Plant Disease Classification: A Comparative Evaluation of Convolutional Neural Networks and Deep Learning Optimizers

Area of Science:

• Agricultural Science
• Computer Science
• Artificial Intelligence

Background:

• Plant disease classification is crucial for crop yield and food security.
• Deep learning (DL) models have shown promise in automating plant disease identification.

Purpose of the Study:

• To conduct a comparative evaluation of various deep learning architectures for plant disease classification.
• To identify the optimal deep learning model and optimizer for accurate plant disease detection.

Main Methods:

• Comparative analysis of well-known and modified Convolutional Neural Network (CNN) architectures.
• Performance evaluation based on validation accuracy, loss, F1-score, and epochs.
• Training DL models using Keras with TensorFlow backend on the PlantVillage dataset.

Main Results:

• The Xception architecture, trained with the Adam optimizer, achieved the highest validation accuracy (99.81%) and F1-score (0.9978).
• This performance surpasses previous approaches in plant disease classification accuracy.

Conclusions:

• The proposed deep learning approach, specifically the Xception model with Adam optimizer, is highly effective for plant disease classification.
• This method demonstrates novelty and potential for broader agricultural applications requiring transparent detection.