Comparison And Application Of High-resolution Image Fusion From The Same Or Different Sensors

With the improvement of the capability of capturing remote sensing images and the imaging technology, the number and the resolution of images are increasing. And the high-resolution remote sensing images gradually become the main data on the social development, national defense security, agricultural production, disaster monitoring and other needs. The fusion of remote sensing images can integrate the useful information of different images to produce a new remote sensing image whose information is more suitable for remote sensing applications, and it can improve the utilization of remote sensing images. The wide application of remote sensing images promotes a large number of image fusion algorithms, and it is an important issue for us to choice the suitable image fusion algorithm.Based on the original images from the same or the different sensors, remote sensing image fusion can be divided into two board categories:fusion of images using same source data and using different source data. In this paper, we made a research on the comparison and analysis of the appropriate fusion algorithm to fuse high-resolution and multi-spectral remote sensing images from same sensor or different sensors based on the commonly used fusion algorithms. In addition, a research on green land information extraction by choosing fusion images from same or different sensors with the best fusion effect was made. We used an area of Hefei Economic and Technological Development Zone and Feixi County border region as study area and the panchromatic and multispectral images of IKONOS, SPOT6 and GF-1 images as the data source to make study on choosing the appropriate fusion algorithm from same or different sensors and the effects of the green land information extraction from fusion images. The main research comments contents were as follows:1. A study was made on choosing appropriate fusion algorithm using high-resolution remote sensing images from same sensors. We can obtain high resolution and spectral images from the same sensor at the same time which can be fused leading to a wider range of application. From two aspects of spatial and spectral information, we used a panchromatic image and a spectral image of IKONOS satellite as the same sensor data to make qualitative and quantitative analysis on the fused images by seven commonly used fusion algorithms including the HSV transform, Brovey transform, Wavelet Transform (WT) and Principal Component (PC) Transform, Gram-Schmidt, Pan Sharpening, Color Normalized (CN) transform. On this basis, the comparison and analysis of classification accuracy of fused images were made. The results showed that fused image from same sensor by PC transform whose total classification accuracy was high to 88.18% have good effect from three respects of spatial information, spectral information and classification accuracy.2. A study was made on choosing appropriate fusion algorithm using high-resolution remote sensing images from different sensors. The height of acquisition platform, running speed, observation scope and resolution of the acquired images from different sensors are all different, so the remote sensing images from different sensors were harder to be fused than the images which were obtained from the same sensor. In the research, we used a panchromatic image of SPOT6 and a multispectral image of GF-1 as the data sources with high spatial resolution from different sensors to make evaluation on three fusion algorithms including the Wavelet Transform (WT) and Principal Component (PC) Transform and Gram-Schmidt from spatial and spectral information. The direct and indirect image fusions were adopted to make analysis of the fusion algorithms based on the high-resolution image fusion from different sensors. The direct image fusion used the panchromatic image of SPOT6 as the high resolution image and the multispectral image of GF-1 as the spectral image; the indirect fusion used the panchromatic image of SPOT6 as high resolution image and the fused image by the panchromatic and multispectral image of GF-1 as spectral image. The results showed the quality of fused image by WT transform was the best in the experiment of direct image fusion; the qualities of images which were indirectly fused were all superior to the images which were directly fused, but it’s hard to decide the best one in the indirect fusion.3. A study was made on the green land information extraction using high-resolution fusion images derived from the same or different sensors. Fusion of high resolution images from same sensor used the panchromatic and multispectral image of SPOT6 as the data source, and indirect fusion of high resolution images from different sensors used the panchromatic image of SPOT6 and the multispectral image of GF-1 as the data source. And the same resolution of 1.5 meter of the two fused images avoided errors due to different scales. The result showed the classification accuracy of extracted green land information of high-resolution fusion images from same sensors was 97.27%, and the classification accuracy of extracted green land information of high-resolution fusion images from different sensors was 76.88%. The green land information of high-resolution fusion images from same or different sensors was both well extracted. And the effect of green land information extraction in high-resolution fusion images from same sensor was better. The vegetation with the little area which grew in the residential area, the middle of roads and two sides of the road was difficult to be extracted. In the high-resolution fusion images from same or different sensors, the vegetation which grew weaker was not extracted, and some roads of long shape and in the location near vegetation were divided into vegetation by mistake.