Deep learning-based image processing for fast and reliable answers to highly complex questions

Deep Learning is a subfield of Artificial Intelligence (AI). It is based on artificial neural networks (ANN), which are modeled on the human brain. Deep Learning algorithms are trained on large amounts of data and used to recognize and classify structured and unstructured information and patterns. This includes, for example, texts, images, speech and sounds.

Deep Learning based algorithms are to be learned by example and gradually optimize their capabilities. As with human learning, this depends on the wealth of experience. The more data and examples the Deep Learning algorithm has available, the better it becomes.

Advantages of Deep Learning in Machine Vision Applications

Conventional image processing systems are based on programmed, rule-based algorithms. They are capable of performing verification and classification tasks in an automated and repeatable manner that are predictable and thus essentially pre-programmed. However, when unexpected errors or irregularities arise that deviate from the program, or in the case of more complex multivariable testing tasks, traditional image processing reaches its limits. For instance, it cannot distinguish between purely external, tolerable appearance defects and functional defects.

Deep learning-based image processing provides a solution: It is versatile and can distinguish between acceptable deviations and natural fluctuations in similar parts and patterns as well as in complex environments. The algorithm continuously evolves based on new examples without the need for reprogramming the core algorithm.

Whether it’s verification tasks in final assembly, checking foodstuffs, or classifying materials and demanding OCR tasks: deep learning-based image processing is the optimal solution for large deviations, significant deformations, high feature variance, and challenging imaging environments. Through deep learning, the efficiency and accuracy of image processing solutions are significantly improved.

Advantages of Deep Learning:

Through end-to-end learning, deep learning automatically extracts the most relevant features from the data. This reduces the manual effort involved in defining features and leads to more accurate results.

Deep learning models can adapt to changing conditions and environments, allowing for high flexibility in various production settings.

The self-learning algorithms of deep learning can also detect complex errors and anomalies in production processes that often go unnoticed with traditional methods.

The fast and precise inspection through deep learning leads to improved production quality, reduced scrap rates, and increased overall efficiency.

Deep learning solutions can be scaled across various application areas, making them a cost-effective and versatile choice for businesses.

Deep Learning technology from leading manufacturers: user-friendly, precise and highly efficient

Deep learning technologies have long been user-friendly, intuitive, easy to train, and do not require extensive expertise and time-consuming programming.

HALCON from MVTec – Comprehensive standard software for industrial image processing

HALCON is MVTec’s comprehensive standard software for industrial image processing, used in a wide range of industries worldwide. With its own integrated and highly interactive development environment, called HDevelop, HALCON is specifically tailored to the development of machine vision solutions. The software offers a wide range of applications covering the entire workflow of a machine vision application and has been proven in hundreds of thousands of installations worldwide.

HALCON is used in a variety of industries, including automation, logistics, food processing, automotive, medical and pharmaceutical. It provides exceptional performance in all areas of machine vision, including alignment, calibration, completeness checking, identification, inspection, measurement, comparison, and object and position detection. In addition, HALCON supports advanced image processing technologies such as 3D vision and deep learning algorithms.

Advantages for industrial image processing

HALCON is designed as a comprehensive toolbox and offers more than 2,100 operators covering the entire image processing process. This allows all conceivable image processing applications to be developed in an accelerated manner.

With its powerful image processing library and comprehensive features, HALCON enables developers to accelerate their production processes and speed time-to-market for image processing applications.

HALCON offers outstanding performance and supports various computing hardware, special instruction sets such as AVX512 and NEON, and GPU acceleration, resulting in fast and reliable inspection solutions.

The software is compatible with a wide range of operating systems and provides interfaces to hundreds of industrial cameras and frame grabbers. HALCON also supports common standards such as GenICam, GigE Vision and Arm®-based embedded vision platforms.

HALCON is portable to various target platforms and can therefore be ideally used in embedded and customer-specific systems, ensuring long-term investment security

MERLIC from MVTec – Intuitive image processing software

MERLIC is another powerful image processing software from MVTec. It is characterized by its user-friendliness and intuitiveness and enables even inexperienced users without programming knowledge to configure inspection routines quickly and easily. MERLIC offers a graphical user interface with intuitive operating concepts such as “easyTouch”, which enables an efficient workflow as well as time and cost savings.

MERLIC is ideal for a wide range of machine vision applications, including quality control, inspection, identification, measurement, monitoring and more. The intuitive user interface enables companies to quickly respond to changing requirements and efficiently develop customized machine vision solutions.

