Automated Classification of Medicinal Plants Using Lightweight Deep Learning and Transfer Learning

Title

Automated Classification of Medicinal Plants Using Lightweight Deep Learning and Transfer Learning

Creator

Kavitha, S.; Vinay, M.; Joby, Sona; Ribu, P.B.

Description

The identification of medicinal plants plays a pivotal role in traditional medicine, biodiversity conservation, and rural healthcare. Conventional manual identification methods are often time-consuming and error-prone, particularly when differentiating between morphologically similar species or plants at varying growth stages. Recent developments in deep learning, especially convolutional neural networks (CNNs) with transfer learning, have emerged as robust solutions for image-based classification tasks, offering efficiency and high accuracy with limited computational resources. The proposed framework employs a carefully structured deep learning pipeline integrating advanced preprocessing, lightweight architecture design, and domain-adaptive transfer learning. A large real-world dataset of 20,109 medicinal leaf images across 99 classes was standardized through resizing, normalization, and categorical encoding, followed by targeted data augmentation and class-weight balancing to address inter-class similarity and dataset imbalance. A key methodological novelty lies in the use of MobileNetV3 with an optimized transfer-learning strategy, leveraging its inverted residual blocks, Squeeze-and-Excite modules, and hard-swish activation to enhance texture-, venation-, and contour-based feature extraction in plant leaves. Unlike existing plant-recognition studies that rely on heavier CNNs, our approach introduces a computationally efficient, low-latency model specifically tailored for mobile and embedded deployment. Experimental results demonstrate that the proposed MobileNetV3-based model achieved a classification accuracy of 92.88%, with macro- and weighted-average F1-scores of 0.85 and 0.86, respectively. Precision and recall values across most classes ranged between 0.80 and 0.95, confirming the models reliability in differentiating species. The Author(s), under exclusive licence to Springer Nature Singapore Pte Ltd. 2026.

Source

SN Computer Science;Volume;7;Issue;3;Article No.;263;

Date

01-01-2026

Publisher

Springer

Subject

Biodiversity informatics; Deep learning; Feature extraction; Healthcare technology; Medicinal plants; Plant identification; Smart agriculture tools

Coverage

Kavitha S., Department of Computer Science, Christ University, Karnataka, Bengaluru, 560029, India; Vinay M., Department of Computer Science, Christ University, Karnataka, Bengaluru, 560029, India; Joby S., Department of Computer Science, Christ University, Karnataka, Bengaluru, 560029, India; Ribu P.B., Department of Computer Science, Christ University, Karnataka, Bengaluru, 560029, India

Rights

Restricted Access; Hardcopy may be available in the library

Relation

ISSN: 2662995X;

Format

online

Language

English

Type

Article

Identifier

https://doi.org/10.1007/s42979-025-04716-5

https://www.scopus.com/pages/publications/105033842899?origin=resultslist