Method Research On Spatial And Temporal Quantitative Information Fusion With Multi-Sensor Remote Sensing Data

With the increase in kinds and quantities of remote sensing sensor, the application scopes of remote sensing have been continuously expanded demanding for data with higher spatial and temporal resolutions. However, limited by the technical conditions, the configuration designs of remote sensing instruments have been obliged to counterbalance the spatial resolution and swath width. As a result it is often difficult to acquire remotely sensed observations with high spatial resolution and frequent coverage, which limits some potential applications. How to achieve dataset with higher spatial and temporal resolutions within current observational circumstances by using low spatial but high temporal resolution data and high spatial but low temporal resolution data, is a newly developed direction of remote sensing data fusion, namely the spatial-temporal information fusion technology. Thanks to the concise modeling, wide applicability and no prerequisite for any auxiliary data and other merits, the technology has received extensive attention from scholars both at home and abroad in recent years. Therefore, to improve the spatial and temporal resolution of remote sensing data, this thesis proposes spatial-temporal quantitative information fusion models with high accuracy and great robustness. Moreover, we have extended the traditional fusion framework to meet an integrated fusion requirement of quantitative information from multi-sensor.Here presents some closely related topics in this work as following:(1) This thesis takes the improvement of the spatial and temporal resolution of remote sensing data as the principal objective, mainly focused on the thorough analysis of current status of traditional remote sensing fusion methods, existing downscaling methods and emerging spatial and temporal fusion methods. Compared to the first two methods, the emerging spatial and temporal fusion methods have incomparable advantages and application potential in enhancing the spatial and temporal resolution.(2) The necessary preprocessing technology and the effect evaluation method are presented firstly. The basic framework of spatial and temporal fusion for remote sensing data is introduced in detail. The spatial and temporal adaptive reflectance fusion model (STARFM) and the enhanced spatial and temporal adaptive reflectance fusion model (ESTARFM) are systematically compared. Then, the other spatial and temporal fusion methods based on different frameworks have been roughly classified into three categories:methods related to the physical mechanism, methods based on signal theory and methods designed for special applications.(3) A spatial and temporal reflectance fusion model considering sensor observational differences is proposed. How to utilize the spectral difference between different sensors at the same acquisition time? This is the key to carry out spatial and temporal reflectance fusion and reduce the influence of noise. Firstly, inherent correlation functions between different sensors are calculated. The functions could be applied to reduce the negative influences of sensor design and observation conditions when describing the uncertainties of spectral differences. Meanwhile, in order to consider the different response to land cover types, functions of uncertainty differences are built based on within-class fitting using robust M-estimation. Moreover, a more objective and accurate method for similar pixels selection is proposed. Finally, experimental results show that the spatial and temporal reflectance fusion model considering sensor observation differences can improve the accuracy.(4) A spatial-temporal variation-based fusion method is constructed for land surface temperature. Based on the Bayesian maximum a posteriori probability criterion, the likelihood probability and prior probability density function of imagery were generated. The two functions play important roles in preserving the conformance of the fused image to the observed image and restricting the spatial constraints on the image, respectively, and can be balanced by the regularization parameter. Under the variation-based framework, the solution of predictions can be converted to extreme value optimization problem. The conversion therefore guaranteed the fused data being solved as a whole. In the end, by fusing land surface temperature in different scales, the variation-based method is tested and verified with in-situ measurements.(5) This thesis proposes a spatio-temporal integrated fusion model (STIFM) for quantitative information from multi-sensor. Through the analysis of already existed spatial and temporal fusion approaches for merely two sensors, the concept of spatial and temporal fusion for more sensors is firstly proposed. The framework of the fusion and its corresponding expressions of spatio-temporal weighting function are built. Further, the fused LST values from three sensors are tested and verified with parallel in-situ LST observations.