Multi-Task NoisyViT for Enhanced Fruit and Vegetable Freshness Detection and Type Classification

Area of Science:

• Computer Vision and Artificial Intelligence
• Agricultural Technology
• Food Science and Technology

Background:

• Fruit and vegetable freshness is crucial for quality, nutrition, and waste reduction, necessitating accurate assessment methods.
• Traditional manual quality assessment is subjective and inefficient, driving the need for automated solutions.
• Imaging sensors combined with Artificial Intelligence (AI) offer promising avenues for objective and scalable quality monitoring.

Purpose of the Study:

• To evaluate the efficacy of the Noisy Vision Transformer (NoisyViT) model for automated fruit and vegetable freshness detection from images.
• To develop and assess a multi-task NoisyViT model for simultaneous freshness and type classification, enhancing generalization.
• To establish a robust and scalable AI solution for real-time quality assessment across the food supply chain.

Main Methods:

• The NoisyViT model was initially tested on five public datasets, achieving high accuracies for freshness detection.
• A unified dataset, Freshness44, was created by merging five datasets, comprising 44 classes across 22 fruit and vegetable types.
• The NoisyViT architecture was adapted into a multi-task configuration with separate classification heads for freshness (binary) and type (22-class) identification, fine-tuned on Freshness44.

Main Results:

• The single-head NoisyViT model demonstrated high accuracy (over 97%) on individual datasets.
• The multi-task NoisyViT model achieved exceptional accuracies of 99.60% for freshness detection and 99.86% for type classification on the Freshness44 dataset.
• The multi-task model outperformed the single-head NoisyViT and conventional machine learning/CNN-based methods in classification accuracy.

Conclusions:

• The multi-task NoisyViT model, trained on the comprehensive Freshness44 dataset, provides a highly effective and accurate solution for automated fruit and vegetable freshness detection.
• This AI-driven approach offers a scalable and robust system for real-time quality monitoring applicable in supply chains, retail, and consumer settings.
• The study highlights the potential of advanced AI architectures like NoisyViT for addressing critical challenges in food quality assessment and waste reduction.