effectively.