Robust CRW crops leaf disease detection and classification in agriculture using hybrid deep learning models

Area of Science:

• Agricultural Science
• Computer Science
• Artificial Intelligence

Background:

• Plant diseases significantly impact crop yield and quality, necessitating effective diagnostic tools.
• Current crop-specific machine learning (ML) and deep learning (DL) models are often impractical for resource-poor farmers with low digital literacy.
• There is a need for versatile and accessible plant disease detection systems.

Purpose of the Study:

• To develop and evaluate a novel Slender-Convolutional Neural Network (CNN) model for multi-crop plant disease detection.
• To address the limitations of crop-specific models by creating a generalized solution for corn, rice, and wheat.
• To provide an accurate and efficient tool for on-farm disease identification, even in resource-limited settings.

Main Methods:

• A Slender-CNN architecture was designed with parallel convolution layers of varying dimensions for multi-scale lesion localization.
• The model was trained and validated on datasets of corn, rice, and wheat crops, including healthy and infected samples.
• Performance was benchmarked against established CNN models like VGG19, EfficientNetb6, and YOLOv5.

Main Results:

• The Slender-CNN model achieved an overall accuracy of 88.54%, outperforming several benchmark models.
• The model demonstrated high accuracy in classifying individual crop types: 99.81% for corn, 87.11% for rice, and 98.45% for wheat.
• The proposed network showed superior performance compared to VGG19, EfficientNetb6, ResNeXt, DenseNet201, AlexNet, YOLOv5, and MobileNetV3.

Conclusions:

• The developed Slender-CNN model offers an effective and accurate solution for multi-crop plant disease detection.
• Its compact design and high performance make it suitable for deployment in resource-limited agricultural settings.
• The model's versatility and accuracy support improved on-farm disease management and crop production.