A lightweight small-object detection algorithm for remote sensing images based on 4SDC-YOLOv8

In the field of object detection, remote sensing image detection has become a critical research focus. However, existing small-object detection algorithms for remote sensing images often fail to strike an effective balance between detection accuracy and computational complexity, thereby limiting their application on mobile devices. To address this issue, this paper proposes a lightweight small-object detection algorithm for remote sensing images, named 4SDC-YOLOv8, built upon YOLOv8. Firstly, we incorporate the SimSPPF spatial pyramid pooling structure, using pooling kernels of different sizes across multiple scales to improve the detector’s performance. Secondly, we design a lightweight neck network that leverages the dynamic upsampling operator DySample and the partial convolution Partial Conv to introduce the new CSPPC architecture, reducing redundant computations and enhancing both real-time performance and model compactness. Finally, we add a dedicated small-object detection layer, enabling the feature map to capture rich semantic and spatial information, with four detection heads supporting multi-scale small-object detection and improving detection capability for small targets. Experimental results on the DIOR dataset show that, compared with YOLOv8, 4SDC-YOLOv8 significantly reduces model size and computational load by 24.6% and 24%, respectively, achieves an accuracy of up to 90%, and maintains a respectable real-time performance at 145 FPS. Confirm that 4SDC-YOLOv8 effectively balances detection accuracy and computational complexity, thereby offering a promising reference for implementing small-object remote sensing detection on mobile devices.