Building reconstruction from aerial images begins often with the search of building edges. But searching of conjugate building edges in multiple images is not an easy task. Since low-level edge extraction in image space can only take consideration of gray-level changes and make very simple linking of edge pixels based on very simple criteria like direction of gradient, gray value consistency along edges, etc., thus no information from object space is considered in that stage, a large percentage of the extracted gray-level edges are not building edges at all. More over, due to noises, shadows, occlusions, weak contrast or even due to complexity of building surface texture itself, most of the extracted building edges are not even complete. They are broken and shown only as piecewise line segments. Since gray-level edges are projected images of true edges in object space, the search for conjugate building edges in multiple images should be more successful if information from object space could be taken into consideration. It is thus the aim of this article to describe how simple object space knowledge of straight edges could help the search for conjugate line segments in multiple images. Functional model for matching of the image space line segments onto object space straight lines are derived and statistical testing procedures for testing the correctness of the matching are developed.

In this study the NOAA satellite AVHRR Sea Surface Temperature data and the Secretariat of the Pacific Community (SPC) longline fishery data from 1996 to 2001 are analyzed to discuss the effect of the El Niño/La Niña events on the longline fishery of yellowfin tuna in the Pacific Ocean. The Empirical Mode Decomposition (EMD) method is used to retrieve Intrinsic Mode Function (IMF) related to the El Niño/La Niña events. The results indicate that the evolution of El Niño/La Niña event causes the decreasing/increasing of the hook rate and westward/eastward movement of the Catch Per Unit Effort (CPUE) zonal center of the yellowfin tuna around the Equatorial Pacific. Meanwhile, a nearly simultaneous response between El Niño/La Niña event and yellowfin tuna longline fishery occurs in the eastern part of the Equatorial Pacific, while about 7-9 months lag for yellowfin tuna fishery in the western Pacific. We also find that the El Niño/La Niña event shows no clear influence on the yellowfin tuna fishery in the Southwest Pacific near Australia.

Multi- polarization SAR imagery data are important for a wide range application such as retrieval of soil moisture, vegetation mapping, and terrain cover classification. However, these alternative polarization data suffer more serious speckle noise caused by the alternating scanning. Speckle noise is inherently induced from the coherent processing of the backscattered signals and substantially degrades the effectiveness of images interpretation and target recognition/identification. To facilitate scene interpretation and qualitative applications of such SAR images, the speckle noise must be reduced to an acceptable level while preserving the polarimetric properties between different polarization data. In this article, we proposed an effective approach that stems from a processing a series of images. A linear transformation that decorrelates the multi-polarization was carried out, followed by the Lagrange multiplier method to find a set of optimal filtering weights that are related to local image statistics. The effectiveness of the method was verified using the ENVISAT-1 Alternating Polarisation Mode HH/HV images acquired on 16th June 2003 over Taiwan. It is found that the equivalent number of looks (ENL) is greatly improved from 3 to 6, and the strong targets are well preserved without scarifying the spatial resolution.

A field spectroradiometer was used to measure the spectral reflectance of water body. The reflectance was calculated as the ratio of surface water radiance to the standard whiteboard radiance nearly measured at the same time. Water samples were taken simultaneously for determining their chlorophyll-a, suspended solid and transparency. The relationships between those water quality parameters and spectral reflectance were analyzed using stepwise multiple regression to derive optimal prediction models. The multiple regression was also applied to the SE-590 simulated SPOT bands. The SPOT image of the same day was also analyzed using the same method to compare the statistical results. It shows that the multiple regression models using the SE-590 reflectance data can get the best water quality prediction results. The evaluated RMS error of chlorophyll-a, suspended solid and transparency of water quality parameters are 0.57 ug/l, 0.2 mg/l and 0.17 m, respectively, and the RMS errors are 1.03 μg/l, 0.51 mg/l and 0.32 m for SPOT data, respectively. The SE-590 simulated SPOT three bands data obtained the worst results because the RMS errors are 1.77 ug/l, 0.49 mg/l and 0.37 m. Further, The band ratio of red/green and red/NIR were used to detect the chlorophyll-a relative distribution qualitatively, and assures that the ratio is less, the chlorophyll-a is abundant in TEKEE reservoir.

