A Focal Attention-Based Large Convolutional Kernel Network for Anomaly Detection of Coated Fuel Particles
The coating thickness of fuel particles is a critical parameter for ensuring the safe operation of high-temperature gas-cooled reactors. However, existing detection technologies still face limitations in measurement accuracy, efficiency, and automation. Notably, accurate thickness measurement relies...
Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2025-05-01
|
| Series: | Sensors |
| Subjects: | |
| Online Access: | https://www.mdpi.com/1424-8220/25/11/3330 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | The coating thickness of fuel particles is a critical parameter for ensuring the safe operation of high-temperature gas-cooled reactors. However, existing detection technologies still face limitations in measurement accuracy, efficiency, and automation. Notably, accurate thickness measurement relies on the precise identification of anomalous particles, which is hindered by several key challenges. First, incomplete particles in edge regions introduce significant interference. Second, some anomalies exhibit weak morphological features, making them difficult to detect. To address these issues, this study proposes an innovative focal attention-based large convolutional kernel network detection framework comprising three core modules. First, a Vision Transformer backbone incorporating a Large Selective Kernel Module dynamically adapts multi-scale receptive fields to enable coordinated global and local feature perception. Second, the Multi-Scale Feature Fusion Module establishes cross-layer feature interactions to enhance responses to subtle anomalies. Third, the Focal Attention Module employs a dynamic convolutional attention mechanism to strengthen the saliency representation of critical regions. Experimental results demonstrate the effectiveness of the proposed method, reducing the false detection rate and miss detection rate of anomaly detection to 1.96% and 1.9%, respectively. |
|---|---|
| ISSN: | 1424-8220 |