<p>Detecting tiny objects in aerial images has always been a perennial challenge in computer vision. The tiny objects contain limited pixel representation and are susceptible to background noise, making accurate detection difficult. This paper proposes a novel framework for the detection of tiny objects, which is guided by an end-to-end query-based approach. There are two inherent drawbacks of previous query-based detectors: first, query-based detectors are inherently insensitive to the detection of tiny objects; second, the performance of the detectors gradually saturates as the network depth increases. This paper proposes a novel approach to solve the above problems by leveraging the property that queries are one-to-one label assignment rules. Specifically, the interactive query prediction selector first selects queries with high confidence scores as acceptable queries. Then, a contrastive learning information extractor is used to progressively assign samples to the accepted and improved noisy queries. Finally, a progressive search is used to generate refined prediction anchor boxes. We conducted extensive experiments on three publicly available aerial image datasets, namely DOTA, VisDrone, and AI-TOD, to demonstrate the usefulness and robustness of our proposed method. The results show that our proposed method yields significant performance improvements in the tiny object detection task, outperforming the existing benchmark models.</p>