数据驱动与深度学习成为了近年计算机视觉算法的主流。在产业应用中，自动驾驶亦是计算机视觉研究的重点。该领域的一个前沿方向之一是对各种天气环境导致的能见度下降的图像进行增强。具体的，本研究将着眼于雨天和雾天。在该研究领域，既往研究面临以下两个挑战，挑战一，网络模型瓶颈：尽管深度神经网络在高级语义问题上取得了突破，没有充分物理建模的网络模型在图像增强领域普遍存在过拟合或欠拟合导致的性能瓶颈。挑战二，数据质量低下，机器学习与数据驱动的方法高度依赖与数据样本的质量，现有的室外天气研究可以普遍存在无数据集，数据集过小或者数据集质量差的问题。本课题将贯彻物理驱动的核心思想来解决上述提出的两个问题，第一点，通过物理建模，将机器学习的特征空间约束到一个稳定的子空间中，从而解决过拟合和欠拟合的问题。第二点，通过物理建模，创建大量真实的数据集，从而解决样本质量和数量的问题。

In recent years, deep neural networks and learning techniques show promising improvement for various high-level vision tasks, such as detection, classification, tracking, etc. With the physics imaging formation model involved, successful examples can also be found in various physics based vision problems (please refer to the references section). When physics based vision meets deep learning, there will be mutual benefits. On one hand, classic physics based vision tasks can be implemented in a data-fashion way to handle complex scenes. On the other hand, high-level vision task can also be benefited by awareness of the physics principles. This proposal will particularly focusing on modeling the out door weather phenomenons, e.g., rain, snow, haze and fog, which are common and inevitable in outdoor scenes, particularly driving scenes. A naïve machine learning algorithm, particularly deep neural networks, which be adversely affected by the weather. And therefore, this proposal employs a set of research which focusing on solving the problem and eventually improve the reliability of outdoor machine learning and computer vision algorithms.