DDM-YOLOv8s for Small Object Detection in Remote Sensing Images

In order to solve the problem of low accuracy of remote sensing target detection caused by insufficient feature representation, complex background and diversified scales in remote sensing image target detection task, a remote sensing image target detection algorithm based on DDM-YOLOv8s (DySample DASI-MODSConvYOLOv8s) model was proposed. First, a super lightweight and efficient dynamic upper sampler DySample is used to improve the image resolution of the algorithm without adding additional burden. Secondly, the new C2f_DASI module is constructed by using the dimensional awareness selective integration module DASI and C2f, and the adaptive channel selection and fusion are realized in the feature fusion module, which effectively solves the complex background problem in remote sensing images; Finally, the head part is redesigned by Maintaining the Original information - Depth Separable Convolution MODSConv, which maintains as much feature layer information as possible before prediction, so that the extracted features contain more correlation information, greatly improving the detection accuracy and speed. The experimental results on RSOD datasets show that the test mAP of the improved YOLOv8s algorithm is 94.5%, which proves that this method can effectively improve the detection accuracy of small targets in remote sensing images.