Anomaly Detection

Deep Learning for Defect Detection

Deep-learning-based anomaly detection enables automated surface inspection of products and the precise localization of defects. Unlike traditional methods, anomaly detection does not require manual labeling of training data. Instead of working with defined defect patterns, the model learns to detect deviating areas in new images based on the patterns of the defect-free original images.

This technology is particularly advantageous for the inspection of surfaces, as it localizes defects independently and reliably. Whether the issue is deformed components, missing parts, or defective surfaces, anomaly detection detects them reliably and quickly.

FAST & EFFICIENT

Advantages

No labeling required – The method only needs good images for training.
Fast training – Only a small number of images are required for training.
Efficient model adaptation – Training and prototyping are completed in seconds or minutes.
Minimal computing power required for the training phase.
The technology enables quick detection of defects and deviations, allowing for rapid error identification in production without extensive model training.

UNIQUE TECHNOLOGY

Global Context Anomaly Detection

Global Context Anomaly Detection is a unique technology from MVTec that takes the logical content of the entire image into account. This makes it possible to detect even completely new types of anomalies, such as missing components, deformed parts, or incorrectly positioned objects.

This method is particularly useful for the inspection of printed circuit boards (PCBs) or embossings in semiconductor production.

Advantages of Global Context Anomaly Detection

Detection of anomalies in the global context of the image
Powerful algorithms for high detection accuracy
No labeling required – training with “good” images only
Low training effort: only a few defect-free images are needed
Fast inference times and quick model adaptations

The image shows bottle necks being inspected for defects using deep learning and anomaly detection. Anomalies such as deformed or defective bottle necks are identified and localized by the technology.

Application Examples

Defect Detection in Printed Circuit Boards (PCBs)

In semiconductor production or during printed circuit board (PCB) inspection, anomaly detection identifies subtle defects, such as missing components or deformed parts, without the need for a complete defect database.

Defect Detection In Prints

For quality control of printed images (e.g., in the packaging industry), anomaly detection identifies irregular printing patterns or missing elements on labels and packaging.

OUR KNOWLEDGE & SERVICES

Benefit From Our Expertise

PRACTICAL INSIGHTS & EXPERT KNOWLEDGE

Discover Our White Papers!

Explore our white papers to gain practical insights and expert knowledge on industrial machine vision. Download them now to understand key technologies, current trends, and real-world applications that support informed technical and strategic decisions.

To the white papers

INDIVIDUALLY TAILORED

Evaluation Of Your Application

Do you want to know if we offer the right solution for your industry? Send us your software application design and our experts will check it.

Learn more

HALCON & MERLIC

Integration In Our Software

HALCON provides a powerful development environment where deep learning models for anomaly detection can be quickly created and implemented in the production line.

Learn more about HALCON

MERLIC offers easy-to-use tools that enable even non-programmers to utilize deep learning models for defect detection.

Learn more about MERLIC

HOW DOES IT WORK

Helpful Tutorials

Please note: Once you watch the video, data will be transmitted to Youtube/Google. For more information, see Google Privacy.

DATASETS FOR YOUR RESEARCH

Anomaly Detection Datasets

MVTec offers a range of datasets for anomaly detection, including MVTec AD for industrial inspection benchmarking. MVTec AD 2 expands existing benchmarks with 8 new scenarios and over 8,000 high-resolution images. The MVTec 3D AD dataset focuses on unsupervised 3D anomaly detection and localization.

Go to our datasets