Research on screening of abnormal exposure dropper based on random forest

In recent years, the rapid development of the high-speed railway, and video inspection are important means of railway safety detection. However, in the process of video shooting, there are overexposure and underexposure, which lead to the wrong classification of dropper fault. It is urgent to solve this problem to improve the accuracy of fault detection. Therefore, an image filtering method is proposed, which can determine whether there is abnormal exposure dropper image by calculating the exposure and the maximum of the minimum gray value of the row. Then the random forest method is used to classify the collected sample data and remove the abnormal exposure dropper image, to reduce the fault detection error rate. Through the field data experiment, the classification accuracy reaches 96.5%.