Nuclear Techniques ›› 2019, Vol. 42 ›› Issue (5): 50102-050102.doi: 10.11889/j.0253-3219.2019.hjs.42.050102

• SYNCHROTRON RADIATION TECHNOLOGY AND APPLICATIONS • Previous Articles     Next Articles

Improving the reconstructed image quality in ptychography with momentum acceleration

Xulei TAO1,2,3,Zijian XU1,2(),Chunpeng WANG1,2,Renzhong TAI1,2()   

  1. 1. Shanghai Institute of Applied Physics, Chinese Academy of Sciences, Shanghai 201800, China
    2. Shanghai Advanced Research Institute, Chinese Academy of Sciences, Shanghai 201210, China
    3. University of Chinese Academy of Sciences, Beijing 100049, China
  • Received:2019-01-15 Revised:2019-03-07 Online:2019-05-10 Published:2019-05-16
  • Contact: Zijian XU,Renzhong TAI E-mail:xuzijian@sinap.ac.cn TAI Renzhong;xuzijian@sinap.ac.cn;xuzijian@sinap.ac.cn TAI Renzhong;tairenzhong@sinap.ac.cn
  • About author:TAO Xulei, male, born in 1992, graduated from Wuhan University in 2014, doctoral student, focusing on X-ray microscopy and coherent diffraction imaging|TAO Xulei, male, born in 1992, graduated from Wuhan University in 2014, doctoral student, focusing on X-ray microscopy and coherent diffraction imaging|XU Zijian, E-mail: xuzijian@sinap.ac.cn TAI Renzhong, E-mail: tairenzhong@sinap.ac.cn|XU Zijian, E-mail: xuzijian@sinap.ac.cn TAI Renzhong, E-mail: tairenzhong@sinap.ac.cn
  • Supported by:
    Supported by National Natural Science Foundation of China (No.11575283, No.11875316, No.U1832146), National Key Development Plan(No.2017YFA0403400)

Abstract: Background

The image retrieval algorithm of ptychography is a kind of conventional gradient descent algorithm. This algorithm begins with a set of guessing functions, and the signals collected in the experiment are continuously input into the iteration process to achieve final convergence. However the reconstructed process often falls into stagnation because of the low gradient around the local minima and the extent of the gradient descent is too small.

Purpose

This study aims to solve the convergence stagnation problem of conventional gradient descent algorithm by using momentum ptychographical iterative engine (mPIE).

Methods

The concept of “momentum” was introduced into the ptychography algorithm to update conventional algorithm, called the momentum ptychographical iterative engine (mPIE). Adding momentum to an algorithm means that the iterative process has the characteristic of inertance. The extent of gradient descent relies on the previous steps and the gradient of current step. With the momentum, convergence can escape stagnation, and the reconstructed process can continue even around the local minima with a gradient is very small. The same set of coherent diffraction imaging experimental data was used to test the reconstructed image quality of mPIE algorithm and compare with that of extending ptychographical iterative engine (ePIE) algorithm.

Results

The experimental results showed that mPIE algorithm outperformed ePIE algorithm with smaller convergence value of objective function, and better final reconstructed image quality.

Conclusion

The result indicates that a better reconstructed result can be achieved with mPIE algorithm compared with the traditional ePIE algorithm.

Key words: X-ray ptychography, Mixed state, X-ray microscopy, Iterative phase retrieval algorithm

CLC Number: 

  • TL99