基于色彩空间和深度残差网络ResNet-50的复杂岩性油气藏岩石样本智能分类及识别

岩石样本的分类识别是油气和矿产资源勘探中的重要环节。目前，仍然以人工识别的实验方法作为主要方法，普遍存在主观性强、周期长、成本高等典型问题。机器学习的分类算法在图像分类领域已经得到广泛应用，然而由于岩石样本图像具有明显的差异性特征，甚至同类岩石样本图像也具有一定的色差，直接应用现成智能算法进行分类，验证集的准确度仅为85%左右。所以，基于色彩空间下岩石样本图像的颜色特征曲线，提出了一种基于颜色类别和深度残差网络ResNet-50的智能分类及识别方法。首先，以7种不同岩性的岩石图像为样本，提取样本的RGB颜色特征，应用无监督K-means聚类算法，按颜色分为3个大类，再通过有监督精细K-近邻(K-nearest neighbor, KNN)算法对颜色类别进行验证，平均分类精度为99%。然后，对于不同颜色类别下的岩石样本，利用深度残差网络ResNet-50进行分类识别。结果表明，不同颜色类别的岩石样本平均训练精度为93.15%,验证精度为88.21%,可以作为岩石样本分类的有效方法。

Intelligent classification and recognition of rock samples based on color space and deep residual network resnet-50

The classification and identification of rock samples is an important step in the exploration of oil and mineral resources. At present, the experimental method of manual identification is still used as the main method, which generally has the typical problems of strong subjectivity, long time, and high cost. Therefore, an intelligent classification and recognition method based on color space and deep residual network RESNET-50 is proposed. Firstly, the RGB color features of the samples were extracted from the rock pictures of seven different lithologies. Then, the unsupervised k-means clustering algorithm was applied to divide the samples into three categories according to color. Then, the supervised fine KNN algorithm was used to verify the color categories, and the classification accuracy was over 98%. Then, for rock samples under different color categories, the deep residual network RESNET-50 is used for classification and recognition. The results show that the recognition accuracy of various rock samples with different color categories are higher than 88.21%, and the training time is reasonable, with high accuracy and reliability.