Application Potentiality Assessment Of Urban Land Cover Classification On High Spatial Resolution Remote Sensing Data

High spatial resolution remote sensing data are increasing and playing an important role in all works of life. However, it becomes a topic research to find suitable remote sensing data for meeting the corresponding application requirements. At present, the study of this aspect is not systematic and mature at home and abroad.This thesis, from the point of basic quality parameters evaluation of high spatial resolution remote sensing data, established function model between basic quality parameters and classification overall accuracy of data. The classification overall accuracy is predicted according to the function model, and using the classification overall accuracy to indicate data application potential, which help user to find the suitable remote sensing data to meet the demand of research and application. The study of this thesis mainly embodied in the following three aspects:1) High spatial resolution remote sensing data optimal segmentation result and classification accuracy evaluation method. The objected-based image analysis(OBIA)is better than traditional pixel-based image analysis for high spatial resolution remote sensing data. Better classification accuracy is obtained using OBIA method. OBIA method includes two key steps: first, image segmentation, and then classification.Classification using optimal segmentation result, in this thesis, can obtain best classification accuracy. In obtain the optimal segmentation result step, multi-scale unsupervised evaluation method is used to obtain the optimal segmentation result, and band weight assignment method determined the single-scale optimal segmentation result and difference indicator is improved in this method. Put forward three scales optimal segmentation result is treated as a new optimal segmentation result under single scale, to identify new optimal segmentation results under two-scale and three-scale, and repeat this process three times, take the lowest GS(Global Score) value under three-scale optimal segmentation result as the optimal segmentation result for the whole image. Random forest method which belongs to decision tree classification method is used to classify the optimal segmentation result of the image data, and has more advantage than ordinary decision tree. In object-based classification accuracy evaluation,confusion matrix of internal accuracy of reference object is established to evaluation object-based classification result using reference objects as sample.2) The effect of high spatial resolution remote sensing data’s basic quality parameters to classification accuracy. The basic quality parameters which effect high spatial resolution remote sensing data application potential can be divided into three categories: the space characteristic parameters, the radiation characteristic parameters and the geometrical feature parameters. The study selected MTF(Modulation Transfer Function), SNR(Signal-to-Noise Ratio) and BBRE(Band-to-Band Registration Error)as basic quality parameters from the three categories to research the relationship with classification overall accuracy of data. For this study, the study area data of high spatial resolution remote sensing data from WorldView-2, use Gaussian low-pass filter model to change MTF value of the data, and analysis classification overall accuracy OA value change of data under different MTF values situation, then variation is concluded between them, the results have shown: classification overall accuracy OA value of urban land cover types changes with the change of MTF value; when MTF value decrease, the OA value has the tendency to reduce. When MTF value between 0.12 and0.25, the change trend of OA value is not obvious; when MTF<0.15, the OA value has obvious downtrend. The OA value changes with the change of MTF value, and approximately conforms to logarithmic function relationship. Using the same remote sensing data, the OA value’s change is analysis under different SNR values through controlling the SNR value of the data, then variation is concluded between them, the results have shown: classification accuracy OA value of urban land cover types changes with the change of SNR value, when the SNR values decrease, the OA values have a decreasing trend; when SNR >46.7, the OA values are over 80%, and the change trend is not obvious; when SNR<41.2, the OA values are less than 80%, and the change trend is absolutely obvious in this interval; classification overall accuracy OA value of urban land cover types and SNR of the data approximately conforms to logarithmic function relationship. Using the same remote sensing data, the OA value’s change is analysis under different BBRE values, through changing the BBRE value of the data, then variation is concluded between them, the results have shown: classification overall accuracy OA value of urban land cover types reduce with the increase of the Band-to-Band Registration Error(BBRE) value; when BBRE value between 0.125 and1, the OA value change greatly; when BBRE value continue to increasing from 1.05, theOA value change trend slowly, but fluctuate situation of the OA value is obvious;overall accuracy OA value of urban land cover types classification and BBRE of the data approximately conforms to power function relationship.3) Establishing application potentiality assessment model of urban land cover classification on High Spatial Resolution Remote Sensing Data. Considering the basic quality parameters impact of MTF, SNR, and BBRE, the function relationship is established between the three basic quality parameters and overall classification accuracy of image data using the multiple regression analysis method with the high spatial resolution remote sensing data of the study area, so as to establish high spatial resolution remote sensing data application potential model for urban land cover types classification, which verifies the validity by the other two study areas.