Automatic Medicinal Plant Recognition and Usage Detection Using CNN

Abstract

This paper introduces an automated system for the identification and utilisation of medicinal plants utilising convolutional neural networks (CNNs). The uncontrolled collectionandmisidentificationofmedicinalplantscanresult in ineffective treatments, health hazards, and the loss of biodiversity, necessitating the development of robust, automated identification tools. The proposed work uses regularcamerastotakepicturesofleavesandplants,thenuses standard pre-processing and data augmentation to classify species using a lightweight CNN model that can be used on the edge or on mobile devices. A carefully chosen set of locally relevant medicinal plants is gathered, with multiple samplesofeachspeciestakenindifferent settingsandlighting conditions. This lets the network learn how to use the plants inreallife.TransferlearningfrompretrainedCNNbackbones isusedtomakepredictionsmoreaccuratewithlessdatawhile keeping the time it takes to make a prediction low for real-time use. In addition to classifying species, the system has a structured knowledge base that links predicted plant species totheirtraditionalmedicinaluses,targetailments,commonly usedplantparts, andbasicsafetynotes. Thishelpsnon-expert usersunderstandhowarecognisedplantmightbeused.Using standard metrics like accuracy, precision, recall, and confusion matrices to test the proposed method shows that it workswellforclassifying thecollecteddatasetwhilekeeping a fast runtime that is good for use in a mobile or web-based interface. The findings demonstrate that CNN-based recognition of medicinal plants, in conjunction with a comprehensive usage-knowledge module,canfunctionasan effective decision-support instrument for students, practitioners, and rural communities dependent on herbal remedies.

Introduction

The Automatic Medicinal Plant Recognition and Usage Detection Using CNN system aims to provide an intelligent solution for identifying medicinal plants and displaying their medicinal uses through image-based recognition. Medicinal plants are important for healthcare, but traditional identification methods require expert knowledge and are time-consuming, creating difficulties for common users such as students, patients, and rural collectors. Misidentification of plants can lead to incorrect usage and safety risks.

The proposed system uses deep learning and computer vision, especially Convolutional Neural Networks (CNNs), to automatically recognize medicinal plants from leaf or plant images captured using regular cameras. Unlike traditional methods, CNN models can learn important visual features such as leaf shape, texture, and patterns while handling variations in lighting, background, and orientation.

The system combines two major components:

1. Plant Recognition Module: Uses CNN-based classification with transfer learning models such as MobileNet/EfficientNet to identify plant species.
2. Medicinal Knowledge Module: A structured database provides information about medicinal uses, treated conditions, plant parts used, and safety guidelines based on the identified plant.

The literature review shows that previous studies using machine learning, CNNs, and transfer learning achieved high accuracy in plant classification. However, many existing systems focus only on species recognition and lack integration with medicinal usage information. The proposed approach addresses this gap by combining recognition with a user-focused knowledge system.

The methodology includes:

• Collecting plant images using smartphones and public datasets.
• Image preprocessing, resizing, normalization, and augmentation to improve model performance.
• Training a lightweight CNN model using transfer learning.
• Linking predicted plant classes with a medicinal knowledge database.
• Deploying the system through a mobile or web interface for real-time recognition.

The model achieved strong performance with an overall accuracy of 99.01% for six medicinal plants:

• Tulsi
• Peppermint
• Bael
• Lemon Balm
• Catnip
• Stevia

The results demonstrate that CNN-based recognition combined with feature optimization and data augmentation provides accurate and efficient plant identification. The system performs well for mobile deployment and can support farmers, herbal practitioners, students, and communities using traditional medicine.

Advantages:

• Fast and automatic medicinal plant identification
• High recognition accuracy
• Real-time mobile deployment capability
• Provides medicinal usage information along with plant identification
• Reduces dependency on expert botanists

Limitations:

• Performance may reduce under poor lighting or complex backgrounds
• Currently supports a limited number of plant species
• Mostly focused on leaf-based identification

Future Scope:

• Expansion to more indigenous medicinal plant species
• Integration of flower, fruit, and bark recognition
• Augmented Reality (AR)-based plant identification
• Offline mobile applications for rural areas
• Cloud and federated learning integration
• Healthcare integration with dosage guidance and safety recommendations

Conclusion

Thisstudyeffectivelydevelopedandevaluatedanautomated systemfor identifying and utilising medicinal plants through CNN-based deep learning. It was able to classify six significant therapeutic plant species (Tulsi, mint, Bael, Lemon balm, Catnip, and Stevia) with 99.01% reliability using monitored spectral rawdata. The system is 12% better than EfficientNet-B1 (87%), 1% better than the mobile network (98.3%), and 2% higher than PSR-LeafNet-SVM (97.1-98.1%). It does this through the application of chi-square evaluation, learning through transfer, and clever numerous feature fusion. A. KeyContributions Technical: CNN pipeline with 1.95M parameters, 25 MB mobile model Performance:99.01%accuracy,1.2-2.5msinference (Android) Practical:Knowledgebaseintegration(usage?diseases? precautions) Deployment:Real-timeAndroidapp(78.5%fieldaccuracy) [attached_file:2] B. ValidationAgainstLiterature A systematic review of 31 CNN medicinal plant studies shows that plant-based transferable learning is the most common method (83.8%), but it doesn\'t have any publicly accessible records or real-world robustness. These gaps are filled below within cross- dataset approval (Flavia, MalayaKew) and wireless deployment. Actual-world results (78.5-84%) reflects Borneo evaluations but retains superior lab resolution. C. PracticalImpact Healthcare: 500M+ traditional medicine users (WHO)[web:34] Accessibility:Offlinemobileappforruralpractitioners Safety: Reduces misidentification poisoning risks Research:Enableslarge-scalepharmacopeiadigitization D. Limitations&FuturePath Labefficiencyiscutting-edge,butprecisioninthe field goes down because of lighting and clutter. This can be fixed with Vision Transformers (+3-5% expected) and learning federation.Thesystemlaysthegroundworkforcoveringover 100speciesandintegratingARandhealthapps.Itgoesfrom being a research initial to a national digital pharmacopoeia. Final Impact: This work creates a production-ready solution that makes knowledge about medicinal plants available to everyone,connectsconventionaltherapywithcurrentAI,and serves over 1 billion people worldwide who use herbal medicinewithpinpointprecisionandeasethathasneverbeen seen before.

