An Interpretable Real-Time Plant Disease Detection and Severity Assessment System Based on Optimized CNN Activation function

Abstract

Plant diseases have a significant impact on agricultural productivity and food security worldwide, making timely disease detection techniques essential. Convolutional neural networks (CNNs) have shown promise in plant disease classification, but existing methods primarily use supervised data, are difficult to interpret, and cannot forecast disease severity or offer timely predictions. This paper presents a novel method for real-time plant disease detection, offering interpretability and severity estimation using a lightweight convolutional neural network (CNN) combined with an op- timization algorithm. The proposed activation function improves convergence speed and mitigates the dead neuron problem associated with ReLU activation. Grad-CAM, an AI-based interpretable analysis tool, is used to detect disease foci, and subsequent clustering determines disease sever- ity. Furthermore, the model, designed to support rational decision-making regarding agricultural crops, incorporates fertilization and treatment strategies based on the severity of the problems. Ex- perimental evaluations demonstrate greater accuracy, faster learning speed, and better results than existing methods based on convolutional neural networks.

Introduction

Agriculture plays a vital role in ensuring food security and economic stability, but plant diseases caused by pathogens such as viruses, fungi, and bacteria lead to significant crop losses worldwide. Traditional disease detection methods rely on manual inspection and expert analysis, which are slow, labor-intensive, and unsuitable for large-scale agricultural production. Recent advances in deep learning, particularly Convolutional Neural Networks (CNNs), have enabled automated plant disease detection using leaf images. While many CNN-based models achieve high accuracy, existing approaches often rely on limited datasets, lack real-time capabilities, and do not provide explainability or severity assessment.

To address these limitations, the study proposes a comprehensive AI-based framework that combines optimized CNN models, explainable AI techniques, disease severity estimation, and real-time processing. The main contributions include developing an optimized CNN learning algorithm, designing a lightweight model suitable for real-time edge deployment, integrating Grad-CAM for disease localization, estimating disease severity, and providing decision support for precision agriculture.

Previous research has shown the effectiveness of CNNs for plant disease classification using datasets such as PlantVillage, but many models lack robustness in real-world environments. Activation functions like ReLU are commonly used but can cause issues such as neuron death. Explainable AI methods like Grad-CAM help visualize important regions in images, though their use in plant disease detection remains limited. Additionally, most studies do not address disease severity estimation or provide actionable recommendations for farmers.

The proposed system includes six major modules: image acquisition, preprocessing, optimized CNN-based classification, explainability, disease severity estimation, and decision support. Leaf images are captured using mobile cameras and preprocessed through resizing, noise filtering, lighting normalization, and data augmentation. A lightweight CNN model such as MobileNet or EfficientNet-Lite is used for real-time inference, enhanced with a hybrid activation function to improve convergence and prevent neuron inactivity.

For interpretability, Grad-CAM highlights infected areas on leaf images, allowing users to understand the model’s predictions. The highlighted regions are further analyzed using K-means clustering to estimate disease severity by calculating the percentage of infected areas, which are categorized as mild (<20%), moderate (20–50%), or severe (>50%).

Finally, the system includes a decision support module that recommends appropriate fertilizers and pesticides based on the detected disease and severity level. This helps farmers apply treatments more efficiently while reducing excessive chemical usage.

Overall, the proposed framework provides a real-time, explainable, and severity-aware plant disease detection system that supports precision agriculture and improves crop management through automated analysis and actionable recommendations.

Conclusion

This paper presents an innovative, interpretable, and robust method for predicting the severity of plant diseases in real time based on deep convolutional neural network architectures. By inte- grating explainable AI and a decision support system for agriculture, the proposed system im- proves transparency, ease of use, and efficiency. Future work will focus on crop augmentation, transformer-based power grid construction, and mobile applications.

References

[1] S. Yegneshwar Yadhav, T. Senthilkumar, S. Jayanthy, and J. Judeson Antony Kovilpillai, “Plant Disease Detection and Classification using CNN Model with Optimized Activation Function,” International Journal / Conference Proceedings, 2021. [2] “Plant Disease Detection and Classification Using a Deep Learning-Based Framework,” in Proceedings of the International Conference on Intelligent Systems, Springer, pp. 39–50, 2023. [3] Y. LeCun, Y. Bengio, and G. Hinton, “Deep learning,” Nature, vol. 521, no. 7553, pp. 436– 444, 2015. [4] X. Glorot, A. Bordes, and Y. Bengio, “Deep sparse rectifier neural networks,” in Proceedings of the 14th International Conference on Artificial Intelligence and Statistics (AISTATS), pp. 315–323, 2011. [5] R. R. Selvaraju et al., “Grad-CAM: Visual explanations from deep networks via gradient- based localization,” in Proceedings of the IEEE International Conference on Computer Vision (ICCV), pp. 618–626, 2017. [6] J. MacQueen, “Some methods for classification and analysis of multivariate observations,” in Proceedings of the Fifth Berkeley Symposium on Mathematical Statistics and Probability, 1967.

Copyright

Copyright © 2026 Khushboo Verma, Payal Goviindani, Neha Kumari . 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.