As a kind of road facilities, traffic signs are presented by various symbols with multiple backgrounds to guide, limit, and warn road users. It is one of the important parts in civilian transportations. Once a driver overlooks the signs, especially important ones, due to complexity of actual traffic scenes or influence of adverse weather conditions, which will lead to violation of traffic laws or regulations that may cause traffic accidents, result in casualties and property losses. Therefore, the recognition of traffic signs is a vital part of real requirement of modern traffic systems, which takes not only predominant significance in our daily life, but also high values in academy research. Due to the complexity of specific environment and increasingly severe haze in recent years, high requirements are put forward for traffic sign recognition (TSR). At the same time, the tolerance and adaptability requirements in practical application environment will also be improved. Based on the recognition of traffic signs, in this thesis, we propose an image defogging method by using guided image filtering. Then convolutional neural network (CNN) in deep learning is applied to recognize traffic signs under foggy weather condition, improve accuracy and speed of TSR. Our main contributions of this thesis are: (1) We propose a HSV color gamut defogging algorithm based on CLAHE algorithm. We offer a guided image filtering algorithm for image dehazing. Our result shows that the guided image filtering method is very effective in image defogging. (2) In this thesis, we proffer two deep learning algorithms for our experiments, one is Faster R-CNN, the other is YOLOv5, we compare the two methods to find that YOLOv5 is much suitable for real-time TSR. (3) In this thesis, we propound a new way to identify road signs. The improved YOLOv5 model is applied to satellite imaging for accurately detecting road signs on ground.