HxGN Visual Detection: Machine Learning

AI-powered surface inspection

Product enquiry

Visual detection shown on monitor

HxGN Visual Detection uses CNN (convolutional neural network) deep learning models to train, identify and categorise surface defects. It requires a relatively small quantity of defects samples (in the order of hundreds) to start. As it uncovers undetected defects, the model can be trained to further its ability to identify true positives.

The algorithms in HxGN Visual Detection employ pattern recognition, statistics, deep learning and other image processing techniques to rapidly learn about surface deviations. 

The HxGN Visual Detection machine learning process

1. Label

Upload a range of training set images from your vision system and annotate the defects with a classification such as scratch or dent. Create as many defect types as you are likely to encounter during any production run.

2. Train

Deep learning models optimise the decision paths created with a training set, called supervised learning. The training sessions take just a few minutes or hours and then you’re all set.

3. Inspect

Identify defects within images supplied by vision systems, X-ray machines or images from any local folder. Advanced GPU processing algorithms optimise inspection speed. HxGN Visual Detection provides a full inspection report and can also provide notifications via PC-DMIS.

Deep learning drives greater accuracy

Successful deep learning relies on the depth and sensitivity of the learning model and its environment. HxGN Visual Detection includes a set of models (X-CNN, X-CNN-Tiny, X-CNN-Plus and Segmentation) which are optimised for performance and accuracy. 

The construction of an advanced neural network increases HxGN Visual Detection’s ability to quickly and correctly identify defects, also known as true positives. Powerful algorithms including synthetic image data augmentation, image morphing, normalisation and dimensionality reduction all contribute to this process.

The foundations of these models are based on accepted industry frameworks such as GAN, YOLO, and RCNN.  

HxGN Visual Detection provides an open architecture design that can be easily integrated with custom CNN models. The application is extendable through an API and can be integrated into third party automated manufacturing cells.

  Advanced neural networks graphic icon

Advanced neural networks 
Synthetic image data augmentation, image morphing, and normalisation algorithms.


Easy to use hyperparameters graphic icon


Easy to use hyperparameters
Fine-tuned image analysis without coding.


Built on YOLO (you only look once) CNN principles graphic icon


Built on YOLO (you only look once) CNN principles
Leading neural network for fast object detection.


Choose from a range of CNN models graphic icon


Choose from a range of CNN models
Select the most useful machine learning model for your application.


Automate with third-party systems graphic icon


Automate with third-party systems
Add to vision and metrology workflows via Bridge application.

What is HxGN Visual Detection?

HxGN Visual Detection is an automated surface inspection application capable of detecting defects such as scratches, cracks and dirt, on materials like glass, metal,...

Add surface inspection to your metrology workflows with HxGN Visual Detection

See how HxGN Visual detection integrates with PC-DMIS Vision and Optiv vision systems as part of the wider quality ecosystem.