The Research On Heavy Fog Recognition Based On Remote Sensing Image

Heavy fog recognition and the separation of cloud and fog has always been the hardest part in weather forecasting. Recently, with the rapid development of socio-economic, the harmful effects of fog become increasingly serious. Fog has become a natural disaster. So how to prevent fog disaster, how to monitor and early warn heavy fog, has become the focus of academic research. On the other hand, with the development of science and satellite remote sensing technology, the research of separate cloud and fog and extract fog areas gets more and more attention. Now lots of technology is been used in heavy fog recognition, and provide strong impetus for the research.The paper based on previous studies, summing up the basic principles of Region Growing, Fuzzy C-means Cluster and BP(Back Propagation, BP) Network, using the function set and toolbox of MATLAB software, programming to separate cloud and fog on MTSAT-1R satellite image, recognizing and extracting fog areas. The result of this research provides basis of heavy fog monitoring.First, the paper uses growing rule which based on improved region grayscale difference method to build Region Growing Algorithm, in order to extract fog areas from remote sensing image and reach the purpose of heavy fog recognition. Experimental results show that this Region Growing Algorithm is easy to understand, easy for computer programming, and also segment fast and effectively.Second, the paper uses the function of fcm in MATLAB software, which based on Fuzzy C-means Cluster, to extract fog areas from remote sensing image, in order to provide basis of heavy fog recognition. Experimental results show that this Fuzzy-Cluster-Based algorithm do not need training samples, easy for programming, and have high accuracy of segmentation, but cost more time than Region Growing method.Third, the paper uses improved BP Algorithm to build BP Network, using train function to train the net and sim function to simulate net. So that we can extract fog areas from remote sensing image with this method. Experimental results show that BP Network can separate fog areas from other features properly and produce good results. But this method cost much more time than the others.Last, the paper uses the same image of MTSAT-1R satellite as original image, using above three methods to segment image and extract fog areas, in order to compare these three methods. Experimental results show that Region Growing Algorithm has the highest accuracy and the fastest speed, BP Network method is worst one of the three. This experimental result is fundamental of future heavy fog recognition.