This study is the first part of a research on developing to a practical procedure for estimating NDVI using AVHRR data. In this part (Ⅰ) the focus is on the geometrical registration problem. The registration of multi-temporal AVHRR images is a very important task for the long-term NDVI estimation. Basically the accurate geometrical registration is the base of the future NDVI analysis and monitoring. In this study, we use one referenced orthimage of Landsat TM and NCC matching method to registrate AVHRR images. For avoid the atmospheric effect, the NDVI images estimated with AVHRR are used in the matching step. Meanwhile, the modified 2 threshold values method is used to mask the cloudy pixels. The result shown that the sub-pixel geometrical registration accuracy are gotten, and this producer proposed in this study is practical for low-resolution eteorological satellite images.

Recently due to the advance of image scanning technology, hyperspectral image scanners which have tens or even hundreds spectral bands have been developed. Comparing to the traditional multispectral images, hyperspectral images have richer and finer spectral information than the images we can obtain before. Theoretically, using hyperspectral images should increase our abilities in classifying land use/cover types. However, when traditional classification technologies are applied to the analyzing processes of hyperspectral images, people are usually disappointed by the consequences of low efficiency, needing a large amount of training data, and barely improvement of classification accuracy. In order to solve this problem, technologies of hypersepctral data analyses and processes must be developed. First, this paper illustrates the characteristics of three different spaces (Image, Spectral and Feature) in which hyperspectral data can be inspected. Then, some fundamental statistical theories and graphical presentations of the statistics for hyperspectral images are introduced. Then, spectral differences are analyzed in difference spaces. In addition to spectral differences, differences between two hyperspectral data classes are also analyzed.

Automatic building reconstruction by simple edge matching encounters many difficulties. The major one is that most of the edges extracted from an image by a low-level linear feature extraction method do not belong to any building edge at all. Moreover, an edge which appears to be straight in the image might actually not be straight in the object space at all, or it might even be from two totally different edges in the space. All these make simple edge matching ambiguous and very difficult to obtain a correct result. In this paper we used large-scale aerial images of a suburban area of Tainan for experiment and developing the methods of reconstructing roof models by edge matching with the help of digital map data. We found out that digital map data can be helpful in two ways. One is that the maps can be used to narrow down the search area for finding building edges. The other is that they can be used to fill up gaps in a broken image edge or even to complete a missing edge. The results of this research also show that even with the help of map data, it is still not easy to reconstruct roof models completely. The main reasons are (1) the low­level edge extraction is often faulty, (2) the roof edges extracted are usually interfered by adjacent strong signals or are broken, (3) some roof edges could not be extracted at all. But in general, with the help of digital maps, the successful rate of reconstruction of roofs by simple edge matching can be indeed increased effectively.

The introduction of projective centre into the block adjustment with independent models can strengthen the geometry and provide much better convergency for the aerotriangulation. In this note, the accuracy of the projective centre determination with two different methods is studied. These two methods are the ΔZ method and the one with analytical relative orientation.