Due to the advance of digital photogrammetry, close-range photogrammetry steps into the door of automation. In close-range photogrammetry, artificial targets are frequently used for measurement. It was inefficient and time consuming in the past when there is no choice but manual measurement. Nowadays, automatically determining the photo coordinates from a digital image has been applied for years. However targets are just positioned but not identified by the current systems. Therefore, the positioned targets in an image are not directly associated with the ground targets and the conjugate target points in other overlapped images. This paper proposes an approach to position targets and recognize the identifications of the positioned targets as well. The work can be divided into three steps： searching, localization, and identifying targets. First we designed a special target which consists of two digits and two round-shape targets. Then an ad hoc circular-target detector is used to search all possible round-shape targets in the image. The centers of the detected circular-targets can be precisely located by using the least squares matching. Based on the known relationship between the two digits and the two circular targets, the boundaries of the digits can be determined. Extracting the subimage of the digits and calculating the normalizeιcross correlation coefficients with the standard templates then the digits could be identified. To verify the proposed method, we conducted three experiments The first experiment is to identify targets in an image with uniform white color background. All the targets were successfully and correctly identified in this experiment. The second experiment is designed to find targets in different scales, so that several sets of targets attached in the vertical sides of stairs were imaged and tested. Targets tend to be misinterpreted when their scales are getting small. The third experiment is designed to identify targets in an oblique image. This experiment shows that correct results can be obtained when the image angle is less than 30 degrees.

A three-year study was carried out to develop a model for estimation of rice growth from ref1ectance spectra of ground vegetative cover. Field measurements were conducted at the experimental farm of Taiwan Agricultural Research Institute, Council of Agriculture at Wufeng, Taiwan (240 02'N, 120040'E, elevation of 85m) during the first and the second cropping seasons in 1996, 1997, and 1998. It showed that the typical curve of reflectance spectrum of rice vegetative cover (canopy) has a differential reflectance pattern in the varied wavebands as well as in different growth stages. By using the first order differentiation in cope with peaks and valleys observations from spectral waves, three characteristics wavelengths of 554, 674 (RED), and 754nm (NIR) were selected from turning points of reflectance spectra. Their seasonal changes were curvilinear ref1ectances of 554 and 674nm fell off to the lowest near heading and then climbed up as plants reached maturity； the trend was reversed for the near-infrared reflectance Seasonal changes of normalized difference vegetation index (NDVI) show a similar curvilinear shape in both First (R2 = 0.919) and Second (R2 = 0.931) crops from 1996 to 1998. Relations of the measured physical growth characters and NDVI were exponential, the determining factors (R 2) were 0.811, 0.809, and 0.669 for leaf area index (LAI) leaf number (LN), and leaf dry weight (LDW), respectively. The exponential curves estimate fairly accurately the advancement of growth characters from spectral transformation (NDVI), especially LAI.

Under mid-low wind and strong current conditions, the sea bottom topography in the shallow water can be showed on the radar images. These kind of images had been observed on side looking airborne radar (SLAR)； real aperture radar (RAR)； and synthetic aperture radar (SAR) images. Taiwan Shoal is a wide sandwave area of 10,000km2 in the south-west of Peng-Hu island. In this thesis, the model proposed by Alpers and Hennings (1984) was applied. Water depth measured by Ocean Research III； calculated current field and wind speed measured by anemometer in Dong-Ji island are used as input data. And the comparison between the model estimated result and the ship-measured depth are made. The comparison shows a good agreement between model and in-situ measurement. And the further efforts for two dimensional model can make the method more useful.

Landsat TM data in Fusan, Chichiawan, Sandiman and Lisan watersheds are used to study the affection of geomorphologic factors (aspect slope, elevation, flow accumulation, topographic index) to the geographic distribution of normalized difference vegetation index (NDVI). Raw Landsat TM data are corrected by Minnaert model to mitigate topographic effect. Geomorphologic factors are calculated by Digital Terrain Model. Results show that aspect strongly affect distribution of NDVI. High NDVI pixels appear on south-east aspect. Slopes can be divided into of illuminate and non-illuminate side. Slope angle has comparative low affection to NDVI, but all the data show a peak value in 30° ~ 40° . NDVI increase with elevation in low altitude area, and decrees with elevation in the area higher than 3,000m. Flow accumulation (FAC) value indicating pixel position is either near ridge or bottomland. It helps to find a fact that highest NDVI is on ridge pixel and then decrees toward bottomland along a slope. The topographic index is believed to relate to topsoil moisture content. NDVI are decrees with topographic index in all basins in positive index case.