• Ramya B N
• Akash M Athreyas
• Rahul Sai L
• Sujith A Donti

Accurate identification of medicinal plant species from leaf images captured under unconstrained real-world conditions is essential for pharmacognosy, traditional medicine and biodiversity monitoring, yet remains challenging due to complex backgrounds, variable illumination and intra-class morphological similarity. In this work, we fine-tune a deep residual network (ResNet-50) pretrained on ImageNet, leveraging the Indian Medicinal Leaves Image Datasets [1]—a repository of over 8 000 high-resolution leaf photographs acquired without environmental constraints. To improve generalization across background clutter and lighting variation, we apply an extensive augmentation pipeline (random rotations, color jittering, background residualization) and transfer-learning strategies. On a held-out test set of 80 species, our system achieves 96.5 % classification accuracy, outperforming classical CNN baselines by over 12 %. Class activation maps confirm that the model reliably attends to venation and margin patterns despite background noise. The proposed approach offers a scalable, high-precision tool for automated leaf-based plant identification, with immediate applicability in digital herbariums and mobile field-identification apps.

Indian Medicinal Plant Classification, ResNet-50, Transfer Learning, Data Augmentation, Unconstrained Backgrounds

"A ResNet-Based Classification System for Indian Medicinal Plant Leaves", International Journal of Emerging Technologies and Innovative Research (www.jetir.org), ISSN:2349-5162, Vol.12, Issue 5, page no.f435-f438, May-2025, Available :http://www.jetir.org/papers/JETIR2505648.pdf

"A ResNet-Based Classification System for Indian Medicinal Plant Leaves", International Journal of Emerging Technologies and Innovative Research (www.jetir.org | UGC and issn Approved), ISSN:2349-5162, Vol.12, Issue 5, page no. ppf435-f438, May-2025, Available at : http://www.jetir.org/papers/JETIR2505648.pdf