Cloud Classification Of Satellite Imagery Based On Neural Network Methods

Using Meteorological satellite cloud images and digital data can track changes inclouds, analyze weather and forecast.So classification and identification of cloudimages is very important, which can lay a good foundation for further analysis. First,this paper introduces different kinds of clouds and main feature of them on cloudimages,then provides an overview of development of the cloud classification topic,briefly introduces knowledge of neural networks and Meteorological satellites.This paper uses Back-propagation(BP) neural network to classify the data whichis cut out from the cloud image. BP neural network is one of the most widely usedneural network models,the algorithm is much more mature and typical,in practicalapplications80%~90%of the nueral network models are the changes of BP networkmodel.This paper selects8different features, and according to the classification resultof National Meteorological Satellite Center,the number of result is set to be8.Compare different results of different training parameters, numbers of hidden layersand numbers of hidden layer nodes, choose the best combination to learn samples andclassify. At last,analyze the result and its feasibility.Learning speed of BP network is slow and performance of the classifier bylearning samples is large.In order to solve this problem,this paper introduces a newunsupervised Self-Organizing Feature Mapping(SOM) to cluster the same data. SOMnonlinear projects high-dimensional characteristics onto a low dimensional space sothat the characteristics become easy to distinguish, the visual results is a bigadvantage too.Select same features and set the number of result to be8too,analyzedifferent results of different topology functions. Select the best classification result tocompare with the result of satellite center and original cloud image data, and analyze its feasibility.According to the results of the two methods,both the time of learning and theaccuracy of the results are better from SOM network than BP network.