Custom Convolutional Neural Networks and Automated Machine Learning for Medicinal Plant Identification: A Comparative Review

Abstract

The botanical misidentification of medicinal herbs, particularly the accidental substitution of therapeutic flora with toxic look-alikes, presents critical risks to global public health and pharmaceutical safety. This review synthesizes technical advancements and performance benchmarks across ten target papers, contrasting hand-tuned pipelines—including attention-enhanced EfficientNetV2, Swin-Transformers, and multi-granularity architectures—against emerging Automated Machine Learning (AutoML) frameworks. While stochastic ensembles and custom ResNet variants achieve accuracies exceeding 95%, substantial classification fragility persists when models transition from pristine laboratory imagery to unstructured wild environments. Primary dataset bottlenecks include signal interference from dynamic water reflections, soil-cluttered backgrounds, and the failure of networks to resolve intricate leaf vein geometry in dense, overlapping growth. The strategic technological outlook emphasizes the imperative for lightweight, parameter-efficient architectures suited for edge deployment, coupled with Explainable AI (XAI) tools like Grad-CAM. Such frameworks are essential to mitigate the \"double black box\" of algorithmic opacity and traditional medical theory, providing field botanists with transparent diagnostic reasoning.

Introduction

1. Transformer Turret Heating Issue (Hindalco Plant)

At Hindalco Mahan Plant, a “smoke-like” emission near a transformer was found to be oil vapor caused by leakage and overheating of turret bolts reaching ~200°C. Root cause analysis showed that induced eddy currents from high LV current (~6000 A) were flowing unevenly through turret bolts due to poor electrical contact. Paint on contact surfaces acted as insulation, causing one bolt to carry most of the current and overheat. Initial fixes (bolt replacement, material change, grounding, copper braid) failed. The issue was resolved by cleaning contact surfaces and removing paint, enabling uniform current distribution across all bolts and eliminating localized heating.

2. Climate Change and Health

Climate change is a major global public health threat driven primarily by human activities, especially fossil fuel use. It leads to rising temperatures, extreme weather, sea-level rise, and ecosystem disruption, which increase risks of heat stress, infectious diseases, food/water insecurity, and mortality. Major contributors include energy production, industry, deforestation, transportation, agriculture, and consumption patterns. Impacts include droughts, storms, biodiversity loss, migration, poverty, and malnutrition. The text emphasizes the need for coordinated public health strategies and climate action to mitigate widespread health risks.

3. AI-Based Home Automation Systems

AI-powered home automation integrates IoT devices, sensors, and machine learning to create intelligent homes that adapt to user behavior. These systems automate lighting, temperature, appliances, security, and health monitoring by learning user routines. Technologies like voice assistants, smart mirrors, Arduino/Raspberry Pi systems, and sensors (PIR, flame, LDR) enable real-time control and safety alerts. Applications include energy efficiency, elderly care, fire detection, intrusion alerts, and remote monitoring via smartphones or cloud systems, making homes more efficient, secure, and user-friendly.

4. Green Synthesis of ZnO Nanoparticles for Agriculture

The study focuses on eco-friendly synthesis of zinc oxide (ZnO) nanoparticles using pomegranate peel extract and their application as nanofertilizers for sorghum. Plant phytochemicals act as reducing and stabilizing agents in nanoparticle formation. Characterization (UV-Vis, FTIR, SEM, XRD) confirmed successful synthesis and crystalline structure. Application of ZnO nanoparticles significantly improved seed germination, root/shoot growth, chlorophyll content, and biomass compared to controls. The work highlights sustainable agriculture using green nanotechnology to improve crop productivity while reducing chemical fertilizer use.

5. AI-Based Smart Attendance System

The system automates attendance using AI-based face recognition combined with QR code authentication. Student data and facial images are stored during registration, and CNN models are trained for real-time recognition. QR codes provide secondary verification to prevent fraud. The system records attendance automatically in a database, ensuring accuracy, security, and reduced manual effort. Compared to traditional methods (RFID, fingerprints), the hybrid approach improves reliability, prevents proxy attendance, and enables efficient centralized attendance management.

