YOLO-Slim:A Lightweight Small Object Detection Algorithm for Drone Images

Object detection tasks on drones face the challenges such as extended target distance, small size of objects, and limited computing resources. However, due to the high computational overhead, existing small object detectors are not well-suited for deployment on mobile devices. To address the key issue, the lightweight detector YOLO-Slim is proposed. Firstly, to enhance the model's feature extraction capability while reducing computational cost, the efficient and lightweight dynamic convolution module Efficient Dynamic Conv (EDC) is designed. Furthermore, to address the feature weakening problem, the feature enhancement module C3k2-SimAM with a three-dimensional attention mechanism is introduced. Compared to YOLOv11n, YOLO-Slim achieves a 2.1% improvement in [email protected], a 1.3% increase in [email protected], and a 29.2% reduction in computational overhead (FLOPs) on the VisDrone2019 dataset. The experimental results demonstrate that YOLO-Slim achieves an optimal balance between detection performance and computational efficiency, providing new insights for optimizing real-time small object detection algorithms on drones.