Information Theory-Based Target Detection for High-Resolution SAR Image

In this letter, we propose a target detection approach in high-resolution synthetic aperture radar (SAR) images by using the information measurement of superpixels. This study aims to transform the basic cell of SAR images from pixel to patch through the superpixel algorithm. Moreover, by taking advantage of the rich statistical character of the patch, an information measurement, including self-information and entropy, is utilized to measure the statistical difference between patches. Self-information is utilized to measure the relative information value of patches, while entropy is used to describe the change degree of statistical characteristics between the patch and its surroundings. In this way, the proposed approach is more stable for SAR images with different intensities of speckle noise because damaging information measurement is difficult with speckle noise. Information on distributed targets in high-resolution SAR images can be measured via superpixel-based pixel clustering instead of by single pixels. Therefore, the proposed method can utilize more information to achieve target detection. The experimental data contain simulated SAR images with different intensities of speckled noise and real high-resolution SAR images. The performance of the proposed approach is validated by comparing the proposed approach with two classic constant-false-alarm-rate algorithms on the experimental data.