The Application Of Information Fusion In Spectral Analysis Reconstruction Method

Optical CT has a wide range of applications in tomography of flow field such as temperature and concentration. However, partial covered field or other environmental restrictions makes the application of optical CT in the case of a few projections increasing, so as to put forward higher requirements for image reconstruction. The quality of the reconstructed image is strongly dependent on the performance of reconstruction algorithm. Series expansion method is especially suitable for incomplete projection data of image reconstruction, and the frequency spectrum approach reconstruction technique(FSART) is a typical representative. But FSART's reconstructed accuracy and stability are short of requirements when reconstruct the complex dynamic field. So it is necessary to improve the algorithm in order to achieve real-time measurement.In this paper, on the basis of in-depth understanding of FSART theory, try to add the idea of information fusion to the original algorithm. Consider the distribution of prior knowledge of the field to join the reconstruction process. First, select the double Gauss function model to simulate double asymmetric candle flame temperature field. Then, Fourier series analysis is carried out on the double Gaussian function. According to the function' general distribution, select different approximate series on the x axis and y axis, get the corresponding Fourier series expansion and reconstructed image. Finally, compare the reconstructed image with the original algorithm. In addition, consider joining point measurements in the reconstruction process of the more complex simulation models such as the three Gaussian function. First, take three peak points of the measured values, replace the reconstruction results in a corresponding pixel of the reconstruction value, calculates reconstruction error. And then take one measured value which close to the highest peak on the basis of the obtained three peaks, modify the reconstruction value of the corresponding point again. By parity of reasoning, the numbers of point measurements whose average error is less than 1% and maximum error is less than 5% meet the requirements of reconstruction.From the experimental results, the addition of a prior knowledge makes computational efficiency improved effectively. With the reduction in the number of unknown values, the amount of computation correspondingly reduces. Meanwhile, one direction has a higher amount of information, the reconstruction results is more accurate. In addition, when the number of point measurements gradually increase, reconstruction error decrease, and the quality of the reconstructed image is higher and higher. The comprehensive utilization of these two points makes information fusion in the application of frequency spectrum approach reconstruction technique get perfect embodiment, and open up an effective way for solving the problem of incomplete data under a few projections.