Explainable transformer framework for fast cotton leaf diagnostics and fabric defect detection.

Area of Science:

• Artificial Intelligence
• Computer Vision
• Machine Learning

Background:

• Accurate identification of cotton leaf diseases and fabric defects is crucial for agricultural yield and textile quality.
• Existing AI models often lack interpretability or computational efficiency.

Purpose of the Study:

• To develop a hybrid deep learning model combining CNNs and Vision Transformers for classifying cotton leaf diseases and fabric defects.
• To enhance model interpretability using Explainable AI (XAI) techniques.
• To achieve high accuracy and computational efficiency in AI-driven quality assessment.

Main Methods:

• A hybrid deep learning model integrating CNN-based hierarchical feature extraction and Vision Transformer self-attention (XCottL-FebViT).
• Application of Explainable AI (XAI) for enhanced model interpretability.
• Hyperparameter optimization for computational efficiency.
• Evaluation on four benchmark datasets: CottonLeafNet, SAR-CLD, CottonFabricImageBD, and FabricSpotDefect.

Main Results:

• XCottL-FebViT demonstrated superior performance compared to existing transformer-based models.
• Achieved high training and validation accuracies (e.g., 99.97% training and 99.93% validation for CottonLeafNet).
• Consistent improvements in accuracy, MCC, and F1 Score across datasets.

Conclusions:

• The proposed XCottL-FebViT model offers a highly accurate, interpretable, and computationally efficient solution for detecting cotton leaf diseases and fabric defects.
• The integration of XAI facilitates better understanding and trust in AI decisions for domain experts.
• A practical web-based application enables remote deployment for real-world quality assessment in agriculture and textiles.