A Review of Applications and Design Space Exploration of Image and Video Analytics in Manufacturing Based on Neural Networks

The manufacturing sector offers numerous opportunities for the integration of image and video analytics based on neural networks, such as in quality assurance and maintenance. In this paper, a systematic literature review is conducted to identify and categorize use cases in terms of their role in specific process steps, their main outcomes, the characteristics to be assessed, and the potential for automation. Subsequently, the design space, consisting of network layers, normalization, and activation functions, and their purposes are examined, and commonly used building blocks are summarized. Finally, frequently employed evaluation metrics and performance indicators, including inference time and number of parameters, are presented. Based on the results, guidance is provided for practitioners and network architects in finding suitable use cases and appropriate network structures, balancing performance and computational effort to achieve the desired objective.