Research Of Construction Land Recognition Based On Remote Sensing Data Fusion

Images from SPOT and Landsat satellite series are the main data sources in land use survey. In order to extract the historical period land cover information better and make full use of the remote sensing data sources, it is necessary to merge data from different sensors. Because of the limited optional data sources in a specific period, how to make best use of the existing remote sensing data for information recognition has aroused public concern.West area of Tai hu is a part of the seriously polluted area main surveyed in water special project. In order to extract the land cover information of 1999 in this area and to be compared with the current land cover, this paper used SPOT1 which has the highest spatial resolution at that time and Landsat7 to extract the land cover information. Methods of how to integrate the information of these two type images are studied to take advantage of both dataset and improve the spatial resolution and spectral information for improving the feature recognition accuracy.Multi-sensor data fusion and its impact to the construction land recognition are the main research contents in this paper. The main works and conclusions including,(1) It has systematically illustrated and summarized the qualitative and quantitative evaluation methods for the fusion images. And quantitative evaluating indexes have been selected both from spatial and spectral information aspect to inform the fusion image evaluating system in this paper.(2) The panchromatic image of SPOT 1 and multi-spectral image of Landsat7 are used in image fusion experiment and the results are evaluated by relevant indexes. The evaluation shows:the fusion images by high pass filter method,PCA transformation method and GS transformation method have reserved the spectral information best, weighted method,Brovey transformation method and HIS transformation method can enrich the spatial information and increase the contrast of features in the image, wavelet transformation method based on PCA or HIS which shows the richest texture details has inherited the most spatial information.(3) New images have been constructed by the characteristic bands of construction land, and construction land has been extracted from new images by maximum likelihood method. As a result, except the wavelet transformation method based on HIS transform, other image fusion methods, such as weighted, high pass filter, Brovey transformation, HIS transformation, PCA transformation, GS transformation and wavelet transformation method based on PCA transform, all improved the construction land recognition accuracy. Among them, PCA transformation and GS transformation method improved the construction land recognition best, the accuracy of construction land are increased from 76.92% to 85.03% and 83.50% respectively, and the wrong extracted percent are lowed from 29.99% to 16.25% and 15.30%.This study shows that construction land recognition by history images can be improved by multi-sensors image fusion. Different image fusion methods enhance the different image information, so property image fusion method should be used according to practical application.