
PEKAT VISION offers a powerful set of deep learning modules for industrial inspection — including Anomaly Detection, Surface Detection, Detector, Classifier, and OCR.
These can be combined and extended with scripting to solve complex inspection tasks. For most applications, no programming is needed — PEKAT VISION includes everything required for effective quality inspection.
Deep Learning Modules
Anomaly Detector

The Anomaly Detection module identifies previously unseen defects using unsupervised deep learning. It requires only a small set of defect-free images for training and no manual labeling. This makes it ideal for products with irregular patterns or complex backgrounds.
Surface Detection

The Surface Detection module is designed to detect visible surface defects such as rust, abrasions, cracks, and inclusions — even on heterogeneous or textured materials. It uses supervised deep learning and requires training on images with annotated defects.
Object Detector

The Detector module locates and optionally classifies objects or defects — even those with variable shapes such as fruits, wood knots, or scratches. It requires annotated examples of each object or defect. Object orientation does not affect detection.
Classifier

The Classifier module assigns a single class to an entire image, to objects detected by other modules, or to objects in fixed positions. Each object or image is classified into one predefined category.
Enhancements
Parallelism

Parallelism enables multiple processing branches to be defined independently. Each branch runs in sequence, and their results are combined afterward. As of version 3.14, no programming is required to configure such multi-branch processing.
Code

The Code module allows you to extend PEKAT VISION with Python. Use it for advanced preprocessing (e.g., with OpenCV or NumPy), custom logic, or integration with external systems and interfaces.
Other Features
Confusion Matrix

This powerful tool evaluates the performance of trained models by comparing user annotations with model predictions, offering valuable insights for fine-tuning your models.