Industrial cameras are already being used in a wide range of industries: from equipment, plant and mechanical engineering to medical technology, agriculture and logistics. AI is opening up completely new fields of application, particularly in the area of machine vision. For example, AI-based systems enable the recognition and processing of objects with natural variance, such as food, plants or other organic objects that cannot be covered by conventional image processing. Color, surface, size, weight or shape can vary greatly. However, AI can be trained with the corresponding training data in such a way that a broad spectrum can be reliably recognized, categorized and thus also processed. The development of AI solutions with machine learning (ML) relies on a more flexible approach without rigid rules. Instead of manually developing a program code, the AI in machine learning undergoes a learning process in which it identifies important features itself using suitable image data and relates them to the annotations provided. However, the selection of this data harbors a high potential for error. In practice, the AI is sometimes provided with images with unimportant content, poor lighting, blurring or even incorrect labels, which can greatly distort the recognition accuracy of the desired features.

The key skills for working with ML methods are therefore not the same as for rule-based image processing - and the provision of hardware alone is not enough. The AI must be tested, validated, retrained and finally integrated into the application in order to guarantee a productive workflow.

The right tool for every task

System programmers are not necessarily needed to train AI-based vision systems. IDS, for example, demonstrates this with its IDS NXT platform. The idea behind it: With the right, coordinated tools, any user group can fully exploit the potential of AI vision without having to invest a lot of time and money in building up new core skills. Specialist knowledge for training neural networks and programming your own applications can be packaged in the tools for many simple AI workflows. This allows each user to implement their individual requirements without having to set up their own team of specialists. The software provides each user group with the right tools for their respective tasks and working methods.

The majority of common image processing applications work with relatively simple processes. Capture image, analyze image, make process decisions, initiate action - basic functionalities that differ in just a few details and therefore do not need to be reprogrammed each time. However, the selection of a deep learning (DL) use case, such as "classification" or "object detection", is often already too abstract as an entry point for a project to be able to derive the further action steps required for data acquisition and vision app configuration.