Anomaly Detection in Nanofibrous Materials by CNN-Based Self-Similarity

Area of Science:

• Materials Science
• Computer Vision
• Artificial Intelligence

Background:

• Automated anomaly detection in nanofibrous materials is crucial for cost reduction and efficient quality control.
• Scanning Electron Microscope (SEM) imaging is a key technology for inspecting these materials.
• Current visual inspection methods can be time-consuming and costly.

Purpose of the Study:

• To develop and validate a novel region-based method for automatic anomaly detection and localization in SEM images of nanofibrous materials.
• To improve the efficiency and accuracy of quality control in nanofiber production.

Main Methods:

• A region-based approach utilizing Convolutional Neural Networks (CNNs) and self-similarity.
• Evaluating subregion abnormality by computing CNN-based visual similarity against a dictionary of anomaly-free training data.
• Localization of detected anomalies within the SEM images.

Main Results:

• The proposed method demonstrates superior performance compared to existing state-of-the-art techniques.
• Accurate detection and localization of anomalies in SEM images were achieved.
• Significant reduction in production costs and post-production inspection time is anticipated.

Conclusions:

• The developed CNN-based self-similarity method offers an effective solution for anomaly detection in nanofibrous materials.
• This approach enhances the quality control process in the production of advanced materials.
• The method shows promise for industrial applications requiring high-throughput material inspection.