The cyber city, a replica of the real one, has demonstrated its potentials in urban and environmental planning, design, construction, and management. Building model is one of the most important elements in a cyber city. Traditionally, the reconstruction of building models is performed by using aerial photography. As an emerging technology, the airborne LIDAR (LIght Detection And Ranging) system provides a promising alternative. Hence, this investigation utilizes the airborne LIDAR point clouds for building reconstruction. The major contribution of this investigation is to propose a solution for the reconstruction of complex buildings. This investigation presents a scheme for the reconstruction of building models from airborne LIDAR point clouds and topographic maps by using a divide-and-conquer strategy. The proposed scheme comprises three major parts: (1) decomposition of building boundaries (2) shaping of building primitives, and (3) combination of building primitives. In the decomposition of building boundaries, the airborne LIDAR data is selected to extract the inner structure lines. Then, the split procedure divides the building boundaries into several building primitives using the extracted feature lines. In the shaping of building primitives, the parameters fitting is applied to shape the roof for each building primitive from airborne LIDAR point clouds. The roof shapes include both planar and circular types. Finally, a least squares adjustment considering the co-planarity and co-linearity is used to merge the 3-D building primitives into building models. The proposed method has been tested with the data collected in Taipei city of Taiwan. The reconstruction rate is better than 92% while the omission error is smaller than 5%. The shaping error is 55cm. The planimetric and vertical accuracy of the reconstructed models are better than 40cm and 70cm, respectively. The experimental results confirm that the proposed scheme produces high fidelity models.

Three dimensional building models provide spatial information for decision support. It is more preferable to maintain a building database by the firstly detecting the changes followed by a reconstruction procedure. Change detection is traditionally done using multi-temporal images through the spectral analyses. Those images provide two-dimensional spectral information without including shape in the third dimension. As the availability and quality of emerging airborne LIDAR systems that make the acquisition of shape information convenient, we use new airborne LIDAR point clouds and aerial photos to detect changes for building models. The proposed scheme comprises data pre-processing and change detection on building areas. In the first step spatial registration is performed, ground area and vegetation area are detected, the Lidar points on the ground surface and in vegetation areas are removed, and height differences are calculated. In the second step, shape and spectral information are integrated to determine types of change. The validation for determination of change types shows that the results can reach 85% overall accuracy. To provide comprehensive observations, those unreliable results are scrutinized. The ratios of the “limitation of method”, “limitation of data”, and “data quality effect” categories are 38%, 11%, and 51%, respectively.

Hyperspectral images are able to provide the detailed spectral information necessary for the discrimination of different land targets in various kinds of remotely sensed images. The linear spectral mixing model is a widely used discrimination method for modeling the variety of multiple land targets in hyperspectral images. Basically, the linear spectral mixing model is solved by least squares adjustment to acquire the minimal-error solutions. It is believed that the reduction of the noise that inherently exists in each spectral band of the hyperspectral image can increase the discrimination accuracy. In general, the noise recorded in the hyperspectral sensor is caused by the absorption or scattering effect of the atmospheric particles during the energy transportation process. In this study, a noise filtering preprocessing based on empirical mode decomposition (EMD) is proposed. The purpose is to reduce the inherent noise and further minimize the residuals of least squares solutions for hyperspectral images. The test results with simulated images indicate that the proposed noise filtering process can effectively decrease the abundance errors. Moreover, in the experiment with real image, the residuals of least square adjustment for noise reduced data set are also smaller than that for original data set.

The image classification is the most important issue in Remote Sensing. In essence, extraction of regional feature is an important process on human’s image classification process to form their effective identification of the target objects. However, vision cognition of machine learning follows Pixel-based which disobeys the rule of human’s image classification process. To resolve the forgoing problems, this study proposes a Region-based approach for Regional Object Classification (ROC). This method is a hyper module which applies an unsupervised method (Region-Growing, RG) to a supervised approach (Maximum Likelihood Classifier, MLC). New concept can effectively reduces a Salt and Pepper effect of classification result from very high resolution image by conventional pixel base classifier. Accordingly, a module of Regional concept is applied to improve its classification accuracy. Our outcomes are used by mean of the ADS-40 data to present the classification performance. Finally, this study utilized the ROC classifier and the overall accuracy of region based concepts (95.3%) is better than those of the pixel-based concepts (92.6%) on the evaluation of paddy rice from ADS-40 image. This result shows that an appropriate progress through regional module can effectively improve the accuracy of image classification.

Image registration is a key issue in many image processing applications in remote sensing. Examples of these applications include change detection using multiple images acquired at different times, and fusion of image data from multiple sensor types.SIFT (Scale Invariant Feature Transform), Canny feature matching and least-squares area matching method are proposed in this research. At first, the initial matching point pairs are detected from manual adjustment or the SIFT algorithm, which is invariant to image scale. And then, edges in both images are located by using the Canny algorithm and broken contours are cleaned. Furthermore, more matching point pairs are selected using a cost function that measures the gradient orientation and distance between all possible pairs of the points. Pairing image windows are built and segmented to get radiometric parameters, and the radiometric parameters are used here to modulate the slave image window. Finally, master image window and modulated slave image window are matched by least-squares matching, and conjugate points are found. The Thin-Plate Splines (TPS) and blunder removal methods are used to register master and slave images. Experimental results show that numerous matching points can be obtained correctly and automatically, and different satellite images can be registered precisely.

LiDAR is a very useful tool on monitoring coastal geomorphic change and has been introduced into Taiwan in recent years. For making the best use of the LiDAR-derived data on long-term coastal change in Taiwan, this research intends to reveal the possibility and limitation of overlapping LiDAR-derived DTM with preexisting pictorial materials, in particular Orthophoto Base Map in the coastal area (OBMC, scale 1:1000, published in 1980s). The study area is the Bali coast, located to the south of the Dan-shui River mouth.Two types of LiDAR data, provided by Central Geological Survey, MOEA, - LAS file and 2m DTM were used, and 1-meter contour line was adopted as the datum-based shoreline indicator. DTM were also constructed by interpolating contour lines of Orthophoto Base Map. Results show that to combine these two sets of pictorial materials to detect shoreline change and calculate volume change of beach and sand dunes is practicable. The amount of shoreline change is from 53 m to 65 m over two decades along the study area. Many details should be taken into account, however, when detecting volume change of beach/sand dune due to the different resolution and accuracy of the two sets of data.So far, most pictorial materials in Taiwan are not produced for the sea shore zone in purpose, and cause many problems on shoreline mapping work. Thus, it is expected that the authority can set up standards and/or guidelines of shoreline mapping (including LiDAR) in the near future.