Few-shot defect detection in industrial scenarios: A comprehensive review of challenges, advances, and frontier trends

Few-shot learning (FSL) has emerged as a transformative paradigm in industrial defect detection, enabling robust generalization from limited prior defect-related experience. This review highlights key challenges in FSL for defect detection, including data scarcity, representation bottlenecks in capturing subtle defects, cross-domain generalization barriers, and inherent trade-offs among sample efficiency, model generalization, and computational feasibility. We systematically explore recent advancements in FSL methodologies—such as meta-learning frameworks, generative augmentation, attention-driven architectures, and domain adaptation techniques—analyzing their effectiveness in addressing these challenges. Finally, we outline future research directions, emphasizing pathways to integrate FSL into intelligent manufacturing inspection systems through unified frameworks, human-AI collaboration, and standardized benchmarking. This synthesis provides actionable insights for advancing resource-efficient defect detection in industrial applications.