Scholarlink

<p>Remote sensing images present large-scale variations, dense small objects, and complex backgrounds. In resource-constrained scenarios such as unmanned aerial vehicle-based monitoring, detection algorithms require both high accuracy and efficient real-time inference. However, the high computational cost of existing deep learning methods makes it difficult to balance detection performance, real-time efficiency, and lightweight design. To address this issue, this paper proposes a lightweight remote sensing object detection model termed Multi-scale Cross-channel and Frequency-domain Feature Enhancement-DETR (MCFE-DETR). Specifically, we construct a context-guided lightweight feature extraction module to maintain accuracy and ensure a lightweight model. Additionally, we design a multi-scale cross-channel feature fusion encoder to improve precision and robustness for multi-scale objects. Meanwhile, we embed a wavelet-based frequency-domain enhancement module into the encoder to enhance perception of dense small targets and fine-grained boundaries. Experiments on the Remote Sensing Target Detection (RSOD), Dam Target Detection in Remote Sensing Imagery (DTDRSI), and Satellite Imagery&#xa0;Multi-vehicles Dataset (SIMD) datasets show that MCFE-DETR improves detection accuracy by 3.2%, 1.5%, and 3.0%, respectively, over the baseline, while reducing model parameters and computational complexity by 36.6% and 30.7%. Overall, it achieves a balance between detection accuracy, real-time inference efficiency, and model complexity, satisfying real-time demands of resource-constrained remote sensing applications.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

MCFE-DETR: a lightweight remote sensing object detection model with multi-scale cross-channel and frequency-domain feature enhancement

  • 期刊论文

  • 发表日期：

    2026年5月5日

查看全文
  • Shoubin Wang,
  • Yawei Liu,
  • Guili Peng,
  • Huaipeng He,
  • Tong Wang,
  • Qiuying Niu,
  • Shaojie Yang,
  • Shuo Yang

摘要

Remote sensing images present large-scale variations, dense small objects, and complex backgrounds. In resource-constrained scenarios such as unmanned aerial vehicle-based monitoring, detection algorithms require both high accuracy and efficient real-time inference. However, the high computational cost of existing deep learning methods makes it difficult to balance detection performance, real-time efficiency, and lightweight design. To address this issue, this paper proposes a lightweight remote sensing object detection model termed Multi-scale Cross-channel and Frequency-domain Feature Enhancement-DETR (MCFE-DETR). Specifically, we construct a context-guided lightweight feature extraction module to maintain accuracy and ensure a lightweight model. Additionally, we design a multi-scale cross-channel feature fusion encoder to improve precision and robustness for multi-scale objects. Meanwhile, we embed a wavelet-based frequency-domain enhancement module into the encoder to enhance perception of dense small targets and fine-grained boundaries. Experiments on the Remote Sensing Target Detection (RSOD), Dam Target Detection in Remote Sensing Imagery (DTDRSI), and Satellite Imagery Multi-vehicles Dataset (SIMD) datasets show that MCFE-DETR improves detection accuracy by 3.2%, 1.5%, and 3.0%, respectively, over the baseline, while reducing model parameters and computational complexity by 36.6% and 30.7%. Overall, it achieves a balance between detection accuracy, real-time inference efficiency, and model complexity, satisfying real-time demands of resource-constrained remote sensing applications.