Flexible Inspection of Defects in Liquid Crystal Display Panels: A Review

As the liquid crystal display (LCD) manufacturing industry transitions toward a high-mix, low-volume (HMLV) production model, traditional defect detection methods struggle to adapt due to their limited generalization capabilities and heavy reliance on extensive labeled data. This paper systematically reviews the research progress of flexibility-oriented intelligent inspection algorithms. The article first establishes a theoretical foundation covering the LCD manufacturing process, defect genesis mechanisms, and Mura defect quantification, while emphasizing the prerequisite role of imaging hardware. Subsequently, it provides an in-depth analysis of five key technologies—meta-learning, transfer learning, zero-shot learning, multimodal fusion, and data augmentation—elucidating their respective unique advantages in addressing challenges such as data scarcity, feature heterogeneity, and unknown defect recognition. Synthesizing the characteristics of these key technologies, this paper summarizes and proposes a Hierarchical Flexible Inspection Framework (HFIF). This framework maps discrete algorithmic logic into three distinct layers: Data Foundation, Feature Representation, and Task Adaptation. Furthermore, addressing the limitations of traditional static metrics, this paper establishes a dynamic evaluation standard that incorporates metrics such as the Changeover Stability Index (CSI). Finally, the paper discusses future evolutionary directions, providing a systematic theoretical reference and technical pathway for constructing the next generation of highly adaptive intelligent LCD inspection systems.