Implementation And Optimization Of GPU-Based Lensless Digital Holographic Convolution Reconstruction Algorithm

The lensless digital holographic imaging technology records the interference pattern of the object light and the reference light through the photoelectric sensor.Then the diffraction process of the light can be reprodeuced By means of the computer numerical simulation.As a result,a reconstructed image containing the amplitude information and phase information of the measured object is obtained.The technology is characterized by simple structure,low cost and large imaging field of view,and is widely used in many fields such as three-dimensional shape observation,microparticle analysis and detection,holographic anti-counterfeiting technology and biomedical engineering.However,in the process of holographic reconstruction,due to the high complexity of the digital holographic reconstruction algorithm and the increasing resolution of the holographic image,the amount of reconstructed data increases tremendously.The current technology is difficult to meet the real-time requirements in practical engineering applications.Even though some researchers take advantage of multi-core CPUs to achieve acceleration,the acceleration effect is not obvious due to the limitations of memory resource and power consumption.In view of the problem that the current holographic reconstruction speed is too to meet the needs of real-time applications,this paper proposes a scheme to improve the execution efficiency of digital holographic reconstruction algorithm based on general GPU technology,by combining the logic control capability of the CPU with the parallel computing power of the GPU,so that the digital holographic reconstruction algorithm is implemented in parallel by heterogeneous programming of CPU+GPU.More exactly,through the detailed explanation and derivation of the lensless digital holographic reconstruction algorithm,the execution flow of the whole reconstruction algorithm is designed,and the reconstruction algorithm is implemented on the CPU.It is verified that the execution efficiency of the algorithm on the CPU is relatively low.To offer a solution,the parallel of the serial code is analyzed and the parallel execution flow of the reconstruction algorithm is further designed.According to the process,the reconstruction algorithm is successfully implemented on the GPU,and the parallel code is further optimized from the aspects of data transmission and thread planning.The optimized parallel code is tested and analyzed on different device platforms.The experimental results show that when the same digital hologram is reconstructed,the execution speed of the entire reconstruction algorithm on the GPU is significantly faster than that on the CPU,and the acceleration effect reaches 35 times.Furthermore,the parallel program exhibits excellent scale growth.In other words,the speed advantage of execution on the GPU will increase significantly with the increase of digital hologram resolution.Therefore,it is suitable for time-sensitive applications such as biological cell observation and analysis.