Research On Optimization Calculation Of Image Reconstruction

Image vision has always played an important role in the development of human society.However,with the continuous development of modern technology,the image information to be processed has become more complex and diverse.Therefore,people from all walks of life have higher performance expectations for image processing technology.Image processing technology usually includes image reconstruction,image recognition,image segmentation,target detection and other fields,and this article focuses on exploring how to use optimized calculation ideas to solve the three-dimensional reconstruction and two-dimensional image reconstruction problems in image reconstruction.Embedded computer vision technology is one of the important application fields of optical image 3D reconstruction.Its principle is similar to the basic mechanism of the human eye.The depth information of the corresponding scene can be obtained through binocular vision.It can be used in robot navigation,cultural relics protection,aerial photography,space Exploration and other fields have important potential applications.On the other hand,deep learning technology can use prior knowledge to train neural network model,and then extract valuable abstract information from complex data.Therefore,it shows the development potential of surpassing traditional algorithms in many application fields.Especially in situations where sample data such as medical images and 3D point clouds are expensive and difficult to obtain.This article mainly involves the research of 3D reconstruction of UAV images based on the collaborative optimization of software and hardware and the reconstruction technology of Compressed Sensing Magnetic Resonance Imaging(CSMRI)based on edge optimization.There are relatively few existing researches on FPGA-based 3D reconstruction of optical images based on phase correlation algorithms,and most of them are based on spatial domain features to achieve image stereo registration.The main problems and challenges can be summarized in two points: one is traditional FPGA technology suffers from the limitation of logic resources and it is difficult to implement many advanced algorithms that are highly iterative in nature;second,traditional FPGA-based 3D reconstruction technology is usually not suitable for complex and low-texture areas with high narrow baselines and easily involving river valleys,plains,and mountains.UAV aerial images;third,traditional 3D reconstruction technology is difficult to effectively realize the collaborative optimization between software algorithms and hardware logic.On the other hand,in the field of Compressed Sensing Magnetic Resonance Imaging(CSMRI)reconstruction research,most of the existing methods are difficult to effectively balance the abstract global high-level features and edge features,which can easily cause a large amount of residual aliasing artifacts and obvious excessive Smooth reconstruction of details and other issues.Based on the above problems,combined with the idea of optimization calculation,this paper explores 3D reconstruction technology based on hardware and software co-optimization and compressed sensing magnetic resonance imaging(CSMRI)reconstruction technology based on edge optimization.Compared with the existing research results,the research work of this article can be divided into the following two aspects:(1)Based on the FPGA platform with limited resources,this paper proposes a multi-scale depth map fusion algorithm to overcome the traditional FPGA method's susceptibility to lighting changes,occlusion,shadows and small rotations,and then achieves from low-texture areas,dynamic texture areas,and large parallax ranges.Highly reliable parallax information is extracted from real mountain UAV images with interference factors.In addition,this paper also combines high-parallel instruction optimization strategies and high-performance software and hardware collaborative optimization methods to propose a high-throughput hardware optimization architecture with hierarchical iteration and parallel at the same level,which greatly improves the FPGA-based 3D reconstruction method of UAV aerial images.Robustness,accuracy and comprehensive energy consumption,etc.,finally realize the efficient operation of the multi-scale depth map fusion algorithm architecture on the FPGA platform with limited resources.(2)This paper also proposes a novel edge-optimized dual discriminator generative adversarial network architecture(EDDGAN)for high-performance CSMRI reconstruction by balancing global advanced features and edge features.First,the edge operator is combined to fuse multi-scale edge information to extract effective edge features,and then the abstract global high-level features and edge features are combined,and a three-person game is introduced to control the detailed illusion and stabilize the network training.A large number of studies have shown that the edge-optimized EDDGAN method can focus on edge detail restoration and removal of aliasing artifacts.Compared with the most advanced technology,the reconstructed image generated by this method has a richer edge detail structure.To the best of our knowledge,this article is the first to use the edge optimization idea to apply the dual discriminator generative adversarial network to the field of CSMRI reconstruction.