Research on Fault Detection of Electrical Equipment Based on Infrared Image

摘要

Due to the problems such as interference points in infrared images and the lack of obvious edge feature information, the detection effect of faults in traditional electrical equipment is low. To this end, a model combining the ResNet algorithm with the improved watershed algorithm is proposed to extract the abnormal areas and fault types of electrical equipment. First of all, the infrared image is initially segmented by OTSU algorithm, and the outline of electrical equipment is extracted. Secondly, the RGB of the original infrared image is converted to HSV color space, the hot spot information is extracted, then the hot spot is located using the ResNet-34 network and transfer learning method, and finally, the equipment is segmented by the improved watershed algorithm in the test phase, and the abnormal area and fault type of the fault device are finally extracted. Experiments show that the method greatly improves the accuracy of detection, the model test accuracy of 99.5%, compared with other detection algorithms can effectively save training time, and identify fault types more accurately.