Hybrid CNN-PSO-SVM Framework for Medicinal Plant Recognition Using Leaf Image Classification

Abstract

Accurate identification of medicinal plants continues to be a critical challenge in botanical studies and traditional healthcare systems. Manual identification is highly dependent on specialized expertise and is inherently susceptible to human error. To address this, we propose an automated deep-learning frame-work designed specifically to classify therapeutic plants based on their leaf characteristics. Our approach leverages MobileNetV2 for initial feature extraction, followed by a Particle Swarm Optimization (PSO) mechanism to select the most relevant features, and concludes with a Support Vector Machine (SVM) for final classification. This hybrid CNN-PSO-SVM architecture was evaluated on a custom dataset featuring 19 distinct species of Indian medicinal flora. Additionally, the model was integrated into a unified digital platform, comprising a React-based frontend and a Node.js backend. To enhance robustness, Google’s Gemini Vision API serves as an auxiliary classification method. Our experimental results demonstrate an overall accuracy of 82.45%. Furthermore, to facilitate edge computing on mobile applications, the optimized model was successfully exported as a compact TFLite file, highlighting its efficiency and practical deployability.

Introduction

The work focuses on building an automated system for identifying Indian medicinal plants to reduce reliance on expert botanists and prevent misidentification of species. It proposes a computer vision approach using Convolutional Neural Networks (CNNs), specifically MobileNetV2, combined with Particle Swarm Optimization (PSO) for feature selection and Support Vector Machine (SVM) for final classification. This hybrid pipeline improves accuracy while reducing computational complexity by selecting the most relevant features from high-dimensional CNN outputs.

The system is trained on a dataset of 19 medicinal plant species, including Neem, Tulsi, Aloe Vera, and Ashwagandha, with images collected from field captures and online sources. Images are preprocessed (resized, normalized, and augmented) to improve model robustness, and the dataset is split into 80% training and 20% testing.

The methodology has three stages: (1) MobileNetV2 extracts 1,280 deep features from images, (2) PSO reduces these features to an optimal subset, and (3) an SVM performs final classification. The system is further optimized for deployment on low-power devices using a lightweight TFLite model and integrated into a full-stack application with web and mobile support, plus a fallback classification option using the Gemini Vision API.

Prior research highlights the effectiveness of CNNs and transfer learning for plant classification, while also noting limitations such as high computational cost and overfitting. This work addresses those issues by combining deep learning with swarm-based feature selection and classical machine learning to achieve a balance between accuracy and efficiency.

Conclusion

In summary, this research formalizes a multi-stage CNN-PSO-SVM architecture adept at differentiating 19 varied medicinal plant species. By extracting robust feature maps via MobileNetV2 and strategically discarding non-vital dimen-sions via Particle Swarm Optimization, the underlying Support Vector Machine reached an impressive 82.45% classification accuracy. Aiming for maximal utility beyond purely academic spaces, the entire recognition cycle was hosted within a dynamic web and mobile software stack. This deployment was further refined by producing a lightweight TFLite model, thereby ensuring rapid offline execution times, paired closely with an online Gemini AI contingency pathway to bolster fault tolerance. For subsequent explorations, logical efforts should center on further diversifying the dataset images, addressing fringe botanical defects, and widening the classification net to capture far rarer herbal strains. Exploring different metaheuristic algo-rithms (like Ant Colony Optimization) for advanced latency optimization also represents an exciting avenue for advancing point-of-care botanical software.

References

[1] P. Pandey and N. Pandey, “Medicinal plants of India: An ethnobotanical survey,” J. Med. Plants Res., vol. 8, no. 3, pp. 45–52, Jan. 2014. [2] Y. LeCun, Y. Bengio, and G. Hinton, “Deep learning,” Nature, vol. 521, [3] pp. 436–444, May 2015. [4] K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” in Proc. IEEE Conf. Comput. Vis. Pattern Recog-nit. (CVPR), Los Vegas, NV, USA, Jun. 2016, pp. 770–778, doi: 10.1109/CVPR.2016.90. [5] J. Yosinski, J. Clune, Y. Bengio, and H. Lipson, “How transferable are features in deep neural networks?,” in Adv. Neural Inf. Process. Syst. (NeurIPS), vol. 27, 2014, pp. 3320–3328. [6] J. Kennedy and R. Eberhart, “Particle swarm optimization,” in Proc. IEEE Int. Conf. Neural Netw., vol. 4, Perth, WA, Australia, 1995, [7] pp. 1942–1948, doi: 10.1109/ICNN.1995.488968. [8] R. Akter and M. I. Hosen, “CNN-based leaf image classification for Bangladeshi medicinal plant recognition,” in Proc. IEEE Int. Conf. Autom. Control Syst. Ind. (ICACSI), 2021, doi: 10.1109/I-CACSI52804.2021.9350900. [9] S. Srivastava, P. Bhatt, and R. Sharma, “Transfer learning approach for plant leaf disease and species classification,” Int. J. Comput. Appl., vol. 176, no. 38, pp. 12–17, 2020. [10] P. Ganesh and M. Reddy, “Medicinal plant leaf identification using SVM with texture and shape features,” in Proc. Int. Conf. Commun. Signal Process. (ICCSP), Chennai, India, 2019, pp. 831–835. [11] X. Zhang, Y. Lu, and C. Liu, “Feature selection via PSO-SVM for bioinformatics classification,” IEEE Access, vol. 6, pp. 34–45, 2018. [12] M. Sandler, A. Howard, M. Zhu, A. Zhmoginov, and L.-C. Chen, “MobileNetV2: Inverted residuals and linear bottlenecks,” in Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), Salt Lake City, UT, USA, Jun. 2018, pp. 4510–4520, doi: 10.1109/CVPR.2018.00474. [13] G. Saleem, M. Akhtar, N. Ahmed, and W. Qureshi, “Automated analysis of leaf diseases using CNN and SVM,” Comput. Electron. Agric., vol. 194, p. 106789, 2022.

Copyright

Copyright © 2026 Prof. Harshith K S, Supriya B K , Chinmayi K S, Keerthana C, Ishika M Gowda. 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.