Small object detection algorithm for remote sensing images based on multiscale feature fusion

The one-stage object detection method based on deep learning has been widely used in remote sensing image object detection because of its high efficiency. Although it has achieved good results in the field of remote sensing image object detection, it is aimed at the problems of poor recognition and insufficient accuracy of small objects in remote sensing images or complex scenes. Firstly, the FreqFusion module and BiFPN module based on frequency feature fusion were designed to replace the original neck structure, which effectively enhanced the characterization ability of multi-scale features. Secondly, the CAA cross-dimensional attention mechanism was introduced to reconstruct the detection head to realize the adaptive feature calibration of the channel and spatial dimensions. Finally, the mean average precision (mAP) of FBC-YOLO11 on the DIOR dataset is 83.5%, which is 4% higher than that of the YOLO11 algorithm, and the recall rate is increased by 2.6% compared with the YOLO11 algorithm. The effectiveness of the method in remote sensing small object detection is proved.