Scholarlink

<p>Scene classification is a prominent research area in remote sensing image analysis. However, remote sensing images frequently present challenges such as significant variations in spectral characteristics, intricate spatial structures, and mixed pixels. In response to the challenges outlined above, this paper introduces a novel method for RSS classification called Joint convolutional neural network (CNN) with mixed attention module (JCMANet). Specifically, the mixed attention module improves the comprehensiveness and accuracy of feature representation by introducing spatial and channel attention at the same time. Additionally, the combination of BiGRU with the classifier module of the CNN serves to further enhance the classification performance of the network. The experimental results demonstrate that JCMANet achieves a recognition accuracy of 93.37% on the NWPU-RESISC45 dataset. In addition, the effectiveness of each change in JCMANet was verified by ablation experiments.</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

Remote Sensing Scene Classification Based on Joint CNN with Mixed Attention Module

  • 期刊论文

  • 发表日期：

    2026年4月10日

查看全文
  • Jiangong Ni,
  • Zhigang Zhou

摘要

Scene classification is a prominent research area in remote sensing image analysis. However, remote sensing images frequently present challenges such as significant variations in spectral characteristics, intricate spatial structures, and mixed pixels. In response to the challenges outlined above, this paper introduces a novel method for RSS classification called Joint convolutional neural network (CNN) with mixed attention module (JCMANet). Specifically, the mixed attention module improves the comprehensiveness and accuracy of feature representation by introducing spatial and channel attention at the same time. Additionally, the combination of BiGRU with the classifier module of the CNN serves to further enhance the classification performance of the network. The experimental results demonstrate that JCMANet achieves a recognition accuracy of 93.37% on the NWPU-RESISC45 dataset. In addition, the effectiveness of each change in JCMANet was verified by ablation experiments.