Deep Learning and Shape Aware Based Image Segmentation for Polycrystalline Micrographic image

Xiaojuan Ban^*, Haiyou Huang*, Boyuan Ma, Hao Wang, Weihua Xue

University of Science and Technology Beijing, Beijing, 100083, China

ABSTRACT: The rapid and accurate establishment of the intrinsic relationship of material composition, process, microstructure and properties is the basis of material performance improvement, quality control and material design. Machine learning technology recently has made significant progress in the analysis and mining of structural data. At the same time, for the quantitative analysis of material organization structure presented by unstructured image data, the deep learning technology represented by convolutional neural network has been widely used. Taking Fig. 1 as an example, the algorithm needs to extract and separate each grain accurately to provide support for subsequent analysis of geometric and topological information of grains. However, the grains have similar appearance and complex shape. Due to its similar appearance of each grain, the traditional image segmentation method based on color or threshold is difficult to separate the grain by grain. To solve the above problems, the contributions of this work are: 1) we present a shape aware based deep convolution network, which can lead the convolutional neural network to pay more attention to the shape information in the training process, so as to obtain more accurate image segmentation results; 2) We summarize and discuss the evaluation metrics for image segmentation, and analyze different segmentation algorithms and typical noises through experiments. Finally, the advantages and applicability of various evaluation methods are compared and discussed; 3) We develop an image segmentation software based on deep learning, and propose the design principles of the software for material microscopic image analysis.

Figure 1. Image segmentation task of polycrystalline grain structure.

Keywords: Material Microscopic Image Analysis; Polycrystalline materials; Deep Learning; Image Segmentation; Objective Function;