Fast Processing Techniques for Ground-Based Differential Interferometric Radar Based on Embedded GPU

For ground-based differential interferometric radar systems, aiming at the problems of low operation-speed performance, a rapid processing technology was studied based on embedded GPU. To achieve effective processing of high-speed data streams, an architecture design based on message queues is adopted in this paper, enabling parallel data transmission and processing. Message queue, as an efficient middleware technology, not only ensure reliable data transmission but also enable decoupling of data reception and processing through an asynchronous communication mechanism. This significantly enhances the system's throughput and response speed. To further optimize performance, condition variables are introduced to control the concurrency strategy in the message queue. When there is data to be processed in the message queue, the condition variable wakes up waiting threads to start processing messages; when there is no data to be processed, it suspends the threads. This mechanism ensures the maximization of system resource utilization. Parallel reduction techniques based on shared memory are adopted to achieve rapid selection of persistent scatterer. Shared memory reduces data access times, and the use of parallel reduction techniques is particularly effective when handling large-scale datasets. The results show that for 200 images with dimensions of 634 by 100 pixels, the method presented in this paper improves the processing speed by 16% compared to traditional methods, effectively enhancing the processing efficiency of ground-based differential interferometric radar systems.