Advantages for industrial image processing

MERLIC enables the creation of complete solutions without the need to write a single line of code. Thanks to its graphical user interface and powerful tools, complete machine vision applications can be created easily and intuitively. This leads to an enormous reduction in development and commissioning times for machine vision applications.

MERLIC covers the entire image processing process, from image acquisition to image processing and visualization of the results. The software supports all common industry standards and offers a wide range of hardware compatibility.

MERLIC includes the latest deep learning features as well as all essential methods of classical image processing. The comprehensive tool library enables the simple and intuitive solution of a wide range of image processing tasks, including classifying, measuring, counting, inspecting, reading text and numbers, barcodes and datacodes, position determination as well as 3D vision based on height images.

The software can be extended with user-defined tools developed on the basis of MVTec HALCON. The add-on “Extension Tools” enables demanding users to implement even advanced image processing applications and to extend the functionality of MERLIC.

MERLIC supports parallel processing and execution of different tools. This simplifies the implementation of multi-camera setups and makes more efficient use of the system’s computing power. Different machine vision tasks can be executed in a single run, increasing the efficiency of image processing.

VIDI from Cognex – A powerful Deep Learning tool

As a leading provider of machine vision systems, Cognex offers the advanced Deep Learning Tool “VIDI”. VIDI was developed and optimized specifically for factory automation requirements. It uses powerful AI algorithms to automate and scale complex tasks such as part localization, assembly verification, defect detection, classification, and character recognition.

VIDI enables users to easily integrate Deep Learning into their existing machine vision systems without sacrificing performance. The graphical user interface facilitates neural network training and enables efficient adaptation to different industry requirements.

With the ability to detect even complex defects and patterns, VIDI revolutionizes quality control and defect detection in production lines. Our customers benefit from higher accuracy, faster inspection times and reduced downtime.

VIDI image processing tools

Cognex’s VIDI deep learning tool is a powerful collection of machine vision tools specialized to solve different tasks. VIDI tools share a common engine but differ in their focus, analyzing single points, individual regions or complete images.

This tool is used to find and locate single or multiple features within an image. It can identify and locate complex features and objects, even on turbulent backgrounds. It learns from annotated images and can be trained efficiently by providing images in which the features to be found are marked.

This tool performs optical character recognition (OCR) on an image. It can identify and read characters even if they are highly deformed and on very unstable backgrounds. Like the Blue Locate Tool, the Blue Read Tool is trained from annotated images in which the characters to be recognized are marked.

This tool is used to detect anomalies and aesthetic defects. It can identify scratches on decorated surfaces, incomplete or defective assembly, and weaving problems in textiles by learning the normal appearance of an object, including its significant but tolerable variations. The training data consists of images of defect-free objects.

This tool is used to segment specific regions such as defects or other areas of interest. For example, it can detect air pockets in cast metal or bruised vegetables on a conveyor belt. The tool learns from images of defective regions that are being searched for.

This tool is used to classify an object or a complete scene. It can identify products based on their packaging, classify welds or separate acceptable and unacceptable defects. The Green Classify tool learns to distinguish different classes based on a collection of labeled images.

Advantages for industrial image processing

Engineers can train a deep-learning-based model in minutes based on just a small set of sample images. This enables rapid implementation and process control.

Unlike many other deep learning-based solutions, Cognex’s deep learning tool requires only a GPU card and can operate on limited computing power.

Cognex Deep Learning can be maintained and retrained at the plant level without intervention by the plant engineer or system integrator.

The technology works with high-resolution images, including color and thermal images, and can virtually detect any anomalies.

Cognex Deep Learning provides access to powerful industrial image analysis solutions for non-specialists in machine vision through our interface.

Rule-based image processing or deep learning technology? It depends on the use case.

In many cases, vision systems based on conventional as well as Deep Learning-based software can be used complementarily and complement each other excellently. While rule-based vision systems are better at measurement tasks, identifying and reading uniquely recognizable barcodes, and robot guidance, Deep Learning creates completely new application possibilities for industrial automation that were not technically possible before.

Are you also looking for the optimal solution for your specific application or do you have more in-depth questions about Deep Learning-based image processing systems? Give us a call!
Image Processing
Vision Sensors
Vision Systems
Lenses
Lighting

Any questions? Let us talk!

Call us at +49 4101 805 0028 or write to