Remote Sensing Target Detection Algorithm Based on Lightweight YOLOv8s Model

The detection of images through remote sensing is of great importance in areas such as geographic information system, environmental detection, and military reconnaissance. To meet the need for efficient operation on arithmetic-limited devices, this paper presents a lightweight improvement technique based on YOLOv8s, designed to improve the detection performance of arithmetic-limited devices. First, the feature expression capability is enhanced by introducing the C2f-UIB module to replace the C2f module in the original YOLOv8s backbone network, performing feature transformations in low-dimensional space, and combining the extension layer and projection layer to reduce computational redundancy and accelerate feature extraction. The integration of BiFPN module into the neck network has been demonstrated to optimize multi-scale feature fusion capability, thus enhancing the detection of targets of varying sizes. Experimental results demonstrate that the improved model performs admirably. Publicly available optical remote sensing dataset DIOR, with the GFLOPs reduced to 7.4, which is about 74% less than the original model; at the same time, the detection precision (P) is 0.812, and the accuracy only decreases by 2.3%, which fully meets the real-time requirements. The above improvement scheme not only effectively improves the model’s performance but also provides a feasible solution for target detection on arithmetic-limited devices.