Research on Intelligent Fault Identification Method Based on UAV Power Inspection

To build an intelligent fault detection system based on UAV patrol and improve the efficiency of technicians, a novel multilevel and multi-scale feature fusion network (M2F2N) is developed. Specifically, to assign more weights to important features, such as high-frequency information containing image edge details, we use two parallel channel attention and spatial attention branches in dual-attention feature extraction block (DAFEB) to promote information interaction and learn interdependency across different channels. In addition, a multi-scale future fusion block (MSFFB) is designed based on stacking several DAFEBs to extract more effective features. Finally, the features extracted from different hierarchies are fused and aggregated via concatenation. Adequate experiments validate that M2F2N exceeds the compared models, i.e., SRCNN FSRCNN, VDSR, DRRN, and IDN, in terms of quantitative and qualitative results.