6. Face Mask Detection Using AI

This research reviews AI-based face mask detection systems used for public health monitoring. Traditional manual methods are inefficient, so machine learning and deep learning models (CNN, YOLO, MobileNet, ResNet, SVM, Random Forest) are used for automated detection. Deep learning generally achieves higher accuracy (>98%), while traditional models are faster and less resource-intensive. Key challenges include lighting variation, occlusion, and dataset imbalance. Future improvements involve Vision Transformers, edge AI, and explainable AI for better real-world deployment and transparency.

7. Medicinal Plant Identification Using AI (Literature Review)

The text reviews AI-based identification of medicinal plants, emphasizing risks from misidentification of toxic look-alikes. Traditional manual methods are unreliable, so deep learning (CNNs, ResNet, YOLO, Vision Transformers) and AutoML systems are used. While high accuracies are achieved, major challenges include limited datasets, environmental noise, morphological similarity, and lack of generalization to real-world conditions. Key issues include the “black box” nature of AI models and dataset scarcity. Future directions include explainable AI, multimodal fusion (image + chemical/genetic data), and edge deployment for real-world field use.

References

[1] S. Sendhilkumar, G. S. Mahalakshmi, A. Swaminathan, and K. Sai Anirudh, “Automated Indigenous Plant Recognition and Medicinal Value Extraction System,” in Computational Intelligence in Data Science, vol. 717, M. L. Owoc, F. E. Varghese Sicily, K. Rajaram, and P. Balasundaram, Eds., in IFIP Advances in Information and Communication Technology, vol. 717. , Cham: Springer Nature Switzerland, 2024, pp. 368–380. doi: 10.1007/978-3-031-69982-5_28. [2] S. J. Gamage K.G., A. P.N.P., N. Gamitha Manawadu, and H. De Silva, “AI-Enhanced Medicinal Plant Identification System with Multilingual Social Media Integration,” Int. J. Innov. Sci. Res. Technol. IJISRT, pp. 2603–2612, Jun. 2024, doi: 10.38124/ijisrt/IJISRT24MAY1928. [3] J. Zou and R. Liao, “AI-powered recognition of Chinese medicinal herbs with semantic structure modeling and gradient-guided enhancement,” Front. Plant Sci., vol. 16, p. 1672394, Nov. 2025, doi: 10.3389/fpls.2025.1672394. [4] J. Wu et al., “Artificial intelligence in Traditional Chinese Medicine herbs: a survey,” Artif. Intell. Rev., vol. 59, no. 4, p. 120, Feb. 2026, doi: 10.1007/s10462-026-11513-w. [5] F. Zhao, J. Zhang, Q. Liu, C. Liang, S. Zhang, and M. Li, “Fast Quality Detection of Astragalus Slices Using FA-SD-YOLO,” Agriculture, vol. 14, no. 12, p. 2194, Nov. 2024, doi: 10.3390/agriculture14122194. [6] Y. Liu, H. Liu, L. Li, and Y. Ding, “From image to insight deep learning solutions for accurate identification and object detection of Acorus species slices,” Sci. Rep., vol. 15, no. 1, p. 15734, May 2025, doi: 10.1038/s41598-025-00038-x. [7] J. Miao, Y. Huang, Z. Wang, Z. Wu, and J. Lv, “Image recognition of traditional Chinese medicine based on deep learning,” Front. Bioeng. Biotechnol., vol. 11, p. 1199803, Jul. 2023, doi: 10.3389/fbioe.2023.1199803. [8] K. Yang, G. Ding, J. Liu, and G. Du, “MGCA-Net: Multi-Granularity Cross-Attention Network with Consistency Learning for Medical Plant Recognition”. Jan. 2026, 10.2139/ssrn.6557881. [9] X. Guan, L. Shi, W. Yang, H. Ge, X. Wei, and Y. Ding, “Multi-Feature Fusion Recognition and Localization Method for Unmanned Harvesting of Aquatic Vegetables,” Agriculture, vol. 14, no. 7, p. 971, Jun. 2024, doi: 10.3390/agriculture14070971. [10] G. Zhu, J. Joo, S. Park, and S. C. Kim, “Toxic Chinese herbal medicine recognition in real-world images via multi-scale and attention-enhanced EfficientNetV2,” PLOS One, vol. 21, no. 3, p. e0344262, Mar. 2026, doi: 10.1371/journal.pone.0344262.

Copyright

Copyright © 2026 Syeda Asra Arshia Qadri. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.