Mura defect detection method, device, equipment, storage medium and program product

By using a cognitive decision-making model based on a large language model and a quantitative scoring feedback iterative correction operator call sequence, the problem of poor flexibility and black box in Mura defect detection is solved, achieving high accuracy, interpretability and scenario adaptability, and is suitable for Mura defect detection in panel manufacturing process.

CN122156104APending Publication Date: 2026-06-05ZHONGJIA MICROVISION (SHENZHEN) SEMICONDUCTOR TECHNOLOGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
ZHONGJIA MICROVISION (SHENZHEN) SEMICONDUCTOR TECHNOLOGY CO LTD
Filing Date
2026-02-13
Publication Date
2026-06-05

AI Technical Summary

Technical Problem

In existing technologies, Mura defect detection schemes are inflexible. Traditional fixed-process schemes cannot adapt to different image qualities and types of Mura defects. Parameter adjustment relies on human experience, which is inflexible and prone to missed or over-detection. The end-to-end neural network scheme is a black box model, and the detection results lack interpretability. It is difficult to trace the logic and parameter basis of defect detection, and it requires a large amount of labeled data for training, which is not adaptable to unlabeled scenarios.

Method used

A cognitive decision-making model based on a large language model is adopted. By calculating the global feature parameters of the original image, a structured text description is generated. Combined with the neural symbol characteristics, a detection process is generated. The operator call sequence and parameters are iteratively corrected using quantitative scoring feedback to achieve adaptive optimization and result traceability, and dynamically adapt to different image qualities and types of Mura defects.

Benefits of technology

It improves the accuracy, interpretability, and scene adaptability of Mura defect detection, solves the problems of poor flexibility of traditional solutions and black box problem of neural network solutions, and can effectively cope with complex environmental interference in unlabeled scenarios, ensuring the quality control effect of panel manufacturing process.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122156104A_ABST
    Figure CN122156104A_ABST
Patent Text Reader

Abstract

The application discloses a Mura defect detection method, device, equipment, storage medium and program product. The method comprises the following steps: acquiring an original image of a to-be-detected panel, calculating a global feature parameter of the original image, and generating a structured text description based on the global feature parameter; inputting the structured text description into a cognitive decision model to infer an initial detection process, wherein the initial detection process comprises a calling sequence of atomized image processing operators and corresponding operator parameters; calling a corresponding atomized image processing operator in an algorithm skill library according to the initial detection process, processing the original image through the atomized image processing operator, obtaining a preliminary Mura defect detection result, and quantitatively scoring the preliminary Mura defect detection result to obtain a scoring feedback; iteratively correcting the detection process output by the cognitive decision model according to the scoring feedback until a preset iteration termination condition is met, and then outputting a target Mura defect detection result.
Need to check novelty before this filing date? Find Prior Art