Scholarlink

<p>This paper makes a thorough analysis on a few recently developed neural net- works by experimenting on leaves of some medicinal plants mostly found in different hilly and plain areas of Assam. The sample set used in this research are created by the researchers and were collected by capturing digital photographs and video recordings of the leaves from different places of Assam. The work was carried out with one twenty five (125) different species. Three segmentation techniques Watershed, SegNet and U-Net were used and among all U-Net segmented dataset was considered for final classification. For classification, two customized CNN based models, ResNet 50, Inception V3 networks and Vision transformer (ViT) have been used. A medicinal plant name table with scientific and family name, common name, Assamese name, Karbi name, parts used and disease they cure is also built. At last it is seen that ViT model with U-Net segmented data resulted most efficient output with 98.04% validation accuracy.</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

Analysis of a few neural networks with reference to identification of medicinal plants from their leaves

  • 期刊论文

  • 发表日期：

    2025年11月21日

查看全文
  • Parismita Sarma,
  • Parvez Azizi Boruah

摘要

This paper makes a thorough analysis on a few recently developed neural net- works by experimenting on leaves of some medicinal plants mostly found in different hilly and plain areas of Assam. The sample set used in this research are created by the researchers and were collected by capturing digital photographs and video recordings of the leaves from different places of Assam. The work was carried out with one twenty five (125) different species. Three segmentation techniques Watershed, SegNet and U-Net were used and among all U-Net segmented dataset was considered for final classification. For classification, two customized CNN based models, ResNet 50, Inception V3 networks and Vision transformer (ViT) have been used. A medicinal plant name table with scientific and family name, common name, Assamese name, Karbi name, parts used and disease they cure is also built. At last it is seen that ViT model with U-Net segmented data resulted most efficient output with 98.04% validation accuracy.