Takako Sakurai-Amano

Detection of flood damaged areas in the entire Chao Phraya River Basin from JERS-1/SAR images with a help of spatial information

The water surfaces detected by a conventional intensity based method from JERS-1 SAR images do not always coincide with the flood-damaged areas especially existing paddy fields. The authors propose morphological methods as spatial information. The result roughly agree with the reported flood-inundated areas. We concluded that this kind of methods has an advantage of the disaster monitoring using SAR satellite.

Automatic extraction of rivers in tropical rain forests from JERS-1 SAR images using spectral and spatial information

A fully automated method of extracting narrow rivers in tropical rain forests from JERS-1 SAR images was implemented using both spectral and spatial information. The system was divided into extraction of narrow rivers that appear as a series of line-like blobs and extraction of wide rivers that appear as dark continuous line-like features. The method was assessed by comparing the results from optical JERS-1 VNIR data and by applying 21 scenes in other areas in Amazon basin and Congo basin.

Automated GCP detection for SAR imagery: road intersections

Road intersections are very good ground control points in a SAR image because they can be obtained from intersecting road segments even when the intersections themselves are not observed. In a previous study, a Hough transform-based approach was developed for detecting intersections from a single-look ERS-1 SAR image. A similar Hough transform-based approach has been applied to detect road intersections in a 6-look ERS-1 SAR image. In the case of multi-look images, however, there are no pure road pixels because roads are thinner than the resolution size and the pixels that include road information are greatly affected by surrounding objects. More spatial information and even approximate road map information are required to improve the capability to detect road intersections. It is found that broad intersections ca be automatically determined from a multi- look image and the accuracy of the location of road intersections may be estimated from the heights of sharp peaks in Hough space.

Terrain feature recognition for synthetic aperture radar (SAR) imagery employing spatial attributes of targets

This paper describes an integrated approach to terrain feature detection and several methods to estimate spatial information from SAR (synthetic aperture radar) imagery. Spatial information of image features as well as spatial association are key elements in terrain feature detection. After applying a small feature preserving despeckling operation, spatial information such as edginess, texture (smoothness), region-likeliness and line-likeness of objects, target sizes, and target shapes were estimated. Then a trapezoid shape fuzzy membership function was assigned to each spatial feature attribute. Fuzzy classification logic was employed to detect terrain features. Terrain features such as urban areas, mountain ridges, lakes and other water bodies as well as vegetated areas were successfully identified from a sub-image of a JERS-1 SAR image. In the course of shape analysis, a quantitative method was developed to classify spatial patterns by expanding a spatial pattern through the use of a series of pattern primitives.

Crustal movement imaging by D-InSAR, reflector array, dense GPS, leveling and gravity changes in the plate convergent zone, at the Tokai region, central Japan

In order to obtain more precise and quantitative crustal deformations, we combine the results of spaceborne differential interferometric synthetic aperture radar (D-InSAR), with the dense distribution of continuous Global Positioning System (GPS) observations, precise gravity measurements, and leveling surveys in the Tokai region, where dense and continuous geophysical and geochemical monitoring has been carried out due to the expected huge interplate earthquake. Seasonal uplift-subsidence motions of 2 to 6 cm are obtained from winter to the beginning of summer for both 1995 and 1997, that might be concordant with the leveling survey, although the atmospheric effects should play an important role for these conclusions. At the Sagara corner reflector, we have experimentally sifted the surface of the reflector and obtained fairly reasonable results, although the discrepancy of about 1.5 cm (along the line of sight) between theoretical estimate could be attributed to the unexpected contamination of noise from the surroundings and/or analytical processes.

Detection of singular corner reflectors in residential and mountainous areas from SAR images

In a mountainous and highly populated country like Japan, corner reflectors frequently must be placed near confounding objects. A procedure to detect singular corner reflectors in such situations is described. It is found that the initial reduction of speckle by the Enhanced SFP filter greatly facilitated the detection of these corner reflectors from both JERS-1 SAR images and ERS-2 images.

Comparison of land cover indices of AVHRR data

The authors have compared NDVI and the feature mixing model as a practical semi-quantitative measure of daily vegetation. In the comparison, they used AVHRR data for a warm, humid and relatively flat eastern China. While NDVI only gives a general trend, the results obtained using the feature mixing model directly gives a more plausible picture of the vegetation of the corresponding terrain.

Detection of seasonal water level changes in Amazon forests from JERS-1 SAR images using isolated strong scatterers

A new method of detecting and visualizing narrow open-water rivers from JERS-1 SAR images was applied to a series of twenty continuous scenes of Amazon forests observed in May, August, and October, 1996 and February, 1997. The tributaries observed during high water seasons were longer and wider than those observed during low water level seasons. Comparing the 30 year-average of monthly precipitation and the change in the river width, the estimate water retention period of the part of the Amazon forests is about 1/spl sim/2 months.

Seasonal water level changes in tropical forests in New Guinea detected from JERS-1 SAR images

A method for detecting and visualizing narrow open-water rivers using very strong scatterers was applied to a series of JERS-1 SAR images of tropical forests in New Guinea. These images were observed in November and December, 1995 and February and March, 1996. The results indicate that the narrow river water levels on the Pacific side of the major mountain range stayed relatively constant during this period, while the water levels on the Australian side were low in November, highest during February and March and slightly lower in December.

Scene feature matching analysis of JERS-1 SAR images

A method to examine the shifts in individual surface features in multi-temporal SAR images has been proposed. First, line-like and boundary-like features were extracted from SAR imagery using the pixel swapping method. The extracted feature images were then tied to observe an overall shift in individual features. The authors applied this method to two JERS-1 SAR images of the same path-row observed on different dates. They observed that the shift in flat areas was simple, but that in mountainous regions, there was a clear difference in the direction of the shift in ridges and valleys caused by the slightly different orbital characteristics of the satellite between the two images. In addition, care in choosing appropriate tie points was found to be important. The method used in this investigation is expected to be useful in planning an appropriate strategy for the coregistration of images, such as in SAR images used for change detection in environmental monitoring, and InSAR.