Scholarlink

<p>Image processing has already expanded to substantial importance in today’s world due to its extensive range of applications across several domains, including science, healthcare, and business. One of the ongoing challenges in computer vision is the accurate recognition of plant species, particularly medicinal plants. Traditional methods are error-prone, time-consuming, and reliant on knowledge experts, so there is a need for automation approaches. In this study, we propose an EfficientViT-Net, a hybrid deep learning architecture combining the strengths of vision transformer (ViT) and ResNet50 to enhance feature extraction and classify the medicinal plants accurately and efficiently. The model is designed to overcome the deficiencies of traditional methods by tapping both global contextual information encoded by transformers and local structural information learned using convolutional networks. The suggested model was tested with a benchmark dataset on Indian medicinal plants, including with augmentation methods for enhanced robustness and diversity. Experimental results indicate that EfficientViT-Net obtains 94.7% accuracy, surpassing solely transformer-based architecture models. Furthermore, explainable AI (XAI) techniques, including Grad-CAM, Grad-CAM++, and Eigen-CAM, were employed to provide visual interpretations of the model’s decision-making process by highlighting the most influential regions in the input images. The results show the potential of hybrid approaches to not only accomplish high results but also improve transparency and interpretability in automated medicinal plant species recognition systems.</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

Explainable medicinal plant identification using a hybrid CNN-vision transformer framework

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年8月2日

查看全文
  • Faisal Firdous,
  • Deepak Gupta,
  • Hemant Sood,
  • Suman Saurabh Sarkar

摘要

Image processing has already expanded to substantial importance in today’s world due to its extensive range of applications across several domains, including science, healthcare, and business. One of the ongoing challenges in computer vision is the accurate recognition of plant species, particularly medicinal plants. Traditional methods are error-prone, time-consuming, and reliant on knowledge experts, so there is a need for automation approaches. In this study, we propose an EfficientViT-Net, a hybrid deep learning architecture combining the strengths of vision transformer (ViT) and ResNet50 to enhance feature extraction and classify the medicinal plants accurately and efficiently. The model is designed to overcome the deficiencies of traditional methods by tapping both global contextual information encoded by transformers and local structural information learned using convolutional networks. The suggested model was tested with a benchmark dataset on Indian medicinal plants, including with augmentation methods for enhanced robustness and diversity. Experimental results indicate that EfficientViT-Net obtains 94.7% accuracy, surpassing solely transformer-based architecture models. Furthermore, explainable AI (XAI) techniques, including Grad-CAM, Grad-CAM++, and Eigen-CAM, were employed to provide visual interpretations of the model’s decision-making process by highlighting the most influential regions in the input images. The results show the potential of hybrid approaches to not only accomplish high results but also improve transparency and interpretability in automated medicinal plant species recognition systems.