Abstract: Conveyor belt deviation is one of the common faults of belt conveyor. At present, manual periodic inspection method is widely used to detect faults. It has many shortcomings, such as high danger, low efficiency, high labor intensity of the inspectors, lack of real-time detection function. Meanwhile, when the faults are found manually, serious losses are often caused. Aiming at the above problems, this paper studied the on-line detection technology of conveyor belt deviation fault based on image processing technology, an improved Canny operator edge detection method based on guided filter and OTSU is proposed, then Hough transform is used to extract and analyze the edge straight line features. Finally, the on-line detection and classification fault alarm of belt deviation are realized. The system has the advantages of non-contact, high detection accuracy, good real-time performance, and good application effect. It promotes the continuous and efficient operation of belt conveyor and lays the foundation for realizing unattended of belt conveyor in coal preparation plant.