Abstract
Artificial intelligence (AI) and computer vision (CV) are increasingly advancing food quality inspection, safety assessment, and process automation. However, translating laboratory-based research into reliable and scalable industrial applications remains challenging. This review critically evaluates progress in AI- and CV-based inspection of processed foods, focusing on machine learning and deep learning methods for automated defect detection and classification, quality grading, compositional prediction, contamination assessment, and real-time production-line inspection. This review also covers imaging and sensing methods, including RGB, multispectral, hyperspectral, near-infrared, thermal, and multimodal systems, with a special focus on data fusion for non-destructive food evaluation. The review also examines key challenges to commercial application, including limited and non-diverse datasets, domain shift, environmental unpredictability, computing constraints, scalability, system integration, and model validation. The review also discusses emerging directions in multimodal AI, transformer architectures, edge computing, explainable and self-supervised learning, robotics, Internet of Things (IoT), and digital twins, and their potential to enable intelligent and adaptable food inspection and manufacturing. This review’s novelty lies in its task-oriented integration of imaging and sensing modalities with AI methods for processed-food inspection, along with a critical assessment of model validation, robustness, and the gap between laboratory-reported performance and practical production-line requirements.
Funding Information
  • Qassim University (QU-APC-2026)

This publication has 100 references indexed in Scilit: