Deep Skipping Fusion for Insulator Fault Detection

This work proposes a deep skipping fusion (DSF) method for insulator fault detection. This work consists of three main steps. First, multi-layer deep features are explored by using a convolutional neural network. Then, multi-layer deep features are fused, and a covariance matrix is calculated for the fused features, which can offer the complementary deep information from various layers. Last, the obtained covariance matrix serves as a feature for classification by a support vector machine. Experimental results demonstrate that the proposed method can obtain an outstanding diagnosis performance with limited training samples.