基于核聚类的砂岩图像孔隙分割方法

砂岩孔隙识别是研究孔隙结构的一个重要步骤, 采用通用的图像分割算法不易得到理想的图像孔隙分割效果, 为此提出了一种使用EfficientNetV2-S模型和核K-Means聚类技术对孔隙进行分割的方法。首先, 获得砂岩图像的超像素集合, 使用超像素方法预分割输入的致密砂岩图像, 构建带标签的孔隙与非孔隙图像库; 然后, 应用EfficientNetV2-S模型提取砂岩图像的孔隙和非孔隙的语义特征, 并结合迁移学习的方法, 使用有限的砂岩图像的孔隙和非孔隙样本进行EfficientNetV2-S模型参数学习; 最后, 设计了一种基于K-Means聚类的区域合并方法——NTK-KCoP方法, 根据超像素的语义特征、灰度特征和边缘特征构建目标函数, 再由聚类结果合并超像素得到完整的孔隙区域。砂岩CT图像的实验结果验证了所提出的孔隙分割方法的适用性和有效性。

Sandstone pore identification is an important step in studying pore structure.It is difficult to achieve ideal results using general image segmentation algorithms.This paper proposes a method for pore segmentation using EfficientNetV2-S and kernel K-Means clustering.First, a superpixel collection of sandstone images is obtained, and the input tight sandstone images are pre-segmented using a superpixel method to construct a library of labeled pore and non-pore images.Then, the EfficientNetV2-S model is applied to extract the semantic features of pores and non-pores in sandstone images.The semantic features are combined with a transfer learning method to learn EfficientNetV2-S model parameters using a limited number of pore and non-pore samples in sandstone images.Finally, a region merging method based on K-Means clustering is designed to construct an objective function by combining the semantic features, grayscale features and edge features of superpixels, which are then merged according to clustering results to obtain a complete pore image.Experiments on sandstone CT images verify the applicability and effectiveness of the pore segmentation method proposed in this paper.