摘要
针对在PDC钻头条件下石油录井过程中的岩屑岩性难以用肉眼识别,以及传统分类器准确率低的问题,该文利用和差直方图变换、傅里叶变换、Gabor变换3种不同的岩屑图像特征提取技术,将岩屑样品进行特征提取,并设定LIBSVM分类器的分类参数,根据得到的岩性识别率的高低来确定最优的参数组合。结果表明,SVM设置类型上选择NU_SVC、核函数选择RBF、分类参数gamma设置为0.1、nu设置为0.11的参数组合可获得较高的岩性识别率。其中采用Gabor特征提取方式和LIBSVM分类器对现场岩屑样品进行测试,获得的岩性识别率最高,对现场泥岩和砂岩的平均识别率分别为95%和90%。与神经网络分类器分类结果比较,该方法能更好地分析岩屑的岩性,可望在现场得以推广应用。
In the process of logging, it＇s quite low for traditional sorter to identify the cuttings made by the PDC bit. In order to extract characteristics of cuttings＇ image texture effectively, a method based on LIBSVM based for identifying of cuttings has been developed and described in this paper. The Sum and Difference Histograms, Fourier Transform and Gabor Transform are used in extracting procedure with different combination. To improve the identifying accuracy, the parameters of the LIBSVM classifier are modulated and tested with the samples from Shengli oilfield （Dongying）. The average identifying accuracy about 95% for mudstone and 90% for sandstone have been achieved with the accuracy distribution around 81%-100%. The obtained results show that the method which uses LIBSVM for identifying of cuttings in petroleum logging has great potential to be developed as an effective identification method for flied application.
出处
《中国海洋大学学报（自然科学版）》
CAS
CSCD
北大核心
2010年第9期131-136,共6页
Periodical of Ocean University of China
基金
国家高技术研究发展规划项目(202AA615170)
国家自然科学基金项目(40706037)
中国石化胜利油田有限公司技术开发项目资助
