Search

Faster RCNN is a popular and very efficient object detection algorithm, it has good performance in image recognition in various fields. In this paper, we applied the Faster RCNN algorithm to the equipment recognition and status detection tasks in the electric power room, then proposed a complete technical implementation scheme. First of all, after site investigation and shooting, we collected a total of 5600 pictures in 100 categories of electrical equipment. Then, we achieved the labeled pictures for all categories by employing an open source software named Labelimg. In the next step, the annotated data was randomly shuffled, 70% of which was used as the training set for model training, 10% of which was used as the validation set for model tuning, and the remaining data was used as the test set for model testing. We get less than 0.01% loss in training step and get 91.3% mAP for all test pictures in testing step. Besides, the test for video received a great performance. The well performance on the test pictures and videos showed the effectiveness and implementability of our technical solutions.

This paper proposed a method for detecting the panel of equipment in the power room based on deep learning technology. For a real-time image, we can achieve the purpose of equipment panel detection and segmentation, which facilitate the task of equipment monitoring. We use the Mask R-CNN algorithm to detect how many target devices there are in the picture and to find the category corresponding to each device. In each target device, the foreground and background are distinguished at the pixel level. The method is described as follows. First of all, we collected a total of 4,500 pictures in 60 categories of power room equipment for our experiments. Next, we labeled all the pictures using Labelimg, which is an open source software for data annotation, the marked pictures were preprocessed as the data set to train and test the proposed model. The well performance on the test set shows that our proposed method has a certain implementability and effectiveness.