Research On Compton Backscatter Imaging Based On Energy Spectrum

Radiation imaging technology has been widely used in non-destructive testing, clinicaldiagnosis, security check and many other fields. Compton backscatter (CBS) imaging technologyis an important research field in radiation imaging, in which backscatter radiation information isused. The main features of this technology include that ray source and detectors are on the sameside of the object, and the geometric structure is flexible. Thus CBS imaging technology allowsone side accessible objects and massive objects to be imaged.The traditional imaging way of CBS relies on the use of well collimated detectors to definethe path of scattering, which has the disadvantages of low scanning speed and low ray utilizationrate. CBS imaging way based on energy spectrum has fast scanning speed and high rayutilization, which is the focus of research. The imaging problem of CBS imaging way based onenergy spectrum is ill-posed, so image quality is affected by projection data and reconstructionalgorithm. Researches on system configuration and reconstruction algorithm are significant toimprove image quality.Researches on CBS imaging system configuration, reconstruction algorithm and the GPUacceleration of algorithm are performed. The main achievements of this paper are as follows:1. An optimization strategy of CBS system configuration is proposed. Based on MonteCarlo Method, this paper constructs the imaging system model and then the influences of systemconfiguration on data acquisition and image reconstruction are researched. The results show thatwhen the vertical position of detector array and ray source are near the top of object and thebottom of object, respectively, and the horizontal position of detector array and ray source arenear the object, the imaging system exhibits excellent capability in data acquisition andreconstruction image quality.2. A Compton backscatter reconstruction algorithm based on total variation minimization(CBS-TVM) is proposed. Utilizing the piecewise constant characteristic of most images,employing the total variation minimization as objective function, the reconstruction problem isdescribed as an optimization problem with nonlinear data-consistency constraint. An algorithmto solve this optimization problem is proposed, in which alternating optimization tactics areutilized after separating variables. The results show that compared with unweighted least-squaresmethod and EM-ML (Expectation Maximization-Maximum Likelihood) method, the proposedalgorithm could suppressive noise and preserve the edge simultaneity. Reconstruction error isreduced and image quality is improved. Especially when the number of measurements is reduced,the proposed algorithm shows more advantages.3. Some strategies and optimization parallel program are designed for CBS-TVM algorithm.Iterative algorithm is a time consuming process. This paper designs some strategies andimplements some optimization to parallel program according to characteristics of CBS-TVMalgorithm. The strategies include subsets specification, voxel-based projection and so on. Theoptimization of parallel program contains maximizing the occupancy of GPU, matrix-vector multiplication, memory access and so on. The results show that under these optimizationstrategies, the GPU parallel program could run with high speedup ratio of138.7and ensurequantity of images at the same time. This could greatly cut the amount of reconstruction time.