Paper Surface Defect Detection

Paper Surface Defect Detection

Project For 
Published
Since 2023-05 till 2023-08
Paper Surface Defect Detection

A breakthrough AI project by Saiwa Inc. and AI-Innovate to automate surface defect detection on white and black paper with real-time precision and cost-saving benefits.

Challenge

Maintaining high-quality standards in paper manufacturing is no easy feat—especially when it comes to detecting subtle surface defects. Our client needed a solution that could accurately detect flaws in both white and black paper, two materials with vastly different visual properties.

Maintaining high-quality standards in paper manufacturing is no easy feat—especially when it comes to detecting subtle surface defects. Our client needed a solution that could accurately detect flaws in both white and black paper, two materials with vastly different visual properties.

paper-types

Our Solution

In collaboration with AI-Innovate, we developed an AI-powered defect detection system as part of our Anomaly Detection service. This cutting-edge solution combines deep learning with smart preprocessing to localize and classify paper surface defects in real-time.

Detection on White Paper

We used advanced machine learning models to detect:

  • Cracks

  • Frost marks

  • Mold oscillations

  • Longitudinal irregularities

Our custom-built SegDecNet model—a two-stage architecture for segmentation and classification—enabled precise detection.

white-paper

Detection on Black Paper

Detecting defects on black paper is inherently more difficult due to low contrast. To overcome this, we used our proprietary LLCC contrast enhancement algorithm, significantly improving visibility.

black-paper
Left: Original black paper – Right: Enhanced using LLCC.

Key Technical Highlights

  • Image Tiling:

    Images split into 1024×1024 px tiles for better accuracy and performance

  • Smart Labeling:

    Using Saiwa Annotation Services, we reduced manual labeling while maximizing data quality

  • Dual-Track Training:

    Separate AI models trained for white and black paper to address their unique characteristics

  • High-Speed Inference:

    Models converted to ONNX & TensorRT for GPU-accelerated real-time performance

 

saiwa-labeling
Cracks annotated using our custom labeling tool.

Results

  • Automated Quality Assurance:

    Significantly reduced manual inspection workload

  • Real-Time Visualization:

    Easy-to-use UI for quick defect review and decision-making

  • Cost Savings:

    Reduced production downtime and material waste

Why Saiwa?

At Saiwa, we specialize in AI solutions for industrial challenges. Our defect detection technologies are:

  • Fast: Capable of processing high volumes of data with minimal delay

  • Scalable: Adaptable to various materials—paper, metal, textiles, plastics

  • Customizable: Integrates seamlessly into your workflow with APIs and intuitive interfaces

 

Let’s Work Together

If your business depends on speed, accuracy, and consistent quality, Saiwa can help.

Contact us: support@saiwa.ai

Share: