Abstract: In view of the problem that the traditional machine vision based conveyor belt tear detection algorithm requires high computing power and high power consumption AI module, and the intrinsic safety power supply can not meet its electrical needs, a longitudinal tear detection algorithm of conveyor belt based on one-dimensional dual attention network(DANet-1D) was proposed. The image of conveyor belt surface line laser was collected by industrial camera. The laser stripe feature filter was designed to extract stripe features. A tear detection algorithm based on DANet-1D was designed to reduce the dimensionality of the torn two-dimensional image data. The detection of neural network operated in one-dimensional space, run faster and supported high-resolution images. An intrinsic safe conveyor belt tear detection device was developed and verified. The results show that the accuracy P of the algorithm is 92.54%, and the recall rate R is 91.78%. The average detection time per frame is 12.40 ms. The industrial test successfully detected simulated tearing of the conveyor belt, providing a new detection scheme for longitudinal tear of the conveyor belt.