Abstract: To recognize longitudinal tearing accurately, Self-Organizing feature Maps（SOM）was introduced to detect longitudinal tearing of conveyor belt. After median filtering, the histograms of oriented gradient of the conveyor belt images were extracted as the SOM’s input feature vectors.Then Chi-square distance was adopted to describe similarity of two different feature vectors. SOM detecting model was thus built and training process of the model was described in detail. Finally simulation experiment was done to test the effect of the mode. The result shows that the SOM network works well in recognizing longitudinal tearing of conveyor belt. The research in this paper brodens ways of longitudinal tearing detection.

Key words: conveyor belt, longitudinal tearing, Self-Organizing feature Maps, neural network