Real-time Interferometric Processing Techniques and Acceleration Methods for UAV SAR

Abstract

Micro SAR outperforms other Technologies in radar system study since it is small, light and concealable. Airborne dual-antenna micro synthetic aperture radar (SAR) system collects ground digital elevation information (DEM) from two different SAR images. Across-track interferometer mode based on UAV SAR reduces the mounted platform’s requirement, thus making it more convenient to perform flight experiments due to its flexible deployment. This type of advanced technology is highly demanding, and it is a hit trend of study field among most researchers. Real-time interferometric measurement is required to deal with a massive amount of complex data and process it within a short period. This feature also plays a significant role in multiple fields such as battlefield surveillance, unmanned fighter navigation, and blind landing of helicopters. However, the traditional way of serial processing can no longer satisfy the real-time processing requirements due to its massive amount of algorithm steps and complex computations. Therefore, this paper proposes a real-time InSAR processing algorithm and applied multi-core and multi-thread acceleration to boost the three most time-consuming modules: Flat-earth Phase Removal Module, Phase Filtering Module, and Phase to Height Conversion and Geocoding Module. In addition, using data block and parallel processing methods while doing calculations can significantly increase signal processing efficiency. Finally, this paper validates the effectiveness of the algorithm by testing it on the whole system. The results show that using an 8-thread computer, the parallel processing module’s time is only 15% compared to the original module. Moreover, the overall system operation speed is accelerated two times more than the previous traditional technique, the relative error of elevation is 0.3911 m. In conclusion, this paper validates that parallel processing is an effective way for real-time processing.