AI-powered automatic detection and object recognition

An X-ray image shows the inside of an object; interpreting it is a separate skill. RöntgenTek adds an automatic detection and object recognition layer on top of the dual-energy image-processing pipeline: classical computer vision combined with machine learning / deep learning. From operator decision support to fully automatic decisions — across security screening, food safety and non-destructive testing (NDT).

Let's discuss your project →

Application areas

One recognition core, three domains

The same foundation — recognizing objects and materials in an X-ray image — applies across three industries. Each domain is customized with its own data and rules.

🛡️

Security screening — threat & prohibited-item detection

In conveyor X-ray scanners we provide decision support to the operator: automatic flagging of weapons, knives, dangerous and prohibited items. Using the material information from dual energy (organic / inorganic / metal), we highlight suspicious regions, focus the operator's attention and reduce human error.

🥫

Food inspection — foreign bodies & quality

In-line X-ray inspection for food safety: detection of foreign bodies such as metal, glass, stone, bone and dense plastic; fill / missing-piece and packaging-integrity checks. Unlike optical systems, X-ray sees inside the packaging and the product.

🔬

NDT — non-destructive testing

Automatic detection and classification of defects, voids, cracks and assembly faults in industrial parts. Consistent, repeatable decisions in series production — faster and more objective than manual inspection.

Our approach: image processing + machine learning

We build AI on top of our existing image-processing expertise. Classical computer vision (segmentation, feature extraction) provides a reliable, explainable baseline; deep learning captures the complex patterns where classical methods struggle. We combine the two in the proportion the task requires.

Classical CV + deep learning — segmentation and feature analysis together with CNN-based detection / classification.
Domain-specific data — sample collection, labelling, model training and independent validation.
Real-time inference — at production-line or conveyor speed, without latency.
On-prem / edge deployment — data stays on site, no cloud dependency.
Integration with the existing pipeline — natural fit into the dual-energy image-processing pipeline and operator interface.
Decision support or fully automatic mode — the false-alarm vs. miss trade-off is tuned to the task.
Continuous improvement — retraining on new samples and monitoring of performance.

An honest approach. AI is not a magic box. We validate feasibility with a pilot and real test samples; the result is driven by data quality and the X-ray contrast between materials. We set realistic expectations from the outset.

How it fits into a system

The recognition layer does not work in isolation; it combines with our other competencies into an end-to-end solution:

Dual energy image processing — produces the image and the material information (Zeff, density).
Detector integration — provides a stable, synchronized data stream.
Controller — triggers mechanisms such as alarms, light/sound warnings or pneumatic ejection based on the detection result (in food and NDT lines).

Ways of working

Feasibility & pilot — assessing X-ray contrast and detectability with test samples.
Model development — data collection / labelling, training and validation, definition of target metrics.
Line integration — deploying the model into the existing system with real-time inference.
Support & improvement — ongoing retraining and maintenance with field samples.

Contact

Let's assess your recognition task together

Tell us which objects or defects you need to detect and which equipment you work with — we'll assess feasibility and propose a pilot and a roadmap. If possible, prepare test samples.

Get in touch →