Geometric transformation-based data augmentation on defect classification of segmented images of semiconductor materials using a ResNet50 convolutional neural network

作者：

Highlights：

• This paper analyzes the effect of data augmentation on the performance of a CNN model.

• Different novel and traditional data augmentation techniques are studied.

• Detailed explanation of the use of techniques based on geometric transformations.

• Data augmentation applied to imbalanced datasets clearly increases CNN performance.

摘要

•This paper analyzes the effect of data augmentation on the performance of a CNN model.•Different novel and traditional data augmentation techniques are studied.•Detailed explanation of the use of techniques based on geometric transformations.•Data augmentation applied to imbalanced datasets clearly increases CNN performance.

论文关键词：Data augmentation,Convolutional neural networks,Semiconductor device manufacturing,Inspection systems

论文评审过程：Received 7 January 2022, Revised 23 March 2022, Accepted 31 May 2022, Available online 9 June 2022, Version of Record 15 June 2022.

论文官网地址：https://doi.org/10.1016/j.eswa.2022.117731