Threat Recognition in X-Ray Scans of Luggage
The project
While postdoc at Duke University from 2017-2019, I led the second and third phases of a major research initiative on automatic threat recognition in X-ray scans of luggage. This was sponsored by the TSA, and under PI Larry Carin.
Results and Publications
We achieved breakthroughs in the first realistic application of Deep Learning to the real-world task of threat recognition in Luggage. Additionally, we advanced the science of real-world threat recognition by proposing a novel scheme for improving threat recognition by doing semi-supervised training with real-world data.
Domain-adaptive, Semi-supervised Threat Recognition
During development of these object detection systems, we discovered a quirk in straightforward use of real-world data to improve threat recognition. The capacity of the convolutional feature extractor was enough to learn the differences between real-world and staged data by unmeaningful signatures such as noise. To overcome this, we developed a training procedure based on Domain Adaptive Neural Networks for semi-supervised threat recognition.
2020
- Background Adaptive Faster R-CNN for Semi-Supervised Convolutional Object Detection of Threats in X-Ray ImagesIn Proc.SPIE, 2020
Smiths
With the team from Smiths, Inc in Wiesbaden, Germany, we published the following collaboration:
2018
- Automatic threat recognition of prohibited items at aviation checkpoint with x-ray imaging: a deep learning approachIn Proc.SPIE – Invited Paper, 2018
Rapiscan
With Rapiscan, Inc. of Fremont CA, we published the following collaboration:
2019
- Toward Automatic Threat Recognition for Airport X-ray Baggage Screening with Deep Convolutional Object DetectionAdvances in X-ray Analysis, Volume 64, proceedings of the 2020 Denver X-ray Conference, 2019