SAR Image Change Detection Based on Geometric Mean Operator and Extreme Learning Machine

Synthetic Aperture Radar (SAR) image is independent of atmospheric and sunlight conditions and can be acquired under all weather and all day. As a part of SAR image application, change detection has high practical value, such as urban sprawl detection, hazard assessment of earthquake areas. In this paper, we put forward a new difference image (DI) generated by geometric mean operator which combined with log-ratio, neighborhood ratio and normal difference operator, and retained the advantages of each operator. Then, a change detection method of SAR image based on extreme learning machine (ELM) is used to classify the pixels in two original SAR images to form the final change image. Experimental results on two real SAR image datasets show that the proposed difference image generation method is robust to speckle noise, and the change detection method based on the new DI operator and ELM can effectively detect variation information in multi-phase SAR images.