The Overall Design Of Processing System For Space-based Atmospheric Background Measurement And The Study Of Data Mining

Proceeding from the requirements of quantitative-processing and subsequentapplication for the infrared remote sensing data of atmospheric background, thisdissertation has described the designing and implementation of the whole quantitativeprocessing system, and how to apply the treated data with the idea of data mining. Fourfields are explored including evaluating and filtering data quality, characterizing imagetexture, data mining and classifying and machine learning, and an intelligent imageclassification system was implemented.The main work and results are as follows:In the overall design of processing system for space-based atmosphericbackground measurement, to meet the actual need for quantitative processing of remotesensing data, the overall design of the data processing system for space-basedatmospheric measurement was given and further divided into independent modulesaccording to their different functions, whose functions, implementing methods and basicmodels were designed.In the preprocessing of data for data mining, as an entry point, Radiance error ofatmospheric image was treated as an entry point, the quality of data was evaluated, thecalculation error was analyzed, the effects on data’s quality caused by different errorcomponents were studied. At last the final screening principles of the data wasdetermined considering a variety of factors. And using image texture features todescribe the cloud distribution, which were extracted using the method of gray-levelco-occurrence matrix (GLCM). By comparing each parameter obtained in GLCM,12characteristic parameters of texture were determined and divided into three categories.In the study of image classification method based on support vector machine,according to the distribution of clouds in an image, non-linear support vector machineclassifier (SVMC) was used to classify the data. Further experiments helped modifyingthe classification of texture characteristic parameters, the texture description parametersusing in learning and classifying were determined with a higher classification accuracy.At the same time, an intelligent algorithms was designed for support vector machineclassifier (SVMC). The results of experiments showed that the ability of SVMC is betterthan the traditional classifiers, and can avoid performance degradation caused by invalidparameters.The data processing system for space-based atmospheric background measuringsatellite had already been completed and put into practical application with stableperformance and high precision. The software for classifying radiance images was developed including the functions of data mining, data filtering and texture description.The work in this dissertation provides a reference for the design and constructionof quantitative remote sensing ground processing system in the future. At the same time,a complete method of data mining and classification is achieved based on actual data,which made a contribution for remote sensing image data processing from academicresearch to practical application.