Improved tomato leaf disease classification through adaptive ensemble models with exponential moving average fusion...

Area of Science:

• Agricultural Science
• Computer Science
• Plant Pathology

Background:

• Tomato yield is significantly impacted by various leaf diseases, necessitating early detection for effective management.
• Existing machine learning models struggle with accurate classification of novel tomato diseases.
• Deep learning combined with swarm intelligence offers enhanced accuracy for plant disease identification.

Purpose of the Study:

• To propose a novel ensemble deep learning model for accurate classification of tomato leaf diseases.
• To improve the accuracy and effectiveness of early disease detection in tomato plants.
• To enhance overall crop yield and provide better support for farmers through precise disease identification.

Main Methods:

• Developed an ensemble model integrating Visual Geometry Group-16 (VGG-16) and Neural Architecture Search Network (NASNet) mobile architectures.
• Incorporated an exponential moving average function with temporal constraints and an enhanced weighted gradient optimizer.
• Trained and validated the model on a dataset of 10,000 tomato leaf images across nine disease categories, with 1,000 images for testing.

Main Results:

• The proposed model achieved a high classification accuracy of 98.7%.
• Demonstrated superior performance in terms of precision (97.9%), recall (98.6%), and F1-score (98.7%).
• Achieved a low loss value of 4% and an exceptional receiver operating characteristic curve score of 99.97%.

Conclusions:

• The novel ensemble deep learning approach significantly improves tomato leaf disease classification accuracy.
• The method offers a robust solution for early and precise detection of various tomato plant diseases.
• This advancement has the potential to substantially benefit agricultural productivity and farmer livelihoods.