Abstract: In terms of the ground-object spectrum information and the image-grain information as symbols to distinguish the ground-objects, fractal theory and BP neural networks are used to monitor thick fog at night, which increases the monitoring precision of thick fog obviously. Compared with the traditional Maximum Likelihood Classifying (MLC), the identified precision of clear sky ground, fog areas, cloud areas is increased, especially that of the cloud areas is increased by 10%. The image-grain extraction technology of grey-power box-counting dimension basing on grey degree connected region made extracting the borderline of cloud and fog more reasonable. Also, the merger of kings is discussed。