Scholarlink

<p>The flower recognition model can improve the intelligence level of agricultural production and provide important support for ecological protection. However, the current models have problems such as weak recognition ability, many parameters, and serious overfitting. Therefore, we propose a lightweight floral recognition model, the Grouping Channel Mixing and Enhanced Dense Connection Model (GCM-EDCNet). The model enhances performance through two design approaches: firstly, it employs “ group channel mixing blocks ” to group channels, thereby reducing the number of parameters while strengthening information exchange between different groups; secondly, it utilises “ enhanced dense connection blocks ” to fuse outputs from different blocks, which aids in combining shallow and deep features, thus improving the model’s generalisation capability. The accuracy of the model is 93.88% on the flower classification dataset and 89.6% on the flower recognition dataset, with a specificity of 99.5% and 97.4%, respectively, and the size of the model is only 2.58M. The model greatly improves the flower recognition ability and reduces a host of model parameters. Compared to existing models in flower recognition tasks, it achieves a superior balance between lightweight design and recognition accuracy. Our code and data are available at <a href="https://github.com/guizhouren/GCM-EDCNet/tree/master">https://github.com/guizhouren/GCM-EDCNet/tree/master</a>.</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

GCM-EDCNet: a lightweight model for flower recognition leveraging grouping channel mixing and enhanced dense connection

  • 期刊论文

  • 发表日期：

    2026年1月5日

查看全文
  • Minghan Li,
  • Yun Wu,
  • Chengdong Ye,
  • Zhihao Li

摘要

The flower recognition model can improve the intelligence level of agricultural production and provide important support for ecological protection. However, the current models have problems such as weak recognition ability, many parameters, and serious overfitting. Therefore, we propose a lightweight floral recognition model, the Grouping Channel Mixing and Enhanced Dense Connection Model (GCM-EDCNet). The model enhances performance through two design approaches: firstly, it employs “ group channel mixing blocks ” to group channels, thereby reducing the number of parameters while strengthening information exchange between different groups; secondly, it utilises “ enhanced dense connection blocks ” to fuse outputs from different blocks, which aids in combining shallow and deep features, thus improving the model’s generalisation capability. The accuracy of the model is 93.88% on the flower classification dataset and 89.6% on the flower recognition dataset, with a specificity of 99.5% and 97.4%, respectively, and the size of the model is only 2.58M. The model greatly improves the flower recognition ability and reduces a host of model parameters. Compared to existing models in flower recognition tasks, it achieves a superior balance between lightweight design and recognition accuracy. Our code and data are available at https://github.com/guizhouren/GCM-EDCNet/tree/master.