FEMS-DETR: A Frequency-Enhanced Transformer for Object Detection in Remote Sensing Images

The detection of objects in remote sensing imagery faces numerous challenges, primarily due to cluttered backgrounds, significant scale variations, and densely distributed small objects. To address these issues, this paper presents an object detection framework named FEMS-DETR, specifically designed for remote sensing imagery. The core of the framework consists of two main modules: a Frequency-enhanced Multi-scale Perception Module (FMPM), which improves feature extraction for multi-scale and small objects through parallel multi-scale convolutional processing and frequency-domain transformations, and a Detail-Preserving Downsampling (DPD) module, which employs a three-branch architecture integrated with an attention mechanism to retain critical features during downsampling. Experimental results on the NWPU VHR-10 and RSOD datasets demonstrate that the proposed method achieves competitive results compared to existing state-of-the-art approaches, with mAP scores of 95.00% and 96.22%, respectively. It performs particularly well in detecting small objects and maintains robust performance in complex scenarios.