Scholarlink

<p>We propose a novel Deep-Multiscale Stratified Aggregation (D-MSA) module, which is specifically designed to enhance the extraction and fusion of multi-scale features across various receptive fields. In contrast to conventional convolutional architectures, D-MSA effectively bridges the gap between shallow and deep features by addressing differences in scale and semantic content. Integrated into the YOLO architecture, D-MSA significantly improves the model’s capability to process complex multiscale information without compromising computational efficiency. Experiments demonstrate that the incorporation of D-MSA could lead to a notable improvement in the accuracy of object detection.</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

Deep-multiscale stratified aggregation

  • 期刊论文

  • 发表日期：

    2025年9月26日

查看全文
  • Ziheng Wu,
  • Song Yang,
  • Fengxiang Hu,
  • Jiaxiang Yao,
  • Jun Zhou,
  • Jingyuan Wang,
  • Yongtao Li

摘要

We propose a novel Deep-Multiscale Stratified Aggregation (D-MSA) module, which is specifically designed to enhance the extraction and fusion of multi-scale features across various receptive fields. In contrast to conventional convolutional architectures, D-MSA effectively bridges the gap between shallow and deep features by addressing differences in scale and semantic content. Integrated into the YOLO architecture, D-MSA significantly improves the model’s capability to process complex multiscale information without compromising computational efficiency. Experiments demonstrate that the incorporation of D-MSA could lead to a notable improvement in the accuracy of object detection.