Combining 2D synchrosqueezed wave packet transform with optimization for crystal image analysis

Jianfeng Lu, Benedikt Wirth, Haizhao Yang

TL;DR

This paper introduces a new variational optimization method to analyze atomic-resolution crystal images. By combining a synchrosqueezed wave packet transform (SST) with a physics-informed optimization framework, the method can more accurately identify crystal defects and estimate local elastic deformation, leading to more robust and physically consistent results.

What Problem Does This Paper Address?

Analyzing crystal images at the atomic scale is crucial for understanding material behavior, especially concerning defects like dislocations and grain boundaries. Traditional methods often rely on visual inspection, which is time-consuming and prone to error, especially with the increasing volume of data from advanced imaging and simulation techniques. While automated methods exist, accurately characterizing crystal deformation and precisely locating defects, particularly in the presence of noise or complex structures, remains a challenge. Existing methods often struggle with the physical constraints of crystal mechanics, such as the curl of the deformation gradient being zero in defect-free regions and related to Burgers vectors at defects.

Methods

The core of this work is a two-stage approach:

  1. Synchrosqueezed Wave Packet Transform (SST): The paper leverages a 2D synchrosqueezed wave packet transform (SST) to extract initial information from crystal images. The SST is particularly adept at analyzing wave-like structures and can estimate local wave vectors, which are then used to infer an initial guess for the inverse deformation gradient ($G_0$). This initial guess also helps in identifying potential defect regions.
  2. Variational Optimization: The initial $G_0$ is then refined using a variational optimization method. This method minimizes an energy functional that includes a fidelity term (measuring how well the estimated deformation matches the initial guess) and an elastic energy term. Crucially, this optimization is performed outside the identified defect regions, enforcing physical constraints like the curl of the deformation gradient being zero in these defect-free areas. The method also incorporates topological consistency by allowing for specific jumps in the deformation gradient across identified defect boundaries, consistent with Burgers vectors. The optimization is solved using a nonlinear projected conjugate gradient method.

Datasets

The paper evaluates its method on:

  • Synthetic crystal images: Generated using the Phase Field Crystals (PFC) model, including both noiseless and noisy examples.
  • Real experimental images:
    • TEM images of GaN with isolated defects and large-angle grain boundaries.
  • Photographs of bubble rafts with disordered and blurry grain boundaries.
  • TEM images of Al with twins and high-angle grain boundaries.

The paper does not introduce new benchmark datasets but uses these diverse examples to demonstrate the method's robustness and applicability.

Key Insights and Ideas

  • SST as a powerful initial guess: The SST provides a surprisingly good initial estimate of the deformation gradient, even in noisy images, which is crucial for the subsequent optimization.
  • Physics-informed optimization: By minimizing elastic energy outside defect regions and enforcing the curl-free constraint, the method produces results that are much more consistent with the underlying physics of crystalline materials.
  • Defect region identification: The SST-based "mass" indicator effectively identifies defect regions, allowing the variational method to focus its constraints on the defect-free areas.
  • Topological consistency: The introduction of jump sets to handle topological consistency across defect boundaries is a significant step towards a more complete description of crystal deformation.
  • Robustness to noise: The method demonstrates remarkable robustness to noise, recovering accurate estimates even from heavily corrupted images.

Key Results

  • The optimized deformation gradient ($G$) is significantly smoother and more physically consistent than the initial SST estimate ($G_0$).
  • The method successfully reduces unphysical curl away from defect regions, concentrating it within the identified defect areas, where it correctly relates to Burgers vectors.
  • Visualizations show improved crystal orientation, reduced shear (difference in principal stretches), and volume distortion closer to zero outside defect regions.
  • The method is computationally efficient, with runtimes for large images (1024x1024) on the order of tens of seconds, even with different downsampling rates.

Limitations and Future Work

  • Defect region optimization: The current method optimizes outside defect regions. The authors acknowledge that a more complete solution would involve developing variational models to also estimate the deformation gradient inside defect regions for more accurate defect localization.
  • Nested optimization: A potential future direction is a nested optimization model, where an inner loop optimizes outside defects and an outer loop improves the curl inside defect regions.
  • Computational cost: While efficient, the inversion of the discrete Laplacian operator, especially with non-empty jump sets, is the main computational bottleneck. Further optimizations in this area could be beneficial.

Summary updated

Abstract

We develop a variational optimization method for crystal analysis in atomic resolution images, which uses information from a 2D synchrosqueezed transform (SST) as input. The synchrosqueezed transform is applied to extract initial information from atomic crystal images: crystal defects, rotations and the gradient of elastic deformation. The deformation gradient estimate is then improved outside the identified defect region via a variational approach, to obtain more robust results agreeing better with the physical constraints. The variational model is optimized by a nonlinear projected conjugate gradient method. Both examples of images from computer simulations and imaging experiments are analyzed, with results demonstrating the effectiveness of the proposed method.
BibTeX
@article{lu2016combining,
  author = {Jianfeng Lu and Benedikt Wirth and Haizhao Yang},
  title = {{Combining 2D synchrosqueezed wave packet transform with optimization for crystal image analysis}},
  journal = {Journal of the Mechanics and Physics of Solids},
  year = {2016},
  doi = {10.1016/j.jmps.2016.01.002},
}

Loading related content...

Figure