Abstract: In view of the problem that traditional fault detection of downhole belt conveyor deviation is prone to noise interference, resulting in fault detection results inconsistent with the reality, this paper proposes to combine the line features of ROI edge images for fault detection of downhole belt conveyor deviation. ROI method is used to lock target image edge, build edge detector, design image edge search path, and obtain image edge information. The template matching is carried out around a certain area, and the external square is used as the optimal external contact point. According to the positioning contact pixels, the point set is formed according to the spatial increasing order to avoid noise interference. Experiment results show that the mapping from image to parameter space is realized by calculating the distance between any point of the feature line and the origin according to the principle of ROI edge image feature transformation. At the same time, multi-layer cooperative detection is realized according to MT-CNN method, and the constraint relationship between line coordinates of ROI edge image is combined to realize the detection of conveyor deviation fault. The gray scale map of the method is consistent with the actual map, and there is only a maximum error of 50 mm between the detection result and the actual value, which has accurate detection results, and can provide decision-making basis for the detection of downhole belt conveyor deviation fault.