RS-YOLOv10: Enhancing YOLOv10 for Accurate Small-Object Detection

The capability to detect objects in remote sensing images has proven to deliver exceptional results spanning multiple fields of application, setting it apart from other techniques. However, with limited resources, it faces numerous challenges related to noise, poor information quality, and particularly the detection of small objects. To overcome these challenges, we introduce RS-YOLOv10, a model built on YOLOv10, designed to reduce resource consumption and improve accuracy for remote sensing applications. First, to enhance the backbone network, we apply the Efficient Attention Mechanism, which helps the model prioritize key information within the feature map, the model boosts detection efficiency. In addition, we propose the Weighted Focused Feature Pyramid Network (WFFPN), which effectively minimizes the network while enhancing the efficiency and performance of multi-scale feature fusion. When evaluated on the VisDrone2019 dataset, RS-YOLOv10 demonstrated an approximate 10% decrease in the number of parameters relative to the baseline model and significantly improves metrics, with precision increasing by 2.5%, mAP50 increasing by 1.4%, and mAP95 increasing by 1.1%. Relative to other cutting-edge models, our model shows superior performance, particularly for smaller object detection.