Land Surface Environment Monitoring And Analysis Based On Collaborative Processing Of Active And Passive Remotely Sensed Data

In order to realize the purpose of comprehensive analysis land surface environ-ment from quantitative, qualitative, geometrical and physical integration aspects, it is necessary to research Land Use/Land Cover (LULC) classification, land surface tem-perature retrieval, soil moisture retrieval and ground subsidence extraction. With rich spectrum of color gradation, optical remotely sensed data has an advantage in target detection and recognition, however the observation time of optical data is limited by weather conditions. While, SAR remotely sensed data with characters of suitable for using in all-weather, all-time, multi-polarization, well penetrating ability, rich in texture information, etc, but SAR data will be affect by different frequency, model of polariza-tion, target geometry, dielectric properties and speckle noise, all of these disadvantage could limit the interpretation ability of the image and makes it is very difficult for in-formation extraction by sole SAR data. Both active SAR and passive optical remotely sensed data have their own application advantages, in order to make full use of infor-mation provided by active SAR and passive optical data for environment monitoring, a technology system of land surface environment monitoring based on collaborative processing passive and active remotely sensed data is proposed in this dissertation. The proposed technology system includes four levels of information fusion and classifica-tion, target recognition based on jointly used of multi-feature extracted, cooperative analysis and collaborative parameters retrieval. Taking the urban areas, mining areas and wetland areas as instances, the application of multi-source remotely sensed data information fusion and technique integration for environment monitoring by integrated method of information fusion, feature extraction and target identification, classifier en-semble, land surface parameters retrieval, ground subsidence are researched. The main contributions of my work can be summarized as follows.(1) A collaborative processing active and passive remotely sensed data for land surface monitoring using multi-source information fusion and multi-classifier ensem-ble classification method is proposed. The conclusion is that combination of improved wavelet method and full samples selected Sequential Minimal Optimization algorith-m for support vector classification (SMO) classifier ensemble has a robust to improve overall accuracy of classification. The wavelet fusion algorithm is improved based on Intensity, Hue, Saturation (IHS) method, classifier ensemble method based on samples selecting is examined and selected. Experiment results demonstrate that the improved fusion method can always obtain the best fusion result when compared with other fu-sion methods; multi-classifier ensemble method could improve classification accuracy at most time, while the overall accuracy will be affected by different subsets of samples.(2) Remotely sensed image classification approach based on ensemble learning and feature integration is proposed. Different features are combined through equivalent weights. And parallel and concatenation strategies are selected for classifier ensemble. The results also shown that polarimetric features, texture features and spectral features are more suitable for parallel classifier ensemble while spectral features and polari-metric features combination obtain a high score in concatenation classifier ensemble strategy.(3) A typical target recognition method based decision level fusion is modified. The variance and correlation texture indexes and Normalized Difference Vegetation In-dex (NDVI) are used to modify human settlement extraction using Local Indicator of Spatial Association (LISA). And classification map is generated through a decision fu-sion strategy. The results demonstrate that there is a significant accuracy improvement not only for sole class but also for overall accuracy.(4) The integrated monitoring method from geometrical, physical, qualitative and quantitative aspects, by collaborative processing active and passive remotely sensed data is designed. The mon-window algorithm is realized for land surface temperature retrieval; water-cloud model for soil moisture retrieval based on the joint using of opti-cal and S AR data is realized, two-pass differential interferometry technique is selected to obtain ground subsidence. Through a cooperative analysis, the relationship of LUL-C and land surface temperature, relationship of high level thermal field and LULC and other parameters are analyzed. And more RUSLE model and CA_Markov model are selected to analyze relationship of different parameters and their changes. The result-s of cooperative analysis demonstrate that the research of relationship of LULC and land surface parameter could make full use of the advantage of optical and SAR da-ta, an effective approach to profit application of’multi-temporal and multi-objective monitoring’and integrated modeling of land surface environment, which embodied the advantage of the combining of optical and SAR data for land surface environmental model in application.