The invention provides a machine vision belt tearing detecting and protecting device. A line laser device projects a consecutive laser straight line to the bottom surface of a belt, and CCD (Charge Coupled Device) cameras are used for respectively acquiring images from two sides. The images acquired in real time are transmitted to a tearing detection control computer; the intensity, curvature and other characteristics of the laser streak are extracted to be used for training of a neural network algorithm; and after learning, the network can quickly and accurately detect a tearing accident. In order to acquire clear and stable images at all time, a seal cover is used for sealing a part of the belt conveyer, and an LED light source is used for supplementing light. The double CCD fault-tolerance mechanism avoids noise interference on images due to bug dust, coal slag and other factors, and the structured light technique ensures that the complexity for image processing is reduced from two dimensions to one dimension. The invention has the advantages of low cost, high precision, simple structure and strong practicality, and effectively narrows the belt tearing range by discovering the tearing accident in time.