A Remote Sensing Detection Method for Floating Objects Along Transmission Channels Based on Improved YOLOv8

Transmission line short circuits caused by floating objects like plastic greenhouses or films under adverse weather pose significant risks to grid safety. Detecting these complexshaped objects remains challenging due to small target sizes, occlusion, and low-contrast backgrounds. To overcome this, we propose EEL-YOLOv8 featuring: (1) a multi-branch deep feature extraction module (FODE) integrating dilated convolution and multi-scale pooling to enhance complex object capture; (2) a feature attention module (FO Attention) combining global pooling and multi-scale dilated convolution to amplify key features and suppress background; (3) an LLIoU loss incorporating geometric constraints to optimize box matching. Experiments on a self-built dataset show EEL-YOLOv8 achieves <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$85.6 \% \text{mAP} {@} 0.5$</tex> (a 5.6 % gain over YOLOv8) and runs at 72 FPS, meeting real-time demands while significantly improving detection in small-target, occluded, and low-contrast scenarios. This work provides an effective intelligent solution for enhancing power transmission safety.