砂样岩屑图像特征的岩性智能高效识别

Intelligent and efficient lithology identification based on image features of returned cuttings

在录井过程中,岩屑的岩性分析主要依靠人工,效率较低且稳定性较差,难以在钻进地层过程中快速识别岩性变化。为此,提出基于砂样图像中颗粒岩屑纹理、色泽和形状等特征的岩性智能识别方法。首先,计算砂样图像的像素值梯度并求取颗粒质心,采用分水岭算法获取颗粒岩屑轮廓线并标记;然后,采用图像分割算法从砂样图像中分离出待检测的单个颗粒岩屑图像,建立颗粒岩屑图像样本库;最后,利用注意力机制及特征融合模块改进MobileNetV2网络,提取颗粒岩屑特征并分类,实现单个颗粒岩屑图像岩性识别,进而获取砂样岩性成分比。该方法将以往岩性智能识别过程中常采用的砂样整体识别方式转变为对砂样中单颗粒岩屑的岩性识别,大幅度减少了颗粒岩屑之间的相互干扰。多个油气区块的砂样图像测试结果表明,该方法对灰岩、泥岩、砂岩和页岩的识别准确率均不低于92%,一组砂样图像岩性分析的用时小于10 s。

In the process of cutting logging,the lithology of cuttings is mainly manually analyzed,which is inefficient and unstable,and it is difficult to quickly identify the lithology change of the stratum during the drilling.Therefore,an intelligent lithology identification method based on the texture,color,and shape characteristics of rock particles in returned cuttings images is proposed.Firstly,the gradient of the pixel value of the returned cuttings image is calculated,and the particle centroid is obtained.The contour line of the rock particles is obtained and marked by using the watershed algorithm.Then the image segmentation algorithm is used to separate the single rock particle image to be detected from the returned cuttings image,and a sample library of rock particle images is established.Finally,the MobileNetV2network is improved by using the attention mechanism and feature fusion module to extract and classify the features of rock particles,so as to identify the lithology of a single rock particle image and then obtain the composition ratio of the returned cuttings sample.The proposed method transforms the overall identification method of returned cuttings,which is often used in the intelligent identification of lithology,into the lithology identification method of a single rock particle in returned cuttings,which greatly filters out the mutual interference of rock particles.The test results of returned cuttings images collected from several oil and gas blocks show that the proposed method can reach an accuracy of more than 92%in identifying limestone,mudstone,sandstone,and shale,and the time consumption for lithology analysis of a group of returned cuttings images is less than 10s.