Task-Adaptive Angle Selection for Computed Tomography-Based Defect Detection.

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Sparse-angle X-ray Computed Tomography (CT) is crucial for industrial quality control, but faces a trade-off between scan time and reconstruction quality.
• Adaptive angle selection strategies aim to optimize information acquisition by considering object geometry.
• Deep Reinforcement Learning (DRL) has shown promise for adaptive angle selection in X-ray CT.

Purpose of the Study:

• To develop a task-adaptive angle selection method for sparse-angle X-ray CT.
• To improve defect detection in industrial quality control by optimizing projection angle selection.
• To introduce flexibility in the number of angles used for reconstruction.

Main Methods:

• Leveraged Deep Reinforcement Learning (DRL) for adaptive angle selection.
• Developed a task-adaptive strategy focusing on image-based defect detection.
• Incorporated prior knowledge of typical defect characteristics into the angle selection process.
• Enabled adaptability in the total number of projection angles used.

Main Results:

• The developed method enables easy detection of defects in reconstructed images.
• The task-adaptive approach enhances the effectiveness of sparse-angle X-ray CT for specific industrial applications.
• Achieved improved defect detection by optimizing projection angles based on downstream task requirements.

Conclusions:

• Task-adaptive angle selection using DRL offers a significant advancement for sparse-angle X-ray CT.
• This method enhances the utility of X-ray CT in industrial quality control by improving defect detection efficiency.
• The flexibility in angle count and task-specific optimization represent key innovations.