Efficient and Explainable Medicinal Plant Classification Using Attention-Enhanced MobileNetV2

Raghavendrachar S 1, Nayana R2
1Associate Professor, Department of Computer Science and Engineering, K.S. Institute of Technology Affiliated to Visvesvaraya Technological University, Belagavi - 590018, Bengaluru, Karnataka – 560109, raghavendrachars@ksit.edu.in

2 Students, Department of Computer Science and Engineering, K.S. Institute of Technology Affiliated to Visvesvaraya Technological University, Belagavi - 590018, Bengaluru, Karnataka – 560109, nayanaramesh15@gmail.com

Abstract:
Accurate identification of medicinal plants is important for healthcare, agriculture, and traditional medicine applications. However, manual identification is time-consuming and often requires expert botanical knowledge due to the visual similarity among plant species. Paper presents Herbal VisionNet, an explainable deep learning framework for medicinal plant classification using leaf images. To improve the classification accuracy further, this study to attach Squeeze-and-Excitation (SE) attention mechanism in the network, so that the model can pay its attentions on some specific features of medicinal plants. Grad-CAM (Gradient-weighted Class Activation Mapping) visualization to explain AI is also included to interpret people Awake prediction results and cover the leaf regions responsible for predict. A Flask-based web application is built to recognize medicinal plants in real-time, providing prediction results, confidence scores, medicinal information, and visual explanations. Experimental results show that the proposed framework achieves a validation accuracy of about 99% with computational efficiency. The proposed Herbal VisionNet framework offers a lightweight, accurate, explainable, and practical solution for medicinal plant identification.

Keywords:

Medicinal Plant Classification, Deep Learning, MobileNetV2, Squeeze-and-Excitation Network, Explainable AI, Grad-CAM, Computer Vision.