The Identification and Counting of Fault Lights Based on Deep Learning

When the fault lights were detected, the effect of identification and counting accuracy are affected by the complexity and subjectivity of equipment and man-made. In this research, the deep learning object detection model is applied to this field for the first time. According to the characteristics of fault lights, the Faster R-CNN model is improved in this study: The residual network module (ResNet) with deeper network depth is selected to replace the VGG16 network module in the original model, and the embedded feature pyramid network (FPN) is used to extract more rich and robust features. Soft-NMS algorithm is used to reduce the missed detection of intensive fault lights. The results show that the improved Faster R-CNN model has an accurancy of 91.1%, which is 10.6% higher than that of the unmodified Faster R-CNN model. The improved Faster R-CNN model can be used to detect fault lights.