Research And Implementation Of GPU Accelerated FD-OCT System Real-time Image Reconstruction

The imaging speed of the Fourier Domain Optical Coherence Tomography(FD-OCT)system depends on the camera acquisition speed and image reconstruction speed.The current linear array camera technology can fully realize the real-time collection of large-scale image data.However,the collected image data is particularly large and each line of data is independent of each other,and is reconstructed according to the same algorithm.Therefore,how to make parallel calculation of each line of image data and improve the speed of FD-OCT image reconstruction is the key engineering problem of real-time imaging.In order to solve this engineering problem,this paper uses a new RTX 2080Ti graphics card as the computing processor for image reconstruction of the FD-OCT system.By distributing each line of image data to different computing cores for simultaneous parallel calculations,the calculation is greatly improved Speed to achieve real-time image reconstruction of the system.Compared with traditional GPU graphics cards,RTX 2080Ti graphics cards have more computing cores(4352 core units)and larger memory capacity(11G),single-precision floating-point number computing power is improved by 76%compared to GTX1060 graphics cards,ensuring that image data can be distributed in parallel Perform calculations for more computing cores to achieve real-time image reconstruction.In the programming,use CUDA C language to write the program of image reconstruction.The transmission part of the interference signal data and intermediate variables collected by the FD-OCT system uses page-locked memory.The GPU device can directly copy the data on the host side by accessing the page-locked memory.The transmission rate of the interference signal data into the GPU is faster;The background signal processing part of the FD-OCT system uses shared memory to read and write background signal data.As the memory block in the thread block,the threads in the thread block can be directly accessed,which is an order of magnitude faster than accessing other memory.Therefore,storing intermediate variables that are repeatedly called during background signal processing in shared memory can reduce memory access latency,improve computational efficiency,and speed up the image reconstruction process.Through comparative analysis of experimental results and data,when performing reconstruction tasks on image data with a size of 1000 lines×1024 pixels/line,the calculation speed of the RTX 2080Ti graphics card is 22 times higher than the CPU processing platform speed,and the calculation speed of the GTX1060 graphics card Increased by 2 times,the calculation time of the entire frame is 10.5ms,which can ensure that the system camera(linear velocity is 92kHz/s)collects one frame of image while completing the calculation of one frame of image reconstruction,reaching 92 frames/second of real-time image reconstruction.Finally,the system is applied to the measurement of the thickness of the middle adhesive layer of the multilayer optical window to improve the speed and accuracy of the measurement of the multilayer optical window.