A Real-Time On-Device Defect Detection Framework for Laser Power-Meter Sensors via Unsupervised Learning

2025-09-25

AI Summary

To address the degradation of laser power meter sensor measurement accuracy caused by coating defects (e.g., thermal damage, scratches), this paper proposes an unsupervised, real-time visual inspection method that requires no defect annotations. The method is trained exclusively on normal samples and integrates Laplacian edge detection with K-means region segmentation; synthetic data augmentation is performed using StyleGAN2, and multi-scale feature extraction and anomaly map generation are achieved via the UFlow architecture. Its key innovation lies in establishing an end-to-end deployable anomaly detection framework capable of identifying both known and previously unseen defects. Evaluated on 366 real-world images, the method achieves 93.8% accuracy on defective samples and 89.3% on normal ones, with an image-level AUROC of 0.957 and an inference time of only 0.5 seconds per image—significantly enhancing inspection efficiency and robustness in industrial settings.

Technology Category

Computer Vision: Multi-modal VisionMachine Learning: Hardware-aware MLIntelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

Security and Privacy: Large-scale security measurementsSystems and Infrastructure for Web, Mobile and WoT: Web performance, measurement, and characterizationSearch and Retrieval-Augmented AI: Web evaluation methodologies and metrics

Abstract

We present an automated vision-based system for defect detection and classification of laser power meter sensor coatings. Our approach addresses the critical challenge of identifying coating defects such as thermal damage and scratches that can compromise laser energy measurement accuracy in medical and industrial applications. The system employs an unsupervised anomaly detection framework that trains exclusively on ``good'' sensor images to learn normal coating distribution patterns, enabling detection of both known and novel defect types without requiring extensive labeled defect datasets. Our methodology consists of three key components: (1) a robust preprocessing pipeline using Laplacian edge detection and K-means clustering to segment the area of interest, (2) synthetic data augmentation via StyleGAN2, and (3) a UFlow-based neural network architecture for multi-scale feature extraction and anomaly map generation. Experimental evaluation on 366 real sensor images demonstrates $93.8%$ accuracy on defective samples and $89.3%$ accuracy on good samples, with image-level AUROC of 0.957 and pixel-level AUROC of 0.961. The system provides potential annual cost savings through automated quality control and processing times of 0.5 seconds per image in on-device implementation.

Problem

• Detecting coating defects in laser power meter sensors automatically
• Identifying thermal damage and scratches affecting measurement accuracy
• Developing unsupervised defect detection without labeled defect datasets

Innovation

• Unsupervised anomaly detection using normal coating images
• Synthetic data augmentation via StyleGAN2 for training
• UFlow-based neural network for multi-scale feature extraction