Comparison on fusion algorithms of ZY-3 panchromatic and multi-spectral images

Abstract: ZY-3 (Ziyuan-3) satellite is China's first civil high resolution mapping satellite that can provide services for precision agriculture in our country in real-time and steadily. Based on ZY-3 satellite panchromatic images with 2.1 m spatial resolution and multispectral images with 5.8 m resolution data, the methods of image enhancement on ZY-3 agricultural land data were analyzed, taking agricultural land in the Caidian district of Wuhan, Hubei province as an example. The study used different fusion methods such as HSV transformation, Brovey transformation, Gram - Schmidt spectral sharpening, PC spectral sharpening, Wavelet transformation, and Ehlers transformation, all of which are frequently used. For assessing the performance and effect of these fusion methods, the image quality of the fused image was evaluated by qualitative and quantitative analysis. Qualitative analysis relied on visual contrast. To improve the spatial resolution and keep the original image spectrum information of fusion images, two kinds of statistical aspects: grey scale statistics and texture characteristics of different areas, were analyzed quantificationally. Grey scale statistics included correlation index and spectral angle index. Texture characteristics included entropy index, second moment index, and edge response index. In order to better determine the fusion image's texture features, three block areas of which the spatial structure's complexity was different from each other, such as building area, farmland, and water, were chosen to analyze their entropy and second moment indexes. Furthermore, through the object-oriented classification method, fusion images were classified into six classes to evaluate the performance of the fusion method at the information level. The six classes included water, arable land, woodland, buildings, bare land, and unclassified. In order to further verify the accuracy of classification, the study used the same methods and parameters to analyze other images of the Wuhu province. Based on the above content, this paper analyzed the applicability of 6 common fusion methods for ZY-3 panchromatic and multispectral agricultural land images. Experiments showed that PC transformation and Ehlers transformation had high spectral fidelity, rich spatial information, and high classification accuracy simultaneously. Brovey transformation and HSV transformation lost serious spectral information, and Brovey transformation also lost great spatial information. The overall classification accuracy of Brovey transformation was poor, but the HSV transformation was good. For GS transformation, the overall analysis result of the classification analysis was bad, but the image quality performed well. The Wavelet transformation lost serious high frequency information, but its spectral fidelity kept well, as did its classification result. Classification experiments introduced details that Brovey transformation was suitable for extracting water. PC transformation, GS transformation, and Wavelet transformation were suitable for building information. For arable land information, priority could be given to Ehlers、PC、and Wavelet transformations. For woodland extraction, Ehlers and Wavelet transformations had better accuracy. HSV transformation was suitable for the extraction of bare land. It is noteworthy that different fusion methods have different advantages in image quality and features extraction. Based on the above reasons, we can select relevant fusion algorithms combined with the practical agriculture application and image information. Comparing experiment data comprehensively, the effects of PC transformation and Ehlers transformation are superior to that of other fusion algorithms, suitable to be applied in ZY-3 panchromatic and multispectral agricultural land data.