AI transformation · Aviation security
Training data that makes airport screening smarter
Client: A national aviation security authority (name withheld)
90%
model accuracy achieved using the delivered training dataset
Challenge
Detecting prohibited items in X-ray luggage scans depends on the training data behind the model, and off-the-shelf datasets do not reflect the items and imaging conditions a real security team faces.
Approach
Built a high-precision, pixel-level annotated dataset of X-ray scans, labeling firearms, knives, cutters and other prohibited items for YOLOv5 and YOLOv8 object detection, run entirely on the client's own servers to meet strict data security requirements.
Outcome
Delivered a production-grade labeling and QA pipeline that contributed directly to the detection model's accuracy, on a restricted network with no external data egress.
- YOLOv8
- Computer vision
- On-prem
