ABSTRACT
Convolution neural networks (CNNs) have shown extraordinary performance in the fields of computer vision and image processing. This is because the incredible ability of CNNs to automatically learn feature representation for a particular domain has given them an edge over the traditional computer vision methodologies, which heavily depended on human annotations for feature extraction. The shift of focus from traditional methods to CNN-based methodology has resulted in immense architectural innovations over the course of time. Choosing a proper CNN architecture is an extremely vital step towards solving a critical problem. We have done a thorough literature survey of various CNN architectures developed over time. This chapter acts as a convenient and comprehensive solution for CNN advancements over time, the major architectural improvements in each architecture, and what improvements they brought. Such innovations have enabled a smooth transition into Industry 4.0, which is heavily dependent on task automation through deep learning and computer vision. In this chapter, we have covered major CNN architectural innovations, a brief description about their constituent blocks, structural description, and limitations, and provided use cases that pertain to Industry 4.0.
