Abnormal Detection of Substation environment based on Improved YOLOv3

In order to satisfy the requirements of power operation scene safety monitoring and lessen the human accidents, a novel framework of video surveillance for abnormal detection in substations based on artificial technology was presented. Firstly, a multi-classes detector based on improved YOLOv3 was proposed to realize the detection and localization of multiple targets. Then, types of violation and defect were discriminated by secondary analysis for the workers or equipments based on detection results. Substation safety monitoring system was developed to implement environment abnormal detection, and it improved the automation and intelligentization level of unmanned substation. The experimental results demonstrate that the mAP(Mean Average Precision) reaches 0.85, which is slightly improved compared with YOLOv3. Nevertheless, the proposed algorithm is capable of real-time detection as faster than 15 frames per second (fps).