Recently Light Delectation and Ranging (LiDAR) technology has developed to a new surveying method to acquire surface information. The system includes two types of terrestrial LiDAR and airborne LiDAR. The generation of digital terrain model (DEM) from airborne LiDAR point clouds is one of the most important applications in surveying and mapping. In this paper, we propose an automatic scheme based on object segmentation by Mathematical Morphology methods to filter out object points and use the reserving ground points to generate DEM. The scheme includes 4 main processing steps: (1) generation of initial surface model, (2) outlier removal, (3) object segmentation and (4) selecting ground points. The test data are released from ISPRS Commission III Working Group 3. We follow the quantitative assessment in ISPRS filter test to analyze our processing results. The experimental results and accuracy assessment are presented and show that most of the total errors are lower than the results of some representative filters.

Four multispectral Images (MSI) were acquired upon various viewing angles by the Quickbird-2satellite in passing over airport neighboring fields nearby Fresno, California on July 13, 2003. Theseappear to be the very original, multiangular, multispectral, high spatial resolution images acquired fromspace. The classification experiments are based on the implementation of hidden Markov models (HMM)to design both supervised and unsupervised classification schemes to test the classification performancerelating to the angular and spectral information. For supervised classification, the HMM were applied tomodel the temporal dependencies between multiangular images, while for unsupervised classification, theHMM were adopted mainly to model the spatial dependencies (i.e. contextual information) using proposedobservation-density adjustment methodology. The experimental results reveal that the addition ofmultiangular information to MSI imagery significantly improves the accuracy of scene discrimination. Itis also concluded that HMM is a superior tool for dealing with multiangular, MSI data in comparing to thetraditional supervised maximum likelihood classifier and k-means clustering method.

In this study, a 4.8-ha ground truth site has been established. Within the site, several rice yield scenarios were produced by using combinations of rice varieties, nitrogen rate and drought treatments. Multispectral images by VCDi from ITRC were taken at booting stages of first rice crop season of 2004. Rice yield estimation models, using band ratios of either reflectance or radiance at green, red and infrared band (NIRJRED, NIR/GRN), were derived using ground truth data collected during1999-2002. Yields predicted from image were then compared with yield map measured at 10m x 10mmesh. Results indicated that using airborne multispectral images at booting stage to predicted rice yield was practicable. The prediction accuracy may be improved if better atmospheric correction method were used.

Drought period, in general, the water resource for irrigation must transfer to civilization use, and compel the farmland to fallow. Therefore the water resource arrangement is gradually important in drought management in Taiwan. MODIS image which announced from USA NASA was used in this study, and applying image differencing method to extract thermal band of No.22 was the most useful for drought warning system. Use thermal band of No.22 combination with other bands of MODIS to calculate the monthly drought indicator, such as NTI, NDVI, NDWI, and DI, and taken variation analysis with these indicators in Taiwan. At this study applying multiple regression models to find out the related drought indexes with accumulate rainfall within 30 days, then integrated them to establish the drought-wet index (DWI). Examining DWI with the experimental data, the result shows that it is feasible to apply in drought monitoring in Taiwan. Therefore the DWI could promote the previous model in drought warning system and provide the manager the other information in decision making.

The third band of ASTER sensor provides along track stereo capability. Two images are collected in this band and one looking nadir, one looking backward. Based on the height-base ratio and the 15 meters spatial resolution of the ASTER sensor, the error of the relative DSM generated with the ASTER stereo-pair is expected to range from 10 meter to 30 meters. This study investigates the DEM accuracy with one ASTER scene in the Jeou-Fen-Erh-San area. It is found that the standard deviation is ±11.87m for areas without clouds or other unfavorable factors.

The large-scale maps of built-up area are of special interest if they are to be revised or supplemented with new themes. The very first procedure for map revision is obtaining the exterior orientation. In this research, database information, including a digital map and GIS data, is used for exterior orientation. It would save times using building's comers instead of the control points. Although the accuracy of the building's comers could not be as good as those control points, the uniform-located comers would provide a great quantity of observations which could hopefully increase the reliability. To see if the applied method would work efficiently, several sub-images are abstracted from the original aerial image over urban area, including rectangular buildings with plat roofs, gable roofs and roofs with parapets. The result shows that it is feasible to find exterior orientation by existing digital map and GIS data and would suffice the required accuracy for local revision over 1/1000 map, which is stipulated under 10% of the complement coordinates would be llowed to deviate 20cm exceedingly corresponding to the aero-triangulation points.

Using vanishing point to estimate orientation parameters does not need control point's information. It is applied for the object reconstruction, navigation of autonomous vehicles or robots, camera calibration and length measurement. This research present a method of vanishing point detection based on θ-histogram analysis. It is suitable for close-range building image and without manual interference. If orientation parameters have known can infer object length on single image. For criterion of experimentation result, we compare inferential length with real length and compute RMSE. Average RMSE of using our method in the six model images is ±0.0041m (relative accuracy is 1/48~1/21). RMSE of using our method in a real building image is ±0.0161m (relative accuracy is 1/64~1/27). Even if in the set of lines for vanishing point detection has 13% error, RMSE still can achieve ±0.166 m ( relative accuracy is 1/62~1/28 ). This result shows that our vanishing point detection method is robust.

The rich and detailed spectral information provided by hyperspectral images can be used to identify and quantify a large range of surface materials which cannot be identified by ultispectral images. However, the classification methods that have been successfully applied to multispectral data in the past are not as effective as for hyperspectral data. The main problem is that the training data set does not increase corresponding to the increase of dimensionality of hyperspectral data. Actually, the problem of the "curse of dimensionality" emerges when a statistical classification method applied to hyperspectral data. A simpler, but sometimes very effective way of dealing with hyperspectral data is to reduce the number of dimensionality. This can be done by feature extraction that a small number of salient features are extracted from the hyperspectral data when confronted with a limited set of training samples. In this study, several methods based on the wavelet transforms are developed to extract useful features for classification. Firstly, wavelet or wavelet packet transforms are implemented on the hyperspectral images and a sequence of wavelet coefficients is produced. Then, a simple feature selection procedure associated with a criterion is used to select the effective features for classification. Because the wavelet-based feature extraction optimizes the criterion in a lower dimensional space, the problems of limited training sample size and the curse of dimensionality can be avoided. Finally, two AVIRIS data sets are used to test the performance of the proposed wavelet-based methods. The experiment results show that the wavelet-based methods perform well for dimensionality reduction and also be effective for classification.