ISRYOLO: Improving Small Object Detection in Aerial Imagery via Super-Resolution and Attention Fusion

Detecting objects in high-resolution drone and satellite images is vital for disaster response, urban planning, and environmental monitoring. Small object detection, the task of finding tiny targets that cover only a few pixels (e.g., cars, people, or rooftops), is tough because these objects lack detailed features and often blend into cluttered backgrounds. ISRYOLO is a super-resolution assisted object detection framework that integrates attention mechanisms (CBAM), pixel shuffle upsampling, ConvNeXtBlock, and Atrous Spatial Pyramid Pooling (ASPP) into the detection pipeline. The model enhances spatial feature representation and improves accuracy, particularly for small objects, while maintaining computational efficiency. The results in the VEDAI dataset demonstrate that ISRYOLO outperforms the state-of-the-art model in terms of mean Average Precision (mAP) and recall, making it a promising solution for real-time applications.