Mid-yolo: an efficient object detection network for remote sensing images

Object detection in remote sensing imagery presents significant challenges due to large variations in object scales, complex and cluttered backgrounds, and sparse semantic content. The highly uneven distribution of object sizes—ranging from numerous small targets to medium and large ones—combined with substantial background interference, greatly increases the complexity of the detection task. Traditional methods often struggle to maintain robustness while effectively representing objects across multiple scales in such scenarios. To address these challenges, this paper introduces a novel detection framework, MID-YOLO, which is built upon the YOLOv11 architecture. The proposed model incorporates two key components: an enhanced Spatial Pyramid Pooling module with Multi-Scale Channel-Spatial Attention (SPPF_MSCA) and an Inception-Dilated Attention Module (IDAM). SPPF_MSCA integrates multi-scale pooling with channel and spatial attention mechanisms to effectively highlight salient target features, while IDAM combines multi-scale deep convolutions with dilated multi-head self-attention to jointly model fine-grained local details and long-range contextual dependencies. Experimental results on the public DOTA1.5 dataset demonstrate that MID-YOLO achieves significantly improved detection performance compared to existing state-of-the-art methods.