Fusion Of Hyperspectral And LiDAR Data For Urban Land Use Classification

With the increasing population of the city, urban land use classification has become a very important research topic. Acquiring land use information timely and accurately could provide government an important basis for urban planning, basic geographic information updating, land resources survey, disaster response, environment resource protection, land dynamic monitoring and analytical decision making. With the advantages of macro and fast, remote sensing could acquire urban land use information in different spatial scales, and it has become the most effective tool for urban land use classification. So far, many studies have been exploring the application of remote sensing technology for land use classification. However, with the continuous development of urbanization, the buildings are becoming higher and higher, and the object types are becoming more and more complex. Therefore, there is a serious shadow problem in high-resolution remote sensing images. Furthermore, over the past twenty years, development of remote sensing classification technology could not improve the overall accuracy. The urban land use classification of passive airborne/spaceborne remote sensing has reach a certain limit. Research has shift the emphasis away from classification algorithm improving to multi-source remote sensing data fusion.Compared with multi-spectral remote sensing data, hyperspectral data could access hundreds of bands in the electromagnetic spectrum range. It could obtain continuous spectrum of the surface feature which could classify spectral-similar surface features. Airborne LiDAR could provide precise three-dimensional information. It could extract features with elevation information and could eliminate the shadow problem. However, LiDAR lacks abundant spectral information. The fusion of LiDAR-hyperspectral data could leverage their complementary advantages. It could not only make fully use of the abundant spectral information of hyperspectral, but also the height information of LiDAR.Houston of Texas in USA and Zhangye in Gansu province of China were taken as subjects investigated. Using airborne LiDAR and hyperspectral data, remote sensing fusion and urban land use classification were conducted.The main constructions of this paper are as follows:1. In order to utilize the height information of LiDAR data, the abundant spectral information of hyperspectral data and overcome their disadvantages, this paper attempt to use the fusion of LiDAR height, intensity information, and hyperspectral NDVI and GLCM information data for urban land use classification. The results show that:compared with hyperspectral data-alone, it is effective for urban land use classification using fusion of hyperspectral-LiDAR data, especially in the spectral-similar urban land types with different height. This is mainly because the fusion data not only use the height information of LiDAR data, but also use the abundant spectral information of hyperspectral data.2. For the first time, SVM and object-based classification method are combined for feature-level fusion of LiDAR-hyperspectral data, and then are used for urban land use classification. Pixel-level fusion of LiDAR and hyperspectral data has its limitations, It couldn’t only make fully use of the intensity and elevation information of LiDAR data, but also clouldn’t make fully use the abundant spectral information of hyperspectral data. The results of the study show that:this method not only retains the advantages of SVM method in hyperspectral data classification, but also utilize the attribute rules in object-based classification method, such as length, area, shape. The combined SVM-object based classification methods obtain a remarkable classification result.3. The fusion of LiDAR and hyperspectral data could deal with urban land use classification in cloudy condition, and it also could solve the shadow problem. Passive remote sensing data are vulnerable to the weather (etc. cloudy weather), which could lead the lack of spectral under the cloud. Furthermore, with the increasing of building height, there is a serious shadow effect in high spatial resolution remote sensing images. While as an active remote sensing technology, LiDAR is not easy to be affected by the weather. And the accurate height information of LiDAR data could eliminate weather effect and shadow problems. In addition, SAR and hyperspectral data fusion may also solve this problem, and it will be a research direction,...