Abstract
The low-band synthetic aperture radar (SAR) is a system which can present foliage penetrating (FP) targets and exposed strong-scattering targets while the high-band SAR system can present exposed targets and the texture information except for FP targets. Therefore, the difference feature of co-registered low-band and high-band SAR images can theoretically reveal FP targets with some disturbances. However, the co-registered dual-band SAR images cannot do the difference operator directly for the distinct statistical property of the SAR image pair. The traditional method is to transfer the background texture from the high-band SAR image to the low-band SAR image by the cycle-consistent generative adversarial network (CGAN) method. However, CGAN would generate the discontinuity and mistake in transferring the large-scale texture information and the complex network would heavily increase the time complexity in the application. To overcome this issue, we first innovatively introduce a fast and accurate texture transferring method based on the dual-band SAR image fusion (DBIF), then we further combine the DBIF based texture transferring method with the difference operator and the threshold segmentation, and finally we get a novel DBIF based FP target detection (DBIFFPTD) method. To verify the feasibility of the DBIF based texture transfer method, we discuss the reflection form of different landscape objects at dual-band waves and estimate the theoretical texture transferring image. Experiments on the open dual-band AIR-MD-SAR dataset and the independently measured dual-band SAR dataset show that the novel DBIF performs better than CGAN in transferring texture information of dual-band SAR images. Besides, dual-band linear SAR (LSAR) and circular SAR (CSAR) foliage penetrating experiments are both conducted to show that the proposed DBIFFPTD has a wide applicability in the flight track and is superior to the traditional low-band double-parameter constant false alarm rate (DCFAR) detector based FPTD (DCFARFPTD) method and dual-band CGAN based FPTD (CGANFPTD) method.
Funding Information
  • Natural Science Fund for Distinguished Yong Scholars of Hunan Province (2022JJ10062)
  • National Natural Science Foundation of China (62271492, 62101562, 62101566)

This publication has 25 references indexed in Scilit: