Scholarlink

<p>Remote sensing target detection is crucial for industrial, civilian, and military applications. However, existing methods like YOLOv8 struggle with challenges such as small object sizes, complex backgrounds, and inconsistent lighting in remote sensing images. To address these limitations, we propose EBC-YOLO. EBC-YOLO incorporates several significant advancements. First, it adds an enhanced detection layer in the head to produce higher-resolution feature maps, capturing finer details. Second, the neck utilizes a bidirectional feature pyramid network (BiFPN) to effectively integrate feature information, further enhanced by a novel feature fusion layer designed to improve inclusiveness and extraction efficiency. Lastly, the backbone integrates a CNF module, combining depthwise separable convolutions and inverse bottleneck layers, to replace part of the C2f block, reducing feature loss during downsampling and propagation. Experiments on the VisDrone-2019 and DOTA datasets show that EBC-YOLO outperforms YOLOv8, achieving mAP@0.5 scores of 44.3% and 49.7%, representing improvements of 5.3% and 6.0%, respectively. These findings confirm that EBC-YOLO delivers enhanced detection performance in complex remote sensing environments while reducing computational costs. Furthermore, the proposed improvements have broad implications for real-world applications, such as agricultural monitoring and urban planning, where accurate and efficient object detection plays a critical role in addressing key challenges.</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

EBC-YOLO: a remote sensing target recognition model adapted for complex environments

  • 期刊论文

  • 发表日期：

    2025年2月22日

查看全文
  • Huakun Luo,
  • Yuqi Wang,
  • Youlin Chen,
  • Xi Li,
  • Jianhui Zhan,
  • Deng Zuo

摘要

Remote sensing target detection is crucial for industrial, civilian, and military applications. However, existing methods like YOLOv8 struggle with challenges such as small object sizes, complex backgrounds, and inconsistent lighting in remote sensing images. To address these limitations, we propose EBC-YOLO. EBC-YOLO incorporates several significant advancements. First, it adds an enhanced detection layer in the head to produce higher-resolution feature maps, capturing finer details. Second, the neck utilizes a bidirectional feature pyramid network (BiFPN) to effectively integrate feature information, further enhanced by a novel feature fusion layer designed to improve inclusiveness and extraction efficiency. Lastly, the backbone integrates a CNF module, combining depthwise separable convolutions and inverse bottleneck layers, to replace part of the C2f block, reducing feature loss during downsampling and propagation. Experiments on the VisDrone-2019 and DOTA datasets show that EBC-YOLO outperforms YOLOv8, achieving mAP@0.5 scores of 44.3% and 49.7%, representing improvements of 5.3% and 6.0%, respectively. These findings confirm that EBC-YOLO delivers enhanced detection performance in complex remote sensing environments while reducing computational costs. Furthermore, the proposed improvements have broad implications for real-world applications, such as agricultural monitoring and urban planning, where accurate and efficient object detection plays a critical role in addressing key challenges.