Research on Infrared Image Segmentation of Substation Arrester Based on DeepLabv3+

Fault diagnosis technology based on infrared imagery of electrical equipment is widely used in substations. Segmenting the infrared image of the target equipment from the background can help to narrow the analysis scope to the intended equipment and significantly enhance the accuracy and efficiency of fault diagnosis procedures. However, the task of identifying and segmenting the target equipment within infrared images poses formidable challenges due to inherent characteristics such as low grey scale distribution, low signal-to-noise ratio and low contrast. This paper takes the arrester as the object, and infrared image segmentation method is studied. To amplify the signal-to-noise ratio, a histogram equalization approach is adopted. Additionally, a semantic segmentation model, trained on the DeepLabv3+ network, is devised. The outcome of this research demonstrates an accuracy rate of 98.18% for the endorsed methodology. The proposed method can provide reference for the temperature features extraction and thermal faults diagnosis of electrical equipment in in substation.