EARLY DETECTION OF YELLOW LEAF DISEASE IN SUGARCANE USING OPTIMIZED VISION TRANSFORMERS AND MACHINE LEARNING MODELS

Abstract

Background: Sugarcane cultivation is critical for agricultural economies, but diseases like Yellow Leaf Disease (YLD) pose significant threats to crop yield and quality. Early and accurate detection of YLD is essential for minimizing crop loss and implementing timely interventions. Problem: Detecting YLD in sugarcane leaves remains a challenging task due to the variability in symptoms, environmental factors, and the need for precise, efficient detection methods. Traditional diagnostic techniques often lack the necessary accuracy and scalability for large-scale detection in the field. Methods: This study proposes a system that integrates advanced image processing, optimization algorithms, and machine learning (ML) techniques to detect YLD in sugarcane varieties such as Co 18009, MS 14082, and CO 11015. Image data is combined with microclimatic information (e.g., temperature, humidity, and land temperature) to create a comprehensive analysis. Optimized Vision Transformers (OViTs), including HOA-ViT (Hummingbird Optimization Algorithm), PSO-ViT (Particle Swarm Optimization), and POA-ViT (Pelican Optimization Algorithm), are used to extract key features from leaf images. These features, along with statistical parameters, are analyzed using ML models like BO-ONN (Bayesian Optimization-Optimized Neural Network) and SLR (Stepwise Linear Regression) to classify and correlate disease patterns. Results: The system successfully detects various types of YLD, including Midrib Yellow Disease, Dry Midrib Disease, and Reddish Discoloration. Among the models evaluated, BO-ONN achieved the highest accuracy of approximately 99.50%, demonstrating superior performance in terms of sensitivity, specificity, and precision when compared to other models. The ONN's high accuracy allows for more reliable disease detection, ensuring early intervention. Significance: The integration of image processing, optimization algorithms, and ML models enhances the precision and reliability of YLD detection, making the system a powerful tool for early disease diagnosis. This approach can significantly reduce the impact of YLD on sugarcane crops, improving yield and quality. Conclusion: The proposed system demonstrates the potential of combining advanced techniques for efficient disease detection in agriculture. The high accuracy of BO-ONN highlights its effectiveness in real-world applications, providing farmers and agricultural experts with a reliable method for early detection and management of YLD in sugarcane crops.