Image dehazing play a vital role in several applications related to computer vision. The prime motive of this paper is to provide the overview of the existing deep learning algorithms associated with image defogging. In beginning, the main issues preset in the existing single image technique based on physical models are discussed. Thereafter, the basic concept of atmospheric scattering model and deep learning are discussed. The existing deep learning approaches based on single image defogging are decomposed into 3 broad categories namely Convolution neural network, Recurrent neural network, and Generative adversarial network with their pro and cons are discussed. The synthesised and real datasets used in defogging techniques are discussed with their applications. It also describes the various challenges and issues in the existing image dehazing techniques.

图像去雾在与计算机视觉相关的多种应用中起着至关重要的作用。本文的主要目的是概述与图像除雾相关的现有深度学习算法。首先，讨论了基于物理模型的现有单图像技术中预设的主要问题。此后，讨论了大气散射模型和深度学习的基本概念。将现有基于单图像去雾的深度学习方法分解为三大类，即卷积神经网络，递归神经网络和生成对抗网络，并对其优缺点进行了讨论。讨论了除雾技术中使用的合成数据集和实际数据集及其应用。