Feasibility Study of Detection Specific Urban Features Using High Spectral Resolution Satellite Imagery, LIDAR Data, and Fully Polarimetric RADAR Imagery

The urban area has always been under the influence of population growth and human activities. This process causes the changes in land use/cover. Thus, for optimal management of the use of resources, it is necessary to be aware of these changes. On the otherhand, satellite remote sensing has several advantages for monitoring land use/cover resources, especially for urban areas. In this regards, classifiying urban area over time present additional challenges for correctly analyzing remote sensing imagery. Nowadays, integrating different kinds of data and images, achieved by the different remote sensing sensors are known as a suitable solution for extracting more useful information. The passive optical sensors have been used extensively in mapping horizontal structures. However, radar data could be used as a complementary data, since these data would be gathered in different climatic conditions in 24 hours of a day, as well as some geo and manmade structures have a specific response in the radar frequency. Furthermore, LiDAR data could gather precise measurements from vertical structures. Hence, by integrating optical, radar, and LiDAR data more features and information would be prepared for different kinds of applications. In this research, we used these data sets to detect buildings, roads, and trees in a complex city sense, i.e., San Francisco, by generating 141 features, in passive optical sensors using high spatial resolution WorldView-2 imagery (image bands, vegetation indices, IHS color space, YIQ color space, YCbCr color space, the first order statistical features and the second order statistical features), also in LiDAR data (first and last pulse, first and last intensity, nDSM, NDI, slope 4 neighbor, slope 8 neighbor, aspect, roughness, smoothness, surface curve, profile curve, variance and laplacian) and in RADAR data using RADARSAT-2 (amplitude, phase, intensity, incidence angle, imagery part, real part, radar cross section, polarized HH, polarized VH, polarized HV, polarized VV, ratio element of scattering matrix, alpha, beta, pauli coefficient, krogager decomposition coefficient, freeman decomposition coefficient, yamaguchi decomposition coefficient, antropy, eigen value and anisotropy). We divide our merging data set to four regions. The first region include building feature, the second region include building and vegetation features, the thid region include building and road features and the forth region include vegetation feature. All thsese features merged and produc the cube of data with 141 dimension number. Then, by using the principal component analysis (PCA) feature extraction method, as well as the well-known intrinsic dimension (ID) methods, including second moment linear (SML) and noise whitened HFC (NWHFC) dimensionality of these data sets is reduced. Finally, the supervised classification method k-nearest neighbour (K-NN classifier) was utilized in order to detect buildings, roads, and trees and grouping features according to the earned accuracies. In this regards, the thirty present of ground truth data was used as traning data sets and remaing seventy present as test data sets. In addition, the fusing and merging these data sets (buildings, roads, and trees) reveal the superiority of the implemented method to classify map with overall accuracy by a margin of nearly 90% using proposed approach and support our analyses.