Enhancing rice crop health: a light weighted CNN-based disease detection system with mobile application integration

• Published in: Multimedia tools and applications 2024-05, Vol.83 (16), p.48799-48829
• Main Authors: Tyagi, Shivangi, Reddy, S.R.N., Anand, Rishika, Sabharwal, Aditi
• Format: Article
• Language: English
• Subjects: Accuracy; Applications programs; Computer Communication Networks; Computer Science; Crop diseases; Cultivation; Data Structures and Information Theory; Datasets; Image classification; Image contrast; Image quality; Medical imaging; Mobile computing; Model accuracy; Multimedia Information Systems; Plant diseases; Rice; Special Purpose and Application-Based Systems
• ISSN: 1573-7721; 1380-7501; 1573-7721
• DOI: 10.1007/s11042-023-17449-5

This article introduces an advanced approach for the accurate detection of rice leaf diseases, a critical concern in rice cultivation worldwide. Rice is a staple crop in many countries, and diseases can inflict severe damage on yields. Timely and precise detection is essential for effective disease management. The proposed model offers a solution to enhance detection accuracy while simplifying the process, utilizing a dataset of rice leaf images obtained from Kaggle.com. The dataset, though valuable, presented challenges due to image quality issues, including noise and inadequate clarity. To address these challenges, a two-phase approach was developed. In the pre-processing phase, the Contrast Limited Adaptive Histogram Equalization (CLAHE) technique was applied to adjust image illumination, contrast, and edge sharpness. This significantly improved image quality and paved the way for more accurate disease detection. In the segmentation phase, a hybrid methodology combining HSV and K-means segmentation techniques was employed. This innovative fusion of techniques effectively extracted the Region of Interest (ROI) from images, focusing exclusively on disease-related features, which was vital for precise detection. The core of the system lies in a Light Weighted CNN model, renowned for its efficiency and accuracy. This model was employed to classify rice leaf images, achieving exceptional results. To make this powerful tool accessible and user-friendly, an Android application was developed, enabling users to easily navigate and identify diseases in rice leaves. To validate the effectiveness of the proposed model, comprehensive comparisons were made with standard models such as simple CNN, InceptionResNetV2, MobileNet, and DenseNet. Across all evaluation metrics, including accuracy, precision, recall, and F-score, the proposed model consistently outperformed traditional counterparts, boasting an impressive 99% accuracy rate. This article presents an innovative and practical solution to address a pressing issue in rice cultivation, offering the potential to significantly improve crop health and agricultural outcomes.