基于深度学习的岩石孔隙分割方法  

Rock Pore Segmentation Method Based on Deep Learning

作　　者：陈国军[1] 姜朕 尹冲 王乐康 CHEN Guojun;JIANG Zhen;YIN Chong;WANG Lekang(College of Computer Science and Technology,China University of Petroleum(East China),Qingdao 266580)

机构地区：[1]中国石油大学(华东)计算机科学与技术学院,青岛266580

出　　处：《计算机与数字工程》2023年第5期1157-1162,共6页Computer & Digital Engineering

摘　　要：现有的传统孔隙分割方法不能准确提取岩心图像中细小狭长的孔隙,且容易受到岩石图像噪声的干扰,针对上述问题,提出了一种深度学习网络模型ARC-Unet(Attention and Recurrent-Convolution Unet),用于更加精确的分割岩石孔隙。采用Unet作为基础网络,并在网络上加入注意力机制,用于解决在分割时小面积的孔隙容易被漏分割的情况。将循环卷积模块代替原来的卷积模块,可以拟合更多的岩石特征,提高孔隙分割的准确度。通过在采集并制作的岩石数据集上进行训练并在通过在测试集的分割结果上进行模型评估,改进模型在测试集上的F1达到了88.15%,有着较好的岩石孔隙分割结果。The existing traditional pore segmentation methods cannot accurately extract the small and long pores in the core image,and are easily disturbed by the noise of the rock image.In response to the above problems,a deep learning network model ARC-Unet(Attention and Recurrent-Convolution Unet)is proposed for more precise segmentation of rock pores.Unet is used as the basic network,and an attention mechanism is added to the network to solve the situation that small areas of pores are easy to be leaked during segmentation.Replacing the original convolution module with the cyclic convolution module can fit more rock features and improve the accuracy of pore segmentation.Through training on the collected and produced rock dataset and model evaluation on the segmentation results of the test set,the F1 of the improved model on the test set reaches 88.15%,which has a good rock pore segmentation result.

关 键 词：深度学习 岩石孔隙分割 注意力机制 循环卷积 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]