Lightweight target detection algorithm based on feature fusion

A streamlined object detection approach is introduced, leveraging advanced feature fusion techniques to address the constraints of limited computational resources in unmanned aerial vehicles (UAVs). The proposed method integrates novel Feature-1 and Feature-2 modules to facilitate semantic feature fusion. By incorporating a pulse residual structure and a novel activation function, the overall model complexity and parameter count are significantly reduced. Empirical results demonstrate a 62.6% decrease in parameters and a 46.8% reduction in computational complexity compared to YOLOv7, and similar improvements over YOLOv9c. This advancement ensures efficient target recognition on low-power platforms, enhancing UAV detection speed and performance.