A Fine-Grained Aircraft Detection Method Based on an Asynchronous Push-Pull Network

With the rapid development of high-resolution remote sensing technologies, remote sensing imagery has become indispensable in military reconnaissance and precision strike missions. However, fine-grained object detection in such imagery remains highly challenging due to subtle inter-class discrepancies and pronounced intra-class variations. To address these challenges, this paper proposes a fine-grained object detection framework designed to optimize detection accuracy. The framework integrates an improved Faster R-CNN with a multi-level feature-enhanced cascaded RoI head, an asynchronous push-pull network, and a contrastive learning-based optimization module. Experimental evaluations conducted on the FAIR1M dataset demonstrate that the proposed model outperforms the baseline by 10.8 percentage points in mAP@50, thereby achieving a notable enhancement in fine-grained detection accuracy.