An approach for adaptive automatic threat recognition within 3D computed tomography images for baggage security...

Area of Science:

• Aviation Security
• Machine Learning
• Image Analysis

Background:

• Current aviation security relies on Automatic Threat Recognition (ATR) using 3D X-ray computed tomography (CT) images.
• Existing ATR methods struggle with adaptability to new and emerging threat signatures.
• Adaptive Automatic Threat Recognition (AATR) concepts were previously proposed to address these limitations.

Purpose of the Study:

• To present a practical solution for AATR using X-ray CT baggage scan imagery.
• To develop detection algorithms adaptable to diverse threat characteristics and evolving signatures.
• To enhance the adaptability of security scanners to varying threat materials and object properties.

Main Methods:

• A novel adaptive machine learning methodology was employed.
• The solution incorporates a multi-scale 3D CT image segmentation algorithm.
• A multi-class support vector machine (SVM) classifier was used for object material recognition, alongside an adaptability strategy.

Main Results:

• The proposed AATR approach demonstrated strong performance in both recognition and adaptation.
• The system achieved a probability of detection around 90%.
• A probability of false alarm below 20% was recorded.

Conclusions:

• The AATR system effectively adapts to various materials, including unknown ones not present in training data.
• The approach demonstrates adaptability to different required probabilities of detection.
• The system can also adapt to varying scales of threat objects.