Scholarlink

<p>Herbarium specimens represent invaluable archives of plant biodiversity, preserving extensive genetic resources such as Crop Wild Relatives (CWRs), which are critically important for advancing crop improvement and ensuring food security. These specimens provide essential data on plant morphology, taxonomy, geographic distribution, and ecological adaptation. The digitization of herbarium collections has significantly broadened their utilization, enabling large-scale research in plant science, biodiversity conservation, and agricultural research. However, automating the accurate identification of CWRs from digitized herbarium images remains challenging due to significant variations in plant morphology, specimen quality, and the necessity for precise species recognition. This paper offers a detailed review of recent developments in Convolutional Neural Network (CNN) models explicitly tailored for plant identification tasks. A small-scale experiment was conducted to evaluate the selected models—VGG16, MobileNet, ResNet50, DenseNet121, and InceptionV3—on a curated herbarium dataset of CWRs at the genus level. MobileNet achieved the highest accuracy at 95%, followed by VGG16 at 94%, while ResNet50, DenseNet121, and InceptionV3 each achieved accuracy of 92%, with ResNet50 showing comparatively lower performance. These results demonstrate the feasibility of applying literature-based CNN models to herbarium specimens. Furthermore, the paper highlights practical implementations and their impacts on biodiversity conservation efforts and agricultural practices. This indicates promising avenues for future research and practical applications, with the intention to further expand the dataset and evaluate performance at the species level.</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

Review of herbarium plant identification of crop wild relatives using convolutional neural network models

  • 期刊论文

  • 发表日期：

    2025年7月26日

查看全文
  • Ankur Tomar,
  • Suresh Kumar,
  • Kuldeep Tripathi

摘要

Herbarium specimens represent invaluable archives of plant biodiversity, preserving extensive genetic resources such as Crop Wild Relatives (CWRs), which are critically important for advancing crop improvement and ensuring food security. These specimens provide essential data on plant morphology, taxonomy, geographic distribution, and ecological adaptation. The digitization of herbarium collections has significantly broadened their utilization, enabling large-scale research in plant science, biodiversity conservation, and agricultural research. However, automating the accurate identification of CWRs from digitized herbarium images remains challenging due to significant variations in plant morphology, specimen quality, and the necessity for precise species recognition. This paper offers a detailed review of recent developments in Convolutional Neural Network (CNN) models explicitly tailored for plant identification tasks. A small-scale experiment was conducted to evaluate the selected models—VGG16, MobileNet, ResNet50, DenseNet121, and InceptionV3—on a curated herbarium dataset of CWRs at the genus level. MobileNet achieved the highest accuracy at 95%, followed by VGG16 at 94%, while ResNet50, DenseNet121, and InceptionV3 each achieved accuracy of 92%, with ResNet50 showing comparatively lower performance. These results demonstrate the feasibility of applying literature-based CNN models to herbarium specimens. Furthermore, the paper highlights practical implementations and their impacts on biodiversity conservation efforts and agricultural practices. This indicates promising avenues for future research and practical applications, with the intention to further expand the dataset and evaluate performance at the species level.