A Comparative Review of Recent Architectures of Convolutional Neural Networks

A Deep Convolutional Neural Network (CNN) is an important part of deep learning that has delivered admirable successes in various competitions related to Image Processing and Computer Vision. Certain attractive application fields of CNN vary from Image and Video Recognition, Image Segmentation and Classification, Medical Image Analysis, Natural Language Processing, and Object detection. One of the greatest powerful abilities of deep CNN is the various feature extraction in an automatic way. Recently, developments in the research of CNNs and attractive deep CNN architectures have been described due to the inherence of the huge quantity of data and refinement in hardware automation. A handful of encouraging concepts such as the use of distinct activation and loss functions, regularization, parameters optimization, and architecture modernization, derive progress in deep CNNs. However, the remarkable advancement in the representational ability of the deep CNN is accomplished by architectural modernization. Thus, this review paper presents a brief survey of the advances that can occur in the architecture of CNNs from the very first architecture to the recent one. This paper, therefore, targets the inherent anatomy in the newly disclosed deep CNNs architectures and accordingly describes the strengths and gaps of various deep CNNs architectures.