Optimizing Rice Health: A Comparative Evaluation of Pretrained and Custom Convolutional Neural Networks for Disease Recognition

Aunik Hasan Mridul; Jannatul Fardus Armin; Md. Abu Saleh; Akash Das; Md. Mahidul Islam; Sujan; G. M. Shakil; Md. Abdur Rakib

doi:10.47852/bonviewAIA52026109

Authors

Aunik Hasan Mridul Department of Computer Science and Engineering, Daffodil International University, Bangladesh
Jannatul Fardus Armin Department of Computer Science and Engineering, Daffodil International University, Bangladesh
Md. Abu Saleh Department of Computer Science and Engineering, Daffodil International University, Bangladesh
Akash Das Department of Computer Science and Engineering, Brainware University, India
Md. Mahidul Islam Department of Software Engineering, Daffodil International University, Bangladesh
Sujan Department of Computer Science and Engineering, Rangamati Science and Technology University, Bangladesh
G. M. Shakil Department of Computer Science and Engineering, Patuakhali Science and Technology University, Bangladesh
Md. Abdur Rakib Department of Computer Science and Engineering, Daffodil International University, Bangladesh

DOI:

https://doi.org/10.47852/bonviewAIA52026109

Keywords:

custom model, pretrained model, rice leaf, explainable AI (XAI)

Abstract

The most commonly used staple in the world, rice, is integral to daily living and working. Among diseases that attack rice plants, bacterial leaf blight, blast, brown spot, and false smut are the diseases affecting both the agricultural productivity and the quality of life of the people relying on it. Pesticides that are used in large quantities to treat these diseases are harmful to human health and disturb the natural balance. This study focuses on an automated machine learning-based approach that applies image processing techniques for accurate detection and classification of common rice leaf diseases. Convolutional neural networks were employed for high-accuracy classification, given that the processing, such as feature identification and noise reduction, was carried out within preprocessing phases. Our proposed model performed better compared to traditional deep learning architectures with an accuracy of 97.68% and precision, recall, and F1-score of 98%. In addition, we further examined the model using explainable AI. The results provide a cost-effective, scalable solution for early disease diagnosis with the ability to enable management to reduce crop loss. This study underscores the transformative potential of AI-powered precision agriculture technology in revolutionizing sustainable agriculture practices and boosting food security.

Received: 8 May 2025 | Revised: 22 September 2025 | Accepted: 12 November 2025

Conflicts of Interest

The authors declare that they have no conflicts of interest to this work.

Data Availability Statement

The data that support the findings of this study are openly available in Kaggle at https://www.kaggle.com/datasets/anshulm257/rice-disease-dataset.

Author Contribution Statement

Aunik Hasan Mridul: Conceptualization, Methodology, Software, Validation, Writing – review & editing, Supervision, Project administration. Jannatul Fardus Armin: Conceptualization, Methodology, Validation, Data curation, Writing – original draft, Writing – review & editing. Md. Abu Saleh: Writing – original draft, Visualization. Akash Das: Formal analysis, Investigation, Data curation, Writing – original draft, Visualization. Md. Mahidul Islam: Methodology, Software, Formal analysis, Investigation, Writing – original draft, Writing – review & editing. Sujan: Formal analysis, Investigation. G. M. Shakil: Formal analysis, Investigation, Resources. Md. Abdur Rakib: Software, Resources.