Study On The Application Of Multi-source And Multi-scale Images In The Urban Remote Sensing

Today, abundant remotely sensed data with various spatial, radiation and spectral resolutions from multi-platforms provide rich sources of information for the study of urban eco-environment changes under different scales. In order to overcome the limitation of a particular type of remote sensing data in application and take full advantage of other remote sensing data, image fusion technology has been frequently used to enhance resolution of remote sensing data and perform scale transforming among images from different remote sensing platforms. Such processed images can be used to investigate the changes of urban vegetation fraction coverage under different spatial scales and detect the best scale for the study of urban vegetation fraction coverage. The result can provide suitable parameters for assessment of urban ecological environment.Based on a review of current fusion algorithms for remote sensing image, this research used multi-source remote sensing data with various scales to study image fusion technology. The used remote sensing images include IKONOS, SPOT5, EO-1 ALI, and Landsat ETM+. To avoid the spectral distortion of the Synthetic Variable Ratio (SVR) algorithm, this thesis proposes an improved algorithm by using a low-pass filter and histogram matching performance, which is hence named the Synthetic Variable Ratio base on low-pass Filter and histogram Matching (SVRFM) algorithm. The SVRFM can fuse an image without the limitation of band number. Also the algorithm does not need to select reference bands. The SVRFM can keep the best balance between spectral fidelity and the ability of gaining high frequency information. Moreover, a few other fusion algorithms have also been proposed, including Synthetic Variable Ratio base on Index (SVRI) and the IHS-BT algorithm based on the IHS and BT algorithms. The spectral fidelity and the ability of gaining high frequency information were assessed by using visual examination and statistical analysis. The fused images using the new algorithms were compared with those fused using previous algorithms. The results show that the spectral fidelity of the new algorithm is generally better than the previous algorithms.The images used in this study have different radiation resolutions such as 8 bit, 12 bit and 16 bit. In order to overcome the influence of the difference in dynamic ranges of images on fusion effect, a method for the standardization of radiation resolution is proposed. Based on the unifying of quantized value intervals, the transformation from low to high radiation variability through multiplying the same proportion coefficient can reduce the image information loss caused by the data bits trade-off in the fusion process. Meanwhile, whether image radiometric correction is needed before fusion process was discussed in this thesis. The research shows that the high frequency information gains of the fusion images are the same regardless they are radiometrically corrected before or after fusion, and the increase of the spectral fidelity is not obvious. Accordingly, we suggest no necessary to do radiometric correction before image fusion. This can save computation time and avoid the spectral distortion caused by the radiometric correction if a dark object subtraction method is used. Meanwhile, it provides a solution to the problem of the panchromatic image’s radiometric correction when using the FLAASH.On the basis of research above, we put forward the transformation algorithms to unify the scales for the comparison the data at different scales. Using the vegetation fraction with different spatial resolutions derived from a 1:500 topographic map as the reference map, we compared the accuracy of vegetation fraction extracted from the data radiometrically corrected using different models, and conclude the optimal radiometric correction model for the extraction of vegetation fraction. We used six different vegetation indices to enhance vegetation information from multi-scale remote sensing images, and then we calculated the vegetation fraction by using Gutman and Carlson vegetation fraction calculation models, respectively.Through comparative analysis of the experimental results, we conclude that ICM model is the best radiometric correction model for urban vegetation fraction estimation. For high special resolution remote sensing image, NDVI is the best vegetation index for vegetation fraction estimation. While the best vegetation indices for estimating vegetation fraction from moderate spatial resolution images are the RVI and MSAVI. In terms of the study area, the vegetation fraction estimated by Gutman model is more accurate than Carlson model. At the same time, the scales of remote sensing vegetation fraction were analyzed using the vegetation fraction data acquired from the topographic map. This concludes that a 4 m resolution image would be the best scale for the study of urban vegetation fraction. Vegetation coverage should be estimated before converting one scale to the other if this scale transform is needed.