Deep Convolutional Neural Network For Fog Detection

Fog detection has becomes more and more important in recent years, real-time monitoring information is very beneficial for people to arrange production and life. In this paper, based on meterological satellite data (Himawari-8 standard data, HSD8), Covolutional Neural Network (CNN) is used to detect fog. Since HSD8 consists of 16 channels, the original CNN is extended to multiple channels for HSD8. Multiple Channels CNN (MCCNN) can make the full exploitation of spatial and spectral information effectively. A dataset is created from Anhui Area which consists of ground station data and grid data. Different image sizes and convolutional kernels are used to validate the proposed methods. The experimental results show that the proposed method achieves 91.87% accuracy.