An Improved YOLOv8 Algorithm for Multi-scale Remote Sensing Image Object Detection

To address the low accuracy issue that may be brought on by the multi-scale images in remote sensing image recognition, an enhanced model based on the YOLOv8 model is suggested. To better handle multi-scale targets, the model makes use of the suggested Zoomcat and ScalSeq modules. Additionally, the model's feature extraction capabilities are improved by introducing both the ASF and EMA attention processes. The dataset's ablation and comparison experiment findings show that, in comparison to YOLOv8m, the enhanced model has a 2.2% increase in precision, a 1.8% improvement in recall, a 2.4% increase in mAP, and a substantial 2.8% increase in mAP50:95.