SpinachXAI-Rec: a multi-stage explainable AI framework for spinach freshness classification and consumer...

Area of Science:

• Computer Science
• Food Science
• Artificial Intelligence

Background:

• Spinach is nutritious but highly perishable, leading to spoilage and health risks.
• Traditional freshness assessment methods are subjective, time-consuming, and unreliable.
• Defective consumption decisions impact human health and food safety.

Purpose of the Study:

• To develop an automated, AI-driven framework (SpinachXAI-Rec) for classifying spinach freshness.
• To provide interpretable AI and actionable consumer recommendations for spinach.
• To enhance food safety through accurate freshness validation.

Main Methods:

• A dataset of 12,000 spinach images across three varieties and six categories (fresh/non-fresh) was created.
• Deep learning models (DenseNet121, ResNet50, EfficientNetB0) were trained and evaluated.
• Explainable AI techniques (GradCAM++, LIME) and a rule-based recommender system were integrated.

Main Results:

• DenseNet121 achieved 96% classification accuracy in Stage 1.
• The combined DenseNet121 + ViT-B/16 + SVM model achieved an F1-score of 0.97 and high segmentation precision (Dice 0.89, IoU 0.82).
• The framework successfully categorized spinach into 'Eatable', 'Eatable with Caution', or 'Not Eatable'.

Conclusions:

• SpinachXAI-Rec offers an accurate and interpretable AI solution for spinach freshness assessment.
• The system empowers consumers and industry stakeholders with informed, health-conscious decisions.
• This advancement contributes to safer food systems through AI-driven validation and recommendations.