Modern Methods of Industrial Image Processing

In the seminar "Modern Methods of Industrial Image Processing," you combine classic image processing with AI and deep learning. A live example shows you the complete workflow of a deep learning application, from data preprocessing to license plate recognition.

At the beginning are the basics: image acquisition, binarization, labeling, and blob analysis. You learn how lighting, optics, and camera affect the contrast of your images, for example through spectral adjustment or depth contrast. Added to this are structure and spatial resolution as well as subpixel-precise measurement methods with camera calibration.

The second part deals with machine learning in industrial image processing. You learn classifiers such as Nearest Neighbors, SVM, and Decision Trees, as well as supervised learning and validation with metrics. In deep learning, you see typical use cases and the architecture of neural networks. You also get advice on model selection, training, and monitoring in practical use.

Real practical examples range from license plate recognition to food sorting and surface inspection and OCR. As current trends, the seminar covers Large Language Models, Vision Language Models, and Event Based Imaging.

The seminar is aimed at engineers and technicians. It is meant for people who develop automation, monitoring, and inspection systems or use and specify image processing systems. It is led by Prof. Christoph Heckenkamp and Prof. Thomas Netzsch, both from the Hochschule Darmstadt.

The event runs over two days. On the first day, it goes from 09:30 to 17:30, on the second from 09:00 to 16:30. You choose between in-person dates in Düsseldorf, Munich, and Frankfurt am Main or an online date. At the end, you receive a certificate of attendance. The seminar is also bookable as an in-house training.

• Deepening of the classical methodology of industrial image processing: robustness, contrast, resolution
• Explanation of established classification methods and basics of Machine Learning
• Categories of ML methods: classification, object detection, anomaly detection
• Metrics and methods for evaluating classification performance
• Live example: development of a machine learning application for vehicle license plate recognition, from preprocessing to classification
• The classical methodology of industrial image processing: image acquisition, binarization, labeling, blob analysis
• How contrast arises: illumination, spectral adaptation, color, NIR, fluorescence, spectral imaging
• Depth contrast and radiometric stereo (topographic contrast)
• Structural resolution: detection of small structures, pixel resolution and contrast
• Measurement: spatial resolution and the influence of the binarization threshold
• Subpixel-precise measurement variables and compensation methods
• Geometrically faithful images: rectification, camera calibration, measurement uncertainties
• Machine Learning in industrial image processing: feature selection, classifiers (Nearest Neighbours, SVM, Decision Trees)
• Supervised Learning, Overfitting/Underfitting and validation (metrics, confusion matrix)
• Deep Learning in industrial image processing: typical use cases, neural network architecture, unsupervised learning
• Tips for practical use: model selection, training, monitoring, drift, evaluation
• Practical applications: license plate recognition, food sorting, OCR, surface inspection
• Current trends: Large Language Models, Vision Language Models, Event Based Imaging