Optimized classification of potato leaf disease using EfficientNet-LITE and KE-SVM in diverse environments

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Accurate identification of potato foliar diseases is crucial for global food security and agricultural productivity.
• Traditional image classification methods struggle with inconsistent data from uncontrolled environments, limiting precision agriculture.
• Automated disease diagnosis using computer vision is essential for real-time decision-making in farming.

Purpose of the Study:

• To develop and evaluate a novel model for enhanced potato foliar disease identification.
• To improve the accuracy and efficiency of automated disease detection in variable field conditions.
• To create a computationally economical model suitable for deployment on resource-constrained devices.

Main Methods:

• Integration of EfficientNet-LITE for superior feature extraction with Channel Attention (CA) and 1-D Local Binary Pattern (LBP).
• Implementation of KE-SVM Optimization to iteratively refine classification accuracy by analyzing misclassified instances.
• Development of a compact model (12.46 MB, 3.11M parameters, 359.69 MFLOPs) for computational efficiency.

Main Results:

• The optimized model achieved 87.82% accuracy on uncontrolled data, a significant improvement from the pre-optimization 79.38%.
• Accuracy on laboratory-controlled data reached 99.54% post-optimization, compared to 99.07% previously.
• The model demonstrated enhanced emphasis on pertinent features while maintaining computational economy.

Conclusions:

• The novel model offers a practical and accurate solution for potato foliar disease detection in real-world agricultural settings.
• Its efficiency and accuracy make it ideal for precision agriculture, especially on mobile or edge devices with limited computational power.
• This advancement supports sustainable agriculture through improved crop health monitoring and management.