Ji Kefeng

Road extraction from high-resolution SAR imagery using Hough transform

This paper presents a technique for the extraction of roads in a high resolution synthetic aperture radar (SAR) image using Hough transform. Roads in a high resolution SAR image can be modeled as a homogeneous dark area bounded by two parallel boundaries. Dark areas, which represent the candidate positions for roads, are extracted from the image using a Gaussian probability iteration segmentation, and the roads are accurately detected by Hough transform. For this purpose, we designed an average Hough transform, which is more reasonable than general Hough transform for the extraction of lines. We search the peak values in Hough space and try to reduce its overall computational cost by introducing a global CFAR detector. In this process, to detect roads more accurately, post-processing, including noisy dark regions removal and false roads removal, is performed. We applied our method to MSTAR clutter images of Redstone that have a resolution of about 1 ft /spl times/ 1 ft. The experimental results show that our method can accurately detect roads.

A fast ship detection algorithm in SAR imagery for wide area ocean surveillance

Ship detection is a basic problem to be solved in SAR application of ocean surveillance. To deal with the rapid increasing SAR data, fast algorithm has become a hot topic in the research of ship detection. This paper proposes a fast ship detection algorithm in SAR imagery for wide area ocean surveillance. The algorithm adopts global two-parameter CFAR and adaptive CFAR based K distribution in the coarse and fine detection phase separately. Performances on detection accuracy and efficiency have been analyzed theoretically. Validation results on several space-born SAR images illustrate the fast method can preserve detection accuracy and improve the calculation efficiency.

Simulation of SAR image of ship

Simulation of SAR (Synthetic Aperture Radar) image of ship is very important for ship detection and classification of airborne and space-borne SAR platforms. Based on the research of high frequency RCS (Radar Cross Section) prediction, the methods of SAR image simulation of ship are investigated in this paper. First, a triangle-facet model of a ship is modeled using 3D Studio MAX, then the RCS of the ship is computed using the combination of PO (Physical Optics) and PTD (Physical Theory of Diffraction), finally, with SAR image formation processing, the predicted complex scattered field is turned into the SAR image of the ship. The validity of high frequency RCS prediction is verified through the RCS prediction result of a rectangular plate, and the validity of SAR image simulation of ship is verified through the simulated SAR images of a simple geometry structure made of four rectangular plates and a real ship model. Keywords-SAR; ship; simulation

A method to detect layover and shadow based on distributed spaceborne single-baseline InSAR

Layover and Shadow are inevitable phenomenena in InSAR, which seriously destroy the continuity of interferometric phase images and present difficulties in the follow-up phase unwrapping. Thus, its significant to detect layover and shadow. This paper presents an approach to detect layover and shadow using the auto-correlation matrix and amplitude of the two images. The method can make full use of the spatial information of neighboring pixels and effectively detect layover and shadow regions in the case of low registration accuracy. Experiment result on the simulated data verifies effectiveness of the algorithm.

The statistical analysis of SAR target based on the correlation algorithm

Utilizing MSTAR measured target data, we analyze statistical trend in the entire measurement training set every 1 deg in azimuth based on correlation algorithm. After segmentation and normalization, each test image was correlated with all the training images to generate correlation and classification statistics and correlation plots. The correlation plots varied approximately sinusoidally with aspect, and all the training sets show that a target was highly correlated at both the correct aspect angle and the correct angle rotated 180 deg, and this two correlation scores corresponded to the two local amplitudes of the correlation plot.