Paw-Net: Stacking ensemble deep learning for segmenting scanning electron microscopy images of fine-grained shale samples

Binqian Yin, Qinhong Hu, Yingying Zhu, Chen Zhao, Keren Zhou

Abstract: Highlights•The widely employed U-Net solution has been investigated to discover its shortcomings in shale rock images.•Both the focal and dice loss functions are combined to address the object imbalance problem in the geological field.•A novel stack ensemble learning pipeline is proposed to reduce the variance using a single neural network.•Our method can improve the average IOU metric by 9% compared to the state-of-the-art solution without additional data.