Deep Learning Models and Their Architectures for Computer Vision Applications: A Review

Abstract

Deep learning is a modern and inspiring field of machine learning with a built-in ability to overcome from the shortcoming of conventional algorithms, which are dependent on hand-crafted features. Deep learning approaches have become one of the most recognizable solutions for multimedia and internet of things applications, like image segmentation, image enhancement, image classification, image generation, action recognition, pattern recognition, sequence prediction, and object detection. Much research has been carried out to analyze the various variants of deep neural network architectures for the widespread application domain. In this chapter, we focus on the major idea of deep learning and its architecture from basic to advance. Also, we discuss the motivation, methods, principal components, and limitations behind each architecture. This chapter investigates and presents a comprehensive survey of the major architectures of deep learning. It also covers various issues and challenges encountered during the design of the architecture. Furthermore, we discuss the importance and utility of deep neural network architectures in real-world applications. Finally, we conclude this chapter by explaining the future direction of research. © 2021, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.