Scholarlink

<p>Although significant progress has been made in object detection within the field of computer vision, due to factors such as cluttered backgrounds, object overlaps, varying target scales, and arithmetic limitations, it is difficult for existing object detection algorithms to achieve ideal results in small object detection. Therefore, this paper proposes an enhanced small object detection method, MRF-yolo, which makes a series of improvements based on the YOLOv8 as the basic model. First, the multi-level re-parameter fusion module MRF is designed to effectively fuse cross-scale feature information. Second, the PPA attention mechanism is introduced in the backbone network instead of the traditional convolution module to extract local detail features and alleviate the feature dilution problem. Finally, using Wise-IoU v3 as the regression loss function aims to improve the localization accuracy and accelerate the convergence speed. To verify the effectiveness of the proposed algorithm, this paper uses the publicly available datasets Visdrone2019 and AI-TOD for evaluation. The mAP0.5 of MRF-YOLO on Visdrone2019 increases from 38.5 to 45.1%. On AI-TOD, the mAP0.5 increases from 40.2 to 49.1%. The significant improvement in detection accuracy demonstrates the effectiveness and generalization ability of the proposed algorithm.</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

MRF-YOLO: small object detection for UAV and remote sensing images

  • 期刊论文

  • 发表日期：

    2025年10月29日

查看全文
  • Yuying Wang,
  • Wenlu Yang

摘要

Although significant progress has been made in object detection within the field of computer vision, due to factors such as cluttered backgrounds, object overlaps, varying target scales, and arithmetic limitations, it is difficult for existing object detection algorithms to achieve ideal results in small object detection. Therefore, this paper proposes an enhanced small object detection method, MRF-yolo, which makes a series of improvements based on the YOLOv8 as the basic model. First, the multi-level re-parameter fusion module MRF is designed to effectively fuse cross-scale feature information. Second, the PPA attention mechanism is introduced in the backbone network instead of the traditional convolution module to extract local detail features and alleviate the feature dilution problem. Finally, using Wise-IoU v3 as the regression loss function aims to improve the localization accuracy and accelerate the convergence speed. To verify the effectiveness of the proposed algorithm, this paper uses the publicly available datasets Visdrone2019 and AI-TOD for evaluation. The mAP0.5 of MRF-YOLO on Visdrone2019 increases from 38.5 to 45.1%. On AI-TOD, the mAP0.5 increases from 40.2 to 49.1%. The significant improvement in detection accuracy demonstrates the effectiveness and generalization ability of the proposed algorithm.