Sugarcane leaf disease classification using deep neural network approach

Area of Science:

• Agricultural Science
• Computer Science
• Machine Learning

Background:

• Manual disease diagnosis in sugarcane is time-consuming and prone to errors.
• Accurate disease detection is crucial for effective crop management and yield optimization.

Purpose of the Study:

• To develop and evaluate deep learning (DL) models for automated sugarcane leaf disease diagnosis.
• To compare the performance of various Convolutional Neural Network (ConvNet) architectures for disease classification.

Main Methods:

• Trained and tested EfficientNet, DenseNet201, ResNetV2, InceptionV4, MobileNetV3, and RegNetX models on the Sugarcane Leaf Dataset (SLD) with 6748 images.
• Utilized 70% training, 15% validation, and 15% testing data splits, supplemented by 5-fold cross-validation for robust evaluation.
• Assessed models based on accuracy, complexity, and depth.

Main Results:

• EfficientNet-B7 achieved 99.79% accuracy, and DenseNet201 achieved 99.50% accuracy, outperforming other tested models.
• 5-fold cross-validation confirmed the reliability and consistency of the top-performing models.
• No direct correlation was found between model complexity/depth and accuracy, highlighting the importance of dataset adaptability.

Conclusions:

• Deep learning models, specifically EfficientNet-B7 and DenseNet201, offer a highly effective solution for rapid and accurate automated disease detection in sugarcane.
• These DL systems significantly enhance traditional manual diagnosis, enabling timely interventions to reduce crop loss and improve sugarcane production.
• The study underscores the transformative potential of DL applications in modern agriculture for disease management and yield enhancement.