Research On Multi-energy DR Images Fusion Technology Based On Sparse Representation And Dictionary Learning

As one of the techniques for inspecting components,digital radiography(DR)technology can better evaluate the internal structure of components,and it is now widely utilized in aerospace,defense,industrial flaw detection,and other industries,where it plays a vital role.However,single-energy DR images cannot image non-equal-thickness components with simultaneous overall exposure,resulting in a serious lack of structural information of the components to be inspected.So multi-energy DR image fusion is especially important to solve the degradation problem.In this paper,we use sparse representation and dictionary learning theory to fuse the sparse coefficients of multi-energy DR images.To achieve complete and clear imaging of complex non-equal-thickness components without changing the hardware equipment of the X-ray imaging inspection system.The main research work of this paper is as follows:(1)It is proposed a method for extracting effective information picture blocks from DR images and training dictionaries with this information.To begin,we analyze the connection between local binary pattern information and DR image block information amount,as well as the relationship between fuzzy entropy and DR image block information amount.We then calculate the local binary pattern information and fuzzy entropy value of a DR image block.The connection between the local binary pattern and the effective information image block,as well as the connection between the fuzzy entropy and the effective information image block,are made by setting the threshold value.As a result,a dictionary training set with a lot of information is acquired.(2)Design fusion strategy.Based on the traditional image fusion strategy,we propose a DR image fusion method based on the combination of three feature parameters to fuse the sparse coefficient matrix.The max₁l norm of the sparse coefficient,the maximum value of the sparse coefficient,and the average value of the sparse coefficient are the three feature parameters.Due to the fusion strategy of single features,the fused images can lack critical information.The proposed fusion method can avoid this problem.Then,to address the issues that the fusion strategy of multiple features cannot guarantee that the fused image contains all of the information of DR sequence images.And the importance or weight coefficient of each feature is difficult to determine adaptively.So,under the premise of ensuring complete image information and better contrast,a fuzzy entropy-based multi-energy DR image fusion method is proposed.Artifacts and spatial discontinuities in the fused images can be avoided with this procedure.Finally,to obtain the fused image,the fused sparse coefficient vector is inverse sparse coded.The effects of dictionary size,sliding step,and parameter?on experimental outcomes are demonstrated empirically in this paper.Furthermore,it has been demonstrated that identifying useful information image blocks can increase both the subjective and objective assessment index of fusion results while also reducing simulation time.The suggested multi-energy DR image fusion approach is compared and analyzed with various traditional image fusion methods in order to prove the usefulness of the algorithm in this study.The results of the experiments reveal that this research is capable of solving the internal structure analysis of DR systems for irregular structure shapes,broad thickness variation ranges,and complicated internal structure components.It is critical in computer vision,medical image processing,industrial flaw detection,military and other fields.