Research On Synthetic Aperture Focusing Ultrasonic Imaging Algorithm Based On GPU

Ultrasonic imaging and reconstruction tools are commonly used to detect, identify and measure defects in different mechanical parts, has been widely used in medical and industrial non-destructive testing and other fields. The conventional ultrasonic imaging technology has difficulties in finding and evaluating the flaw accurately because of its low resolving capability. In order to get better imaging quality, the method of synthetic aperture ultrasound imaging is introduced. Ultrasonic image generation based on Synthetic Aperture Focusing Techniques (SAFT) can be divided into two stages:(1) the excitation and acquisition stage, where the signals received by each element or group of elements are stored; (2) the beamforming stage, where the signals are combined together to obtain the image pixels.Although synthetic aperture ultrasound imaging improves the quality of the image,it also increases the amount of computation. Therefore, a general hardware platform is difficult to achieve real-time imaging. Aimed at the problem of large amount of computation and poor real-time performance of the synthetic aperture ultrasound imaging reconstruction algorithm. In this text, a parallel processing method based on GPU architecture is proposed to improve the imaging rate of CUDA. The parallelism of the synthetic aperture ultrasound imaging algorithm is analyzed in detail, including the parallel of the analytical signal, the parallel and the post processing of the image. The use of Graphics Processing Units (GPUs) can significantly reduce the computing time of this last stage, which usually includes different functions such as echo signal preprocessing, beam control, delay stack, envelope detection, dynamic focusing or filtering noise reduction.This work mainly studies the parallel application of ultrasonic imaging in the beamforming stage and the application of CUDA based GPU technology. In the same experimental environment, Using CPU+GPU heterogeneous model parallel speedup, with respect to the use of a single thread to reconstruct a higher than about 200 of the acceleration ratio, by reconstructing images with simulated data with the defect data in standard B test block.