Ensemble-Based Plant Disease Detection with Mini TensorFlow on Risc Devices and Chatbot

Abstract

The research trains and evaluates multiple CNN architectures, including Basic CNN, AlexNet, VGG16, and EfficientNet B0, to enhance the accuracy of plant disease identification. Each model was tested using the New Plant Diseases Dataset from Kaggle, which includes various plant species and diseases, in order to assess performance, accuracy, and efficiency. The trained models were subsequently integrated into a Marathi language chatbot to facilitate real-time disease detection and provide agricultural guidance. This study provides valuable insights into the strengths and limitations of different models for precision agriculture, especially in applications that support regional languages to encourage accessible and sustainable farming practices. Additionally, a Marathi language chatbot is incorporated, enabling users to obtain plant disease information instantly through a user-friendly web application.

Introduction

The rapid growth of precision agriculture has increased the demand for intelligent plant disease detection systems that support sustainable food production. Deep learning techniques, particularly Convolutional Neural Networks (CNNs), have significantly improved disease detection accuracy and efficiency. This study evaluates four CNN architectures—Basic CNN, AlexNet, VGG16, and EfficientNet B0—on a Raspberry Pi 5 edge computing platform to determine their suitability for real-time agricultural applications. To improve accessibility, the system also integrates a Marathi-language chatbot, enabling farmers to receive disease-related information and recommendations in their native language.

The research uses the New Plant Diseases Dataset from Kaggle, containing images of healthy and diseased leaves from crops such as tomato, potato, and corn. Images undergo preprocessing steps including resizing, normalization, noise reduction, and data augmentation to enhance model performance and reduce overfitting. The four CNN models are trained and evaluated based on accuracy, efficiency, and real-time deployment capability. To further improve reliability, an ensemble learning approach using soft voting combines predictions from multiple models.

The optimized ensemble model is converted to TensorFlow Lite and deployed on a Raspberry Pi 5. A connected camera captures leaf images, allowing real-time disease detection without relying on cloud connectivity. The Marathi chatbot provides disease diagnosis, treatment recommendations, preventive measures, and crop management guidance, making the system farmer-friendly and accessible in resource-constrained environments.

The study highlights the strengths of each CNN model. Basic CNN, AlexNet, VGG16, and EfficientNet B0 all achieve accuracy levels above 90%, with EfficientNet B0 offering the best balance between accuracy and computational efficiency for edge devices. A web-based interface enables users to select crop categories, upload images, and view disease detection results. The platform also supports multilingual output, improving accessibility for farmers from different linguistic backgrounds.

Existing plant disease detection systems face challenges such as poor image quality, high computational requirements, increased latency, and limited dataset diversity, which can reduce accuracy and practical usability. The proposed architecture addresses these issues by combining edge computing, deep learning, IoT-based image acquisition, cloud-supported visualization, and chatbot assistance into a single integrated framework.

Conclusion

The project develops an ensemble-based plant disease detection system using lightweight deep learning models on Raspberry Pi 5, optimizing performance with Mini-Tensor Flow for efficient real-time analysis. A comprehensive dataset enhances model accuracy, while a Marathi-language chatbot offers instant disease diagnosis [6] and management advice, improving accessibility for farmers. The system is designed to be scalable and user-friendly, ensuring usability and security in resource-constrained agricultural environments

References

[1] Rushikesh S. Tanksale and Sunil B. Mane, “Efficient Plant Disease Detection on RISC Devices: A comparison of Basic CNN, AlexNet, ResNet-50, and MobileNet Models using MiniTensorFlow,” International Journal of Intelligent Systems and Applications in Engineering, vol. 12, no. 16s, pp. 374–383, 2024. [2] Krizhevsky A, SutskeverI, Hinton GE. ImageNet Classification with Deep Convolutional Neural Networks. Advances in Neural Information Processing Systems 2012 [3] Food and Agriculture Organization (FAO), The Impact of Disasters and Crises on Agriculture and Food Security 2020. Food and Agriculture Organization of the United Nations. 2020. [4] M. Hughes, J. Salguero, R. Rodriguez, et al., “Deep Learning for Plant Disease Detection and Classification Using Field Images,” Computers and Electronics in Agriculture, vol. 183, 2021. [5] R A Boukabouya, A Moussaoui, and M Berrimi, “Vision Transformer Based Models for Plant Disease Detection and Diagnosis”. In: 2022 5th International Symposium on Informatics and its Applications (ISIA). 2022, pp. 1–6 [6] Chug, Anuradha Bhatia, Anshul Singh, Amit Singh, Dinesh, (2022). A novel framework for image based plant disease detection using hybrid deep learning approach. Soft Computing. 27. 10.1007/s00500-022-07177 7. 4 [7] Chen, J., Chen, J., Zhang, D. et al. A cognitive vision method for the detection of plant disease images. Machine Vision and Applications 32, 31 (2021) [8] M Shoaib et al. “An advanced deep learning models based plant disease detection: A review of recent research”. Frontiers in Plant Science 14 (2023), pp. 1158933–1158933. [9] J Zhang, H Xia, and J G Yang., “Deep Learning for Remote Sensing Data: A Technical Tutorial on the State of the Art”. IEEE Geoscience and Remote Sensing Magazine 9(2) 2021.

Copyright

Copyright © 2026 Ankit Raut, Omkar Aher, Rohan Chavan, Rushikesh Pawar, Prof. Nitin Zinzurke. 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.