Object-oriented Scale Conversion And Uncertainty Analysis For Remote Sensing Products

In recent years with the development of the remote sensing applications, various forms of data have appeared which makes data integration and multi-source data assimilation become a brand new difficulty for the development of the remote sensing application. In the meantime, the key to resolve this difficult problem is the research about the scale difference of the data. The scale research for remote sensing data can be divided into two parts, scale conversion and scale effect. In earlier days, most researches of scale are based on pixel level, which only consider the spectral information of surface features while ignoring the spatial structure and morphological characteristics among the objects. More importantly, the object-oriented researches for image analysis are mostly limited in the stage of image information extraction. Even fewer researches focus on data applications based on object-oriented. For the lack of research in this area, Heihe River Basin has been chosen as the study area for this paper,with the analysis of ground measurement data, High-resolution image data like TM and Low-resolution of ground surface features parametric products(including MODIS and homemade FY-3 temperature product), to build an object-oriented scale conversion model for ground surface characteristic parameters. With the help of an intermediate parameter to relate the surface feature class, image objects and surface characteristic parameter, parameters and objects are linked to complete object-oriented scale conversion, and conversion results can be applied to comparative and uncertainty analysis finally. The results of research are as followed:(1) A summarized available image segmentation algorithm has been put forward,which achieves multi-scale image segmentation, and provides access to the network of objects at different scales by setting different segmentation parameters. The object-oriented information extraction has been achieved, by using fuzzy classification extracted feature information at different scales of object layer. The Kappa coefficient is 0.7944, and overall classification accuracy is 0.8698 by using confusion matrix to analyze and the evaluation of classification is impressive.(2) The object-oriented model for remote sensing scale conversion is built through combining land measured data and satellite image data. Firstly, the intermediate parameters(NDVI and land surface emissivity) are collected from remote sensing data,and the inversion of characteristic parameters is achieved, on the basis of the inverted land category information and land surface characteristic parameters(LAI and LST).Especially for the inversion of LAI, with surface measured point data and satellite data process regression analysis, the inversion results are correlated with the ground measured data up to 0.955, and the accuracy of satellite data retrieval has been highly improved. Secondly, in the area of calculation of the correspondence relationship between large-scale of point data(MODIS LAI and FY-3 LST) and small-scale of area data(Thematic Mapper image), the scaling object-oriented is completed through inter-object merger method which is to make surface characteristic parameters; In the pilot phase to select the specific leaf area index and surface temperatures of the two parameters of surface features, each of the reference data correlation between large-scale remote sensing products were 0.688,0.380. Based on the results of the differences between the correlation coefficient, the evaluation and analysis for universal scale conversion and inadequate of method are introduced in the paper.(3) In the process of object-oriented scale conversion, different segmentation parameters lead to different classification results, which lead to the different inversion distribution of characteristic parameters. With the analysis of the results, the uncertainty during the changes of the various stages of scaling has been described by the comparative data in the paper.In summary, this paper is based on object-oriented information extraction to achieve the object-oriented scale conversion. Furthermore, the process of converting at various stages about the scale effect and the uncertainty of the results has been analyzed,for realization of multi-source remote sensing data assimilation, remote sensing product authenticity test and study universal scaling method is of significant value certainly.