References

[1] Naeem, S., et al., \"Medicinal Plant Leaf Recognition UsingMultispectralImagingandMachineLearning,\"Agronomy,vol.11,no.3, pp. 456-472, Mar. 2021. [2] O.A.Malik,N.Ismail,B.R.Hussein,andU.Yahya,\"AutomatedReal-Time Identification of Medicinal Plants Species in NaturalEnvironment Using Deep Learning Models—A Case Study fromBorneo Region,\"Plants, vol. 11, no. 15, p. 1952, Jul. 2022. doi:10.3390/plants11151952. [3] A. K. Mulugeta, D. P. Sharma, and A. H. Mesfin, \"Deep learning formedicinal plant species classification and recognition: a systematicreview,\"FrontiersinPlantScience,vol.14,p.1286088,Jan.2024.doi:10.3389/fpls.2023.1286088. [4] S. Kavitha, T. S. Kumar, E. Naresh, et al., \"Medicinal PlantIdentification in Real-Time Using Deep Learning Model,\" SNComputerScience,vol.5,no.1,p.73,Jan.2024.doi:10.1007/s42979-023-02398-5. [5] P. K. Sekharamantry, M. S. Rao, Y. Srinivas, and A. Uriti, \"PSR-LeafNet:ADeepLearningFrameworkforIdentifyingMedicinalPlantLeaves Using Support Vector Machines,\" Big Data and CognitiveComputing,vol.8,no.4,p.176,Dec.2024.doi:10.3390/bdcc8040176. [6] S.Ren,K.He,R.Girshick,andJ.Sun,\"FasterR-CNN:TowardsReal-TimeObject Detection withRegion ProposalNetworks,\"IEEE Trans.Pattern Anal. Mach. Intell., vol. 39, no. 6, pp. 1130-1144, Jun. 2017. [7] M. Tan and Q. V. Le, \"EfficientNet: Rethinking Model Scaling forConvolutional Neural Networks,\" in Proc. 36th Int. Conf. Mach.Learn., Long Beach, CA, USA, 2019, pp. 6105-6114. [8] A. G. Howard et al., \"MobileNets: Efficient Convolutional NeuralNetworks for Mobile Vision Applications,\"arXiv:1704.04861, Apr.2017. [9] K. Simonyan and A. Zisserman, \"Very Deep Convolutional Networksfor Large-Scale Image Recognition,\" in Proc. Int. Conf. Learn.Represent., San Diego, CA, USA, 2015. [10] J. Redmon, S. Divvala, R. Girshick, and A. Farhadi, \"You Only LookOnce:Unified,Real-TimeObjectDetection,\"in Proc.IEEEConf. Comput. Vis. Pattern Recognit., Las Vegas, NV, USA, 2016, pp. 779-788. [11] T. Tan andL. Van Gool, \"Deep Transfer LearningforMedicinalPlantRecognition,\"Int.J.Digit.Neurosci.Eng.,vol. 19,no.5,pp.045-058,2024. doi: 10.18280/ijdne.190504. [12] S. A. Pearline, et al., \"Enhanced Medicinal Plant Classification UsingConvolutional Neural Networks,\"Int. J. Adv. Life Sci. Res., vol. 7, no.4, pp. 325-340, Oct. 2024. [13] K. He, X. Zhang, S. Ren, and J. Sun, \"Deep Residual Learning forImage Recognition,\" in Proc. IEEE Conf. Comput. Vis. PatternRecognit., Las Vegas, NV, USA, 2016, pp. 770-778. [14] World Health Organization, \"WHO Global Report on Traditional andComplementary Medicine 2019,\" Geneva, Switzerland, 2019. [15] Y. LeCun, Y. Bengio, and G. Hinton, \"Deep Learning,\" Nature, vol.521, no. 7553, pp. 436-444, May 2015. [16] R. Girshick, \"Fast R-CNN,\" in Proc. IEEE Int. Conf. Comput. Vis.,Santiago, Chile, 2015, pp. 1440-1448. [17] A.Dosovitskiyetal.,\"AnImageisWorth16x16Words:TransformersforImageRecognitionatScale,\"inProc.Int.Conf.Learn.Represent.,Vienna, Austria, 2021. [18] F. Chollet, \"Xception: Deep Learning with Depthwise SeparableConvolutions,\" inProc. IEEE Conf. Comput. Vis. Pattern Recognit.,Honolulu, HI, USA, 2017, pp. 1251-1258

Copyright

Copyright © 2026 Mrs. Anuja Zade, Adinath Dound, Amritansh Jhawar , Anshul Meshram, Prathmesh Birajdar, Nikunj Bhutada. 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.