Electrical Violation Behavior Recognition of Substation Staff Based on Deep Learning

LI-XIA SUN, JING-TAO BAI

Abstract


In order to reduce the incidence of electrical violation during the operation of substation staff, this paper combines transfer learning, residual network and k-nearest neighbor (KNN) classifier to establish ResNet-18-KNN model. ResNet-18-KNN model is suitable for the identification of electrical violation of substation staff. Firstly, the ResNet-18 pre-trained model of the residual network is obtained through publicly available resources on the Internet. Then, the data of image is enhanced and expanded using image dehazing algorithm, image contrast enhancement, image denoising and other methods to solve the problem of insufficient and unbalanced data. After that, the ResNet-18 network is trained using extended data and features are extracted using the trained ResNet-18 network. Finally, the extracted features are used to train KNN. Experimental results show that the proposed method has the best performance compared with the ResNet-18 network, AlexNet network, SqueezeNet network, GoogleNet network and other models. ResNet-18-KNN can give an accurate judgment on whether the substation staff have electrical violation.

Keywords


Deep Learning, Transfer Learning, Image Enhancement, Electrical Violation, Behavior RecognitionText


DOI
10.12783/dteees/peems2019/34028

Full Text:

PDF

Refbacks

  • There are currently no refbacks.