Multi-temporal images and cadastral parcels were combined in a pilot study for detecting rice crops and updating the GIS database of landuse parcels. Both unsupervised and supervised classification scheme in a pixel-by-pixel mode and a parcel-based mode that using a sing1e time or combined multi-temporal images were adopted in Lu-Kang study area. Cadastral parcels were rasterized into a 2.5m grid for the classification. It is shown that the best result was obtained by applying the parcel-based supervised classification with NDVI difference images, evaluated on basis of clssification accuracy and index. Images from two other test areas in Chang-Hwa County are analyzed using this method for verifying the possibilities of adopting common criteria. Results include: (1) A classification accuracy of 88.71 % with 0.77 can be achieved for the 2nd crop period in 1997. An accuracy of 89.01% with 0. 78 was obtained if the interpreted criteria was adjusted according to the growing period of North, Middle, South area of Chang-Hwa. (2) The classification accuracy was 85. 64% with 0.71 for the 1st crop period in 1999 using the same classification method. A improvement of accuracy of 0.8% was obtained by using three different interpreted creteria.

This paper begins with elaborating the deduction and solutions of Radargrammetric Condition Equations on a single SAR (Synthetic Aperture Radar) image, which is the time-varying SAR airborne flight locus represented with polynomial collaborating with the mixed adjustment method. As for the blunder detection in observation data, in order to apply τ -test in detecting the possible blunder in observation data to assure its quality, the orientation parameter used to be described with two-order polynomial in previous researches. In this paper, one appropriate high-order polynomial is chosen in the first place to proceed with the optimization of the orientation parameter originally presented by two-order polynomial. The result of the experiment has shown that the check point precision is 4.5-5.5 meter horizontally and 5.0 meter vertically, which proves to be better than two-order polynomial. This is going to make positive contributions toward the precision promotion in the subsequent researches on geocoding of SAR image.

A Kodak DCS210 digital camera is calibrated with the test field established in the exterior of a building in National Chiao-Tung University, Hsin-Chu. The three dimensional coordinates of 79 points located on the building are measured with a total station. The accuracy of the coordinates is within 1 cm. All these points are chosen from the corner or the intersection of edges. The experiments indicate that the sub-pixel level accuracy (0.15 pixel) can be achieved with proper mathematical modelling. In the object space, the relative accuracy of 1/330 with respect to the average object distance can be achieved. Regarding to the lens distortion, radial distortion is the largest to the magnitude of several pixels.

The purpose of this paper is to develop a new method to extract features from the scanned thematic map. Features to be extracted are the urban, road, river, and lake. The method is divided into two parts. One is the color classification, and the other is the feature extraction of binary classification image. To perform color classification, a supervised neural classifier, called Dynamic Learning Neural Network (DLNN), is used, while feature extraction is done by means of segmentation and erosion. The test data is a scanned image of 1/50000 thematic map in Southern Taiwan. Results demonstrate that the method is both effective and efficient in extracting features of interest in the scanned image.