Ground-based air quality monitoring stations were set up traditionally in accordance with population density. It could only sample the air pollutants at selected stationery locations with high operation and maintenance cost. Nowadays, remote sensing technology has been widely applied to monitor the ambient atmospheric environment, this research is intended to develop a portable hyperspectral camera (HyCAM-I) to monitor the air Pollutant Standard Index (PSI) remotely. With the establishment of an air quality estimation model from the hyperspectral data and the sampling calculated hourly PSIhr, we can measure the air quality easier in any ungaged region by mobile HyCAM-I rather than using the traditional site-specific monitoring data. For building up the air quality index estimation model, this study adopted the supporting vector regression (SVR) model and the multivariate linear regression (MLR) model to calibrate the relationship of the measured hyperspectral data and the PSIhr index. Three bands of 500nm, 550nm and 600nm were used as the input variables to estimate the outputs of the PSIhr. Cross validation method was used to verify the model efficiency. The results showed that the estimation of the SVR model may have over learned, and caused the estimation errors of the SVR model more unstable and greater than those of the MLR model. However, the average estimation error of the MLR model is still acceptable but need to be verified by improving the device with expanding the samples in the future to enhance the reliability.

Unmanned aerial vehicles (UAVs) integrating with GPS and INS provide a safe and fast method for data acquisition. The UAVs which can implement automatic pilot promote the automation of data collection. In UAV systems, the GPS antenna and the INS system are not aligned with the perspective center, so that the GPS and INS records should be revised according to the geometry of UAV systems for exterior orientation references. And the cameras equipped with UAVs are often belonging to the non-metric camera, whose interior orientation parameters can be acquired by close-range photogrammetry softwares. However, there are several different camera models used in the softwares and the interior parameters calibrated by different softwares would not be the same, so that the interior parameters of the non-metric camera should not be regard as constant in aerotriangulation. Self-calibration can not only calibrate the camera in close-range photogrammetry but also model and compensate the departures from collinearity in aerotriangulation to promote the positioning accuracy. This study uses the images acquired from a safe UAV system, a Quadrotor UAV, and compares the results by using different aerotriangulation procedures. In this paper, the optimal accuracy can be obtained by using self-calibration in bundle adjustment with all measurements been pre-corrected for radial and decentering lens distortion. And the suboptumal accuracy can be obtained by using Brown’s (1976) added parameters in bundle adjustment, better than the results of using bundle adjustment. But using Ebner’s (1976) added parameters in bundle adjustment cannot help promoting the positioning accuracy.

Currently, the application on the new generation of high-resolution images (such as QuickBird image) to produce of “Paddy Rice Thematic Map” has become widely accepted. But in the application of high-resolution images, the difficulties are encountered on the confused categories and the surplus of variables. These uncertainty factors will involve poor classification outcomes. Therefore, the purpose of this study is to discuss the confused (uncertainty) influence problems of grass, forest and paddy rice classification results on QuickBird image. Hence, this study has three stages to discuss this problem: (1) The high resolution image data by texture information (semi-variogram) is created on the features. (2) The fuzzy membership method is integrated the each category features of spectral and texture information which can be present a normalization data sets. (3) The hierarchical relationships (or connection knowledge) of multi-categories training samples through Apriori Association Rules and Shanno Entropy method are developed to identify the paddy rice, grass and forest on classification rules. This study contributed two points: (1) the process of this study can be separated and presented features knowledge on different categories of object, and this knowledge can be defined a certainty-accuracy of paddy rice GIS map. (2) Using these techniques, the GIS map can provide a better strategy of classification performance.

Recently, remote sensing technology has progressed to permit quicker access to individual tree positions using two-dimensional (2D) information. Therefore, insufficient 2D information has led to subsequent limited applications of the technology. We used ground-based LiDAR technology to map three-dimensional (3D) individual tree positions in the Liukuei forest. The 3D laser scanner obtained highly accurate scans of cloud data, and provided basic information on individual tree spatial distribution. We compared different data collections of tree spatial distribution and forest measurement characteristics from the original field survey using the LiDAR technique. The results showed that the absolute value of horizontal error of less than 0.1 m is 28.26% (13 trees) and vertical error of less than 0.1 m is 32.61% (15 trees). In addition, the absolute value of horizontal error and vertical error of less than 0.5 m were 76.09% (35 trees) and 82.61% (38 trees), respectively. However, there was strong correlation between the LiDAR DBH and field DBH (coefficient of determination R2=0.9685). The LiDAR detection location and an increase in the detecting quantity could obtain sufficient point cloud data. The linear correlation R2 value of the field measured and LiDAR detection was 0.9905, which showed that 3D laser scanning technology provided effective quantification data to reconstruct a tree competition index and tree growth model in a forest ecosystem.

This study aims to assess the land-use dynamics of Shihmen reservoir catchment using photogrammetry and Markov model. The objective was to investigate the effects of human development on landslide. First, ERDAS Leica Photogrammetric Suite (LPS) and Geographic Information System (GIS) were used to interpret the land-use types, and further to examine the land-use changes during 2004 to 2007. A transition probability matrix was then calculated to realize the dynamic relationship between human development and landslide. Finally, the Markov model was developed based on current land status for simulating the future land use changes. The results showed that the study area was classified into five land-use types (i.e., forest, human development, road, water, and landslide). Both human development and landslide areas revealed an increasing trend during 2004 to 2007. The result of transition analysis pointed out that human development has the most conversion into landslide as comparing with other land types. The simulation of future land-use change using the Markov model showed that, the occurrence of landslides in the Shihmen reservoir catchment would be increased based on the current land dynamic.

In this paper, a robust feature extractor technique for FORMOSAT-2 image rectification is applied. The extracted features are required to be computationally attractive and invariant to image rotation, scale change and illumination. Modified Scale Invariant Features Transform method can find good features and matches from different areas, land cover types of FORMOSAT-2 imagery. A proposed procedure improves processing efficacy and performance of image rectification. The result got good performance, precise, and more reliable and removed uncertainty of image processing.