RS-TinyNet: A multi-dimensional attention network for tiny object detection in remote sensing imagery

Detecting tiny objects in remote sensing (RS) imagery has long been a challenging task due to their minimal spatial information, weak feature representations, and dense distributions across complex backgrounds. Despite numerous efforts devoted, mainstream detectors still underperform in such scenarios. To bridge this gap, we introduce RS-TinyNet, a novel detection framework specifically tailored for detecting tiny objects in various RS scenarios. The core design of RS-TinyNet is a multi-dimensional collaborative attention (MDCA) mechanism, which jointly integrates channel-spatial dependencies and global-local contextual cues to enhance the saliency and discriminability of tiny objects and suppress background interference. Moreover, MDCA is lightweight and flexible, allowing seamless integration into different stages of the detection pipeline for progressive feature refinement. Comprehensive experiments conducted on the public RS dataset AI-TOD demonstrate that RS-TinyNet surpasses existing state-of-the-art (SOTA) detectors by 5.8% AP, 6.1% AP₅₀, and 8.4% AP₇₅. These results demonstrate that the proposed multi-dimensional attention network offers an effective and practical solution for tiny object detection in challenging RS scenarios.