This investigation is to compare the geometric precision of Rigorous Sensor Model and Rational Function Model for QuickBird images. In the Rigorous Sensor Model, the on-board data are used to initialize the satellite orbit. Then, we employ the ground control points for preliminary orbit fitting. Finally, a least squares filtering technique is applied to collocate the orbit. In the Rational Function Model, the Rational Polynomial Coefficients provided by the satellite company are used to build up the transformation between object space and image space. Then the bias of the transformation is compensated by an affine transformation using ground control points. Experimental results indicate that, both of the RSM and RFM may reach an accuracy of 2 pixels when 9 ground control points are employed.

The evaluation of topographic change has been done on basis of 3D laser data and aerial photogrammetric survey. The newly-developed, time-saving approach is by using airborne and ground-based Lidar. A points-cloud of high precision and density distribution can be obtained in a few minutes for the sensitive latent landslide area. With a limited control survey of precise targets using GPS and total stations, multiple scans of ground-based Lidar can be registered together to form one cohesive 3D model. And, thus, overlaid the different period topographic data for estimating the changes of the target areas. This is conducted to observe the deformation of a large-scaled dip-slope landslide of Jiu-fen-er mountain, which was triggered by the big shock of the Chi-Chi earthquake at Nantou County of central Taiwan on 1999/09/21. Control points are systematically distributed on the 1000m by 2000m landslide surface. Surveys are made by transits of total stations, GPS, airborne Lidar and ground-based Lidar data. The terrain changes on ground surface have been detected in thus a scheme. It is found the largest and average changes of height between the earthquake events are about 60m and 38m. The effectiveness of applying 3D laser scanning is proved in this study.

Remote sensing applied on monitoring environmental disaster and ecological impact has been widely used in Taiwan. However, the recovery of vegetation was identified as one of the key factors contributed to establish an environmental risk assessment model. In reality, it was so difficult to choose an optimal vegetation index equation used to quantify the vegetative factor. Spatial Modeler Language（SML） based on ERDAS IMAGINE software was used to develop eight different vegetative index transformative equations from different aspect of eight researchers. Also, 118 disaster areas rehabilitated by vegetation were chosen as studied samples in this study. Test of homogeneity and analysis of covariance （ANOVA） were also utilized to identify the most suitable equation for estimating vegetation index in studied area. As a result, soil adjusting vegetation index (SAVI) equation developed by Huete was much better than others on the basis of considering atmosphere scattering and ground soil image reflection. This result offers a good reference to help us choose SAVI as one of the best model can be used to estimate the vegetation index. This index became a key factor contributed to create the environmental risk assessment model.

When applied to remote sensing imagery classification, the conventional neural network classifier represents training information in a one-pixel-one-class basis. Therefore, the class mixture of a pixel cannot be taken into account resulting in poor classification accuracy. Based on the original dynamic learning neural network (DL), this paper introduces a fuzzy dynamic learning neural network (FDL) in which the training information is represented as fuzzy sets. In order to add the fuzzy information to the neural network, a fuzzy c-means clustering algorithm is used to assign the degree of membership of each pixel during the training stage. In this paper, the FDL is applied to a SAR image classification to evaluate the selection of the algorithmic parameters including the membership weighting exponent and the measure of dissimilarity. Furthermore, comparisons between the DL and FDL are made. The effectiveness of the combination of the neural network and fuzzy sets theory is demonstrated by an example of SAR image classification. Experimental results show that the separatability between the similar classes and the classification capability of the class-mixed pixels are improved. Moreover, the classification results match better with the ground truth.