Unsupervised Autoencoder Approach for Precise Line-Type Mura Detection and Classification

Mura refers to surface defects or uneven brightness in panel manufacturing and is classified by severity into light Mura and serious Mura. Due to limited data, traditional object detection is not feasible. Instead, we propose an unsupervised method to classify serious Mura and accurately localize de...

Full description

Saved in:
Bibliographic Details
Published in:IEEE International Conference on Consumer Electronics-China (Online) pp. 507 - 508
Main Authors: Chang, Ting-Yu, Lin, Chia-Yu
Format: Conference Proceeding
Language:English
Published: IEEE 16.07.2025
Subjects:
Anomaly localization
Autoencoder
Autoencoders
Computational modeling
Consumer electronics
Defect detection
Inspection
Location awareness
Manufacturing
mura detection
Object detection
Quality control
Unsupervised learning
ISSN:2575-8284
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Mura refers to surface defects or uneven brightness in panel manufacturing and is classified by severity into light Mura and serious Mura. Due to limited data, traditional object detection is not feasible. Instead, we propose an unsupervised method to classify serious Mura and accurately localize defects. We combine an autoencoder with computer vision to simulate a supervised model. This approach not only improves defect reconstruction quality but also achieves 90% precision while improving recall by 30%. Our method enhances defect detection accuracy, providing a data-efficient, scalable solution for quality control in panel manufacturing
ISSN:2575-8284
DOI:10.1109/ICCE-Taiwan66881.2025.11207809