Auto-interpretation for Bridges over Water in High-resolution Space-borne SAR Imagery

An automatic method for detecting and interpreting bridges over water in high-resolution space-borne synthetic aperture radar imagery is proposed. Firstly, the textual features for image classification are computed, including response for Gabor filter, tree-structure wavelet coefficient and statistics of gray level co-occurrence matrix. Then the SAR imagery is classified to low-reflection area, vegetation covered area and built-up area using support vector machine classifier. By analyzing targets space distribution, shape and gray characteristic in low-reflection area, the Regions Of Interested (ROI) are detected. For each ROI, five key parameters of bridge are estimated based on imaging model of radar, including direction, length over water, width, elevation over water, thickness of body and the real position for orthographic projection. Experiment with TerraSAR-X image indicates that the method is effective.