Improved multiscale attention based deep learning approach for automated sugarcane leaf disease detection using BSRI...

Area of Science:

• Agricultural Science
• Computer Science
• Biotechnology

Background:

• Sugarcane leaf diseases significantly impact crop yield and economic returns.
• Accurate and early disease detection is crucial for effective crop management and resource optimization.

Purpose of the Study:

• To develop and evaluate a novel deep learning model for automated sugarcane leaf disease classification.
• To improve the accuracy and efficiency of disease detection in sugarcane cultivation.

Main Methods:

• A Multi-scale Attention-based Dense Residual Network (MADRN) was designed, integrating dense residual learning and multi-scale attention.
• The MADRN model was trained and validated on two datasets: a public Kaggle dataset and a blended dataset including BSRI images.
• Preprocessing included resizing, normalization, and data augmentation; performance was compared against baseline models like CNN, VGG16, MobileNetV2, and XceptionNet.

Main Results:

• MADRN outperformed all baseline models across both datasets, achieving up to 94.78% accuracy on the Kaggle dataset and 92.25% on the blended dataset.
• The model demonstrated superior performance in accuracy, precision, recall, and F1-score, indicating its effectiveness in capturing disease-specific features.
• A web-based application was developed for real-time, user-friendly disease detection, facilitating practical implementation.

Conclusions:

• The MADRN model offers a robust and accurate solution for sugarcane leaf disease classification, outperforming existing methods.
• This deep learning approach shows significant potential for precision agriculture, enabling timely interventions and sustainable crop management.
• The developed tool provides a scalable and practical solution for disease management in the agricultural sector.