Scholarlink

<p>Medicinal plants are known to contain therapeutic compounds, whether used directly for healing purposes or as a precursor for producing beneficial drugs. This distinction is important as it distinguishes scientifically validated medicinal plants from those that have not been proven. With an exhaustive review of the existing literature, it was observed that the existing models did not perform well in real-world settings because very limited datasets are available for real-world scenarios. Hence, at the onset of this work, a large-scale Real-Environment Medicinal plantS (REMEDS) dataset was prepared. It comprises 8,000 images under 15 classes with similar-looking interclass samples, making this dataset even more challenging. Subsequently, this work proposes a novel and efficient deep-learning model named VitNas for the classification of medicinal plants. It combines the capabilities of two pre-trained models, Vit_B32 and Nasnet Large, to leverage their strengths while minimizing their limitations. Our results demonstrated a significant increase in classification accuracy of 99.7% on the test set of the REMEDS dataset. This model performed well on REMDES and worked efficiently on existing lab-prepared datasets. Moreover, the proposed model carries an additional LLM layer that provides valuable information on the medicinal properties of the predicted class values, thereby enabling individuals to benefit from their therapeutic qualities.</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

VitNas: a deep learning model for efficient identification of medicinal plants using real-world datasets

  • 期刊论文

  • 发表日期：

    2025年7月16日

查看全文
  • Zeba Anwar,
  • Sarfaraz Masood

摘要

Medicinal plants are known to contain therapeutic compounds, whether used directly for healing purposes or as a precursor for producing beneficial drugs. This distinction is important as it distinguishes scientifically validated medicinal plants from those that have not been proven. With an exhaustive review of the existing literature, it was observed that the existing models did not perform well in real-world settings because very limited datasets are available for real-world scenarios. Hence, at the onset of this work, a large-scale Real-Environment Medicinal plantS (REMEDS) dataset was prepared. It comprises 8,000 images under 15 classes with similar-looking interclass samples, making this dataset even more challenging. Subsequently, this work proposes a novel and efficient deep-learning model named VitNas for the classification of medicinal plants. It combines the capabilities of two pre-trained models, Vit_B32 and Nasnet Large, to leverage their strengths while minimizing their limitations. Our results demonstrated a significant increase in classification accuracy of 99.7% on the test set of the REMEDS dataset. This model performed well on REMDES and worked efficiently on existing lab-prepared datasets. Moreover, the proposed model carries an additional LLM layer that provides valuable information on the medicinal properties of the predicted class values, thereby enabling individuals to benefit from their therapeutic qualities.