The Study Of Cloud And Smoke Plume Detection Using Kmeans Clustering And Multithreshold Approaches

Forest fires destroy the natural resources and the environment which human survive in and seriously threat to the town around the forests. So, it is very important to research fire detection approaches. Satellite remote sensing technique has many advantages, such as high temporal and spatial resolution and monitoring periodically. People can study the surface features all-weather and all-time by using remote sensing. The EOS/MODIS sensor has high resolution and wide spectrum and it can give us very great help.Cloud and smoke detection is an important step for remote sensing monitoring forest fires. Identification of cloud and smoke is by no means a trivial task using spaceborne data. The most traditionally used method of identifying cloud and smoke is to assign different colors to different channels or channel combinations with absolute thresholds. The multithreshold approach is based on differences in the reflectance or brightness temperature of MODIS channels. However, the threshold method is subjective and not adapt to different seasons and regions that can easily lead to mistake or miss. This study developed new remote sensing methods for detecting cloud and smoke. Unlike many previous studies dealing mainly with multithreshold, the method proposed here combines Kmeans clustering and the multi threshold approach. On the basis of landmark spectrum analysis, MODIS data was categorized into several types initially by Kmeans. Then a multithreshold method was applied to fine the clustering result and eliminate interference, such as land and snow.The method is tested with MODIS data in different time under different underlying surface conditions. By visual method to test the performance of the algorithm, found that the algorithm can effectively detect smaller area of cloud and smoke and exclude the interference of underlying surface, which lays a good foundation for the next fire detection.