Detecting hot spots on solar cells in mountainous photovoltaic power plants is challenging due to high false and missed detection rates. This paper proposes S-YOLO, a lightweight model based on YOLOv8, for efficient hot spot detection. The model introduces a Spatial Confusion Backbone to reduce parameters and computational complexity while enhancing feature extraction with attention mechanisms. A Multi-scale Feature Pyramid Network (MSBFPN) improves small-target detection, and the SIoU loss function replaces CIoU for faster convergence and precise regression. Compared to YOLOv8, S-YOLO reduces weights by 41.3%, parameters by 21.2%, and increases accuracy by 2.3%. Its lightweight design and high accuracy make it ideal for drone-based inspections in complex environments.