Cloud Classification Of Satellite Imagery Based On ELM And SVM

Meteorological satellite can continuously observe the earth surface and cloudlayer on a large scale. The cloud images sent back from meteorological satellitesprovide a wealth of meteorological information, which contributes to analyze weathervariations, especially the rainfall. However, with the development and explosivegrowth of data source of cloud images, the research and application of relevant tool foranalyzing and processing is badly lagged behind. There exist many problems in theapplication of traditional classification algorithm to the remote sensing cloudclassification, such as the much too large treatment scale, the complex analyzingprocess and tendency to trap in local minimum. In particular, the classification speedand accuracy is far from satisfied. So, the automatic classification techniques ofsatellite cloud image which have distinct advantages of accuracy and rapidity andsimplicity have been the development direction and the focus of research nowadays inall countries in the field of remote sense.Based on this, this paper applies a good learning algorithm for feedforwardneural networks——Extreme Learning Machine to construct the classifier. In addition,the Support Vector Machine is applied to do the same experiment for comparison withextreme learning machine. The main research work and innovation are listed asfollowing：(1)Firstly, this paper introduces the topic selecting and significance, thendescribes the history and present situation of cloud classification research in detail，andexamines the research measures of clouds classification deeply．(2)Introduces the concept of meteorological satellite and satellite cloud image.Analyzes the properties of remote sensing image and the cloud classification theory.Introduces the types of clouds as well as the performance characteristics in the satellitecloud image and details the samples used in this experiment and reading method.(3)Studies systematically the algorithm of ELM. Explicates the algorithm‘sadvantages and peculiarity. Applies the ELM to clouds classification innovatively.Based on the experiment results, the effect of choice of the node number of the hiddenlayer is analyzed, including the classification accuracy and learning speed.(4)For comparison, this paper uses the SVM algorithm to design the classifier and do the experiment by the same samples. Points out the advantages anddisadvantages of the two classification algorithm.By summarize the classification results, we can conclude that the ELM can beapplied to clouds classification effectively and has a distinct advantage in learningspeed but the classification accuracy of ELM is lower than SVM.