Detection of Insulator Strings in Aerial Photography Based on Lightweight CenterNet
Jiamin Mai, Shenghui Wang, Xiuru Wang · 2023
In order to realize the rapid and accurate detection of transmission line insulator strings in aerial images, an improved detection method based on lightweight CenterNet was proposed. First, the lightweight network EfficientNetB0 was used as the backbone network of CenterNet to reduce the number of parameters of the model; Second, the feature enhancement structure was constructed to make full use of the global information; Then, the weight-based attention feature fusion modules was adopted to replace the common feature fusion mode in the upper sampling process; Finally, based on transfer learning and efficient network training methods, the generalization ability and training efficiency of the model were improved. The experimental results showed that compared with the original model, the average accuracy of the model on the public dataset (CPLID) was 93.9%, which was increased by 4.2 percentage points, and the image prediction speed was raised from 53.3 frames / s to 7.7 frames / s.