基于残差网络的高温合金微观组织图像分割方法

材料微观组织图像分析是材料研究的重要环节,其分析方法的精准性和快速性对新材料的设计、研制和现有材料的优化、寿命评价都非常重要。因此,如何建立更快速更精准的微观组织分割方法成为微观组织图像分析和性能评价的关键。针对传统的微观组织图像分割技术对于高温合金材料分析精度不高等问题,通过对卷积神经网络结构进行优化,提出了一种基于Res_Unet网络的微观组织图像分割方法。实验验证结果表明,本文的方法不仅解决了深度学习在材料组织图像小样本数据上的实现问题,还显著提高了材料微观组织图像的分割精度。

Study on the Microstructure Images Segmentation Method of Superalloy Based on Res-Unet

Analyzing material microstructure images is an important part of researching material performance. The accuracy and celerity of analysis methods have an important impact on the design and development of new materials, the optimization of current materials and the evaluation of their life. Therefore, how to establish a faster and more accurate microstructure segmenting method becomes the key for analyzing microstructure images and evaluating performance. Aiming at the problem that the traditional microstructure image segmenting technology is not accurate for analyzing Ni-based superalloy, this paper optimizes convolutional neural network and proposes a microstructure image segmenting method based on Res_Unet network. It also integrates network improvement and data enhancement in the meantime. Through the experiment to Ni-based superalloy microstructure image, the result shows that this method solves how to actualize deep learning on the small sample data of material structure images and significantly improves the segmenting accuracy of material microstructure images.