HF-DETR: High-Frequency Feature Enhancement for Small Object Detection
Junlin Zhu, Yanqing Zhang
South China University of Technology
阅读操作
确认中在文库中上传 PDF 后可生成中文音频讲解。
摘要与影响
Real-time small object detection in unmanned aerial vehicle (UAV) imagery remains challenging, as high-frequency cues (edges, textures) of tiny targets are inevitably degraded by repeated downsampling. To address this limitation, we propose HF-DETR, a high-frequency enhanced detector that systematically tackles the degradation problem through three complementary mechanisms. A Frequency-Spatial Decoupled Dual-Stream Stem (FSD-Stem) explicitly isolates edge information at the earliest stage using a Laplacian-of-Gaussian operator. A Frequency-Enhanced Feature Reconstruction Encoder (FEFR-Encoder) leverages wavelet-based inverse projection to reconstruct fine-grained details during multi-scale fusion. A Sparse Saliency Micro-Gating (SSMG) module enforces token sparsity to suppress cluttered backgrounds. Experimental results show that HF-DETR improves AP and AP$_{50}$by 4.3 and 6.0 points respectively on VisDrone2019-DET compared with the baseline, while reducing the number of parameters from 10.2M to 8.6M and maintaining real-time performance at 121 FPS. These results validate that explicitly modeling high-frequency cues is crucial for real-time small object detection in UAV imagery.
逐年被引趋势
关键指标
同类平均 = 1
同领域 · 同年份 · 同类型
Google Scholar 与 OpenAlex 的被引统计范围不同,数值存在差异属正常。
AI 辅助阅读
依据:摘要
可就本文提问;依据不足时会说明。
学术脉络
学科主题
计算机 / AIAdvanced Neural Network Applications
Visual Attention and Saliency Detection · Advanced Image and Video Retrieval Techniques