Competencies
AI vision
What does artificial intelligence mean?
The term Artificial Intelligence (AI) is not precisely defined. A definition fails simply because the concept of intelligence itself is ambiguous.
The term Artificial Intelligence (AI) was coined as early as the 1950s and is therefore historically shaped and subject to many influences. The term reflects the vision and the "big picture." AI touches upon several technical and scientific disciplines and is often used (including by us) as a catchy marketing term.
When AI is discussed in measurement technology or image processing, it usually refers to machine learning (ML) or deep learning (DL). The terms build upon each other as follows:
Artificial Intelligence (AI) → Machine Learning (ML) → Deep Learning (DL)
Historical development
Although artificial intelligence is often described as a trend, it is by no means a new phenomenon. As early as the 1950s, scientists shared the conviction that the process of thinking is not limited to the human brain. After research on the topic stalled, particularly in the 1980s, technology companies like Google brought new momentum to the field in the 2000s. Today, artificial intelligence is an integral part of our everyday lives.
What does machine learning (ML) mean?
In image processing, there are basically two different ways to address a problem:
Rule-based programming
- Manually formulating and programming rules after the results have been calculated or defined
Machine learning
- Training a model with data
- Independent learning of patterns from the data
- Classifying or estimating outcome variables
Which approach is better depends on the application and must be carefully assessed or systematically determined. Rule-based approaches are well-suited, particularly in metrology and image processing, for making decisions based on clear measurement characteristics and rules. If the rules are unknown or can only be systematically extracted from images with considerable effort, a machine learning method may be the better approach. Machine learning methods are typically used in image processing for difficult segmentation tasks, optical character recognition (OCR/OCV), pattern and anomaly detection, object and image recognition, and image classification. In modern applications, both approaches are usually combined effectively.
Learning methods for machine learning (ML)
In broad and simplified terms, there are 3 different learning methods for machine learning.
Deep Learning vs. Machine Learning
Modern AI applications are almost exclusively based on deep artificial neural networks (ANNs). The special feature of these networks is their ability to perform complex tasks, completely eliminating the need for manual feature extraction. An ANN is therefore capable of independently fulfilling complex tasks.
- Deep learning is a subfield of machine learning
- DL methods are based on artificial neural networks with multiple intermediate layers
Artificial neural networks (ANN)
Deep learning methods are based on artificial neural networks. These networks are inspired by (biological) neural networks, which are part of a nervous system.
Artificial neural networks are structured in layers. A layer consists of several artificial neurons. Between an input and output layer are several hidden layers. The structure gets its name "deep" neural network because of these hidden layers.
Convolutional Neural Networks (CNNs)
In image processing, so-called "convolutional neural networks" (CNNs) are most commonly used. Inspired by the visual cortex, these networks consist of multiple feature maps. These feature maps correspond to the layers of an artificial neural network and are generated through convolution. The convolution operators produce different features, such as edges. In image processing, a CNN should be able to generalize features and represent them at increasingly higher levels of abstraction.
Source: LeCun, Yann, et al. “Gradient-based learning applied to document recognition.” Proceedings of the IEEE 86.11 (1998): 2278-2324
Source: Lee, Honglak, et al. “Unsupervised learning of hierarchical representations with convolutional deep belief networks.” Communications of the ACM 54.10 (2011): 95-103.
MV.MINDNET, the Deep Learning platform from MABRI.VISION, combines state-of-the-art technology and practical solutions.