中国石油大学学报（自然科学版）2025，Vol.49Issue(3)：97-107,11.DOI:10.3969/j.issn.1673-5005.2025.03.010
基于DC-HED网络和骨架提取的岩心图像边缘检测
Edge detection of petrographic thin section images with DC-HED network and skeleton extraction
摘要
Abstract
The holistically-nested edge detection(HED)network is a widely used deep learning model for image edge detec-tion.However,it suffers from issues such as missing edges,redundancy,and blurring in the detection results.To address these shortcomings,this paper proposes DC-HED,a novel deep network model that integrates dilated convolution(DC)into the HED framework.First,the pooling layers in the last two layers of the original HED network are removed to better pre-serve edge information.Dilated convolution is the incorporated to expand the receptive field and enhance the restoration of edge details,leading to the re-design of the DC-HED network.Subsequently,the Zhang-Suen algorithm is applied to extract the skeleton from the image edge detection results.The DC-HED network and skeleton extraction were tested on edge detec-tion tasks for core cast thin section images(referred to as core images)from the S oil field in northern Shaanxi,China.The experimental results demonstrate that,compared to traditional methods such as the Canny and Sobel operators and the original HED network,the DC-HED network produces more complete and connected edges.Specifically,the DC-HED netword a-chieved a mean squared error of 0.1106,a structural similarity index of 0.7997,and a peak signal-to-noise ratio of 9.5611,all significantly improved over previous methods.Finally,applying skeleton extraction to the detected edges effectively re-moves clutter and produces a clear and continuous central contour of the image edges.关键词
岩心铸体薄片图像/边缘检测/岩心数字化/HED网络/扩张卷积/骨架提取Key words
petrographic thin section image/edge detection/core digitization/HED network/dilated convolution/skeleton extraction分类
石油、天然气工程引用本文复制引用
潘少伟,杨怡婷,尚娅敏,郭智,蔡文斌..基于DC-HED网络和骨架提取的岩心图像边缘检测[J].中国石油大学学报（自然科学版）,2025,49(3):97-107,11.基金项目
国家自然科学基金项目(52074225) （52074225）
