Early stage black pepper leaf disease prediction based on transfer learning using ConvNets.

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Crop diseases cause significant agricultural damage, necessitating early detection methods.
• Black pepper, a valuable medicinal plant, is susceptible to various leaf diseases.
• Computer vision systems offer potential for timely disease diagnosis and prevention.

Purpose of the Study:

• To develop an intelligent transfer learning technique for predicting prominent black pepper leaf diseases.
• To implement a state-of-the-art deep learning model, specifically a convolutional neural network, for disease identification.
• To enhance agricultural practices through early and accurate disease prediction.

Main Methods:

• Utilized transfer learning with a deep neural network trained on the ImageNet dataset.
• Developed a new dataset of real-time black pepper leaf images, annotated by experts.
• Trained and evaluated multiple deep learning models including Inception V3, GoogleNet, SqueezeNet, and Resnet18.
• Optimized hyperparameters such as learning rate, optimization algorithm, and batch size.

Main Results:

• The Resnet18 model achieved the highest accuracy of 99.67%.
• All evaluated models demonstrated high validation accuracy, ranging from 99.1% to 99.7%.
• Low validation loss was observed across the models, indicating effective learning.

Conclusions:

• The proposed deep learning approach significantly improves early-stage leaf disease identification in black pepper.
• This cutting-edge method offers a valuable tool for agricultural disease management.
• Accurate and timely disease prediction can lead to targeted prevention strategies and reduced crop loss.