Research on the Detection Method of Small Fastener Defects in Aerial Transmission Lines Based on YOLOV10N

To address the challenge of low accuracy in detecting small targets during unmanned aerial vehicle (UAV) aerial transmission line inspection, we propose an enhanced target detection algorithm derived from the You Only Look Once Version 10 Nano (yolov10n) model. Here's a revised version of the algorithm description: Firstly, to enhance the network's adaptability to geometric variations, we introduce the Deformable Convolutional Network (DCN) to replace certain standard convolutions in the CSPDarknet53 backbone and 2-Stage Feature Pyramid Network (C2F). This modification allows the network to better capture deformations in small targets. Secondly, we embed the Shallow Feature Enhancement Module (SFEM) to merge the shallow spatial details from the input features with the deep semantic information from the network's neck section. This integration enriches the representation of small targets. Additionally, we utilize the Global Context Block (GC block) to recalibrate the fused features, effectively suppressing background noise and highlighting the target information. Next, to further boost the model's ability to detect targets at various scales, we incorporate the Advanced Spatial Pyramid Pooling Fast (ASPPF) module. This module enhances the multi-scale feature representation, thereby reducing the missed detection rate, especially for small targets. Finally, we integrate the Similarity-based Attention Module (SimAM) to strengthen the model's feature extraction capabilities. By focusing on salient regions, SimAM helps the model identify small targets more precisely. In summary, this enhanced yolov10n-based algorithm provides robust technical foundations for UAVs to conduct intelligent inspections and fault detection on high-voltage transmission lines, particularly in scenarios involving small targets.