花岗岩矿物正交偏光镜下图像人工智能识别研究  

A study on identification of minerals in crossed polarizer images of granites using the artificial intelligence

作　　者：王名越 狄永军[1] 张春禹 WANG Mingyue;DI Yongjun;ZHANG Chunyu(School of Earth Sciences and Resources,China University of Geosciences Beijing,Beijing 100083,China)

机构地区：[1]中国地质大学(北京)地球科学与资源学院,北京100083

出　　处：《矿物学报》2024年第1期11-23,共13页Acta Mineralogica Sinica

基　　金：中国地球物理场与成矿关系(编号:KD-[2019]-XZ-064,KD-[2020]-XZ-044,KD-[2021]-XZ-049)。

摘　　要：人工智能在地球科学领域中的应用是近年来研究的热点,对地球科学的发展有着重要意义,其应用之一就是使用计算机视觉技术实现岩石或矿物的自动化识别分类。然而,目前大多研究直接对岩石薄片图像进行分类,不能够精细定位并识别薄片中多且复杂的矿物目标。虽然现已有许多学者将目标检测技术应用于岩石矿物的图像识别分类中,但这些方法识别的对象大多是岩石手标本图像,只能对图像中的单一对象检测。在识别分类研究领域中,缺少对岩石薄片镜下图像识别的算法及质量较好的相关数据集。为了解决这些问题,本文首先采集了3000张正交偏光镜下花岗岩薄片图像,标注矿物样本10000余个,并通过数据增广方式对数据集进行增强,建立了一个质量较好、具有多样性的数据集。其次本文提出基于Yolov5x的改进算法RDB-Yolov5x。这种方法在特征提取过程中添加了密集连接方式,使用密集连接残差模块(RDB)替代传统的残差结构,有效地保留了图像的语义和位置信息细节。实验结果表明该方法泛化能力较好,在对图像中小尺寸、特征模糊的矿物颗粒的识别中表现出优秀的性能,可以准确有效地对花岗岩中的五类目标矿物(石英、黑云母、白云母、斜长石、钾长石)进行识别,平均精度均值m AP达到94.1%。较对比方法,在IoU阈值为50%时提高了0.5%,阈值为95%时提高了1%。The application of artificial intelligence in field of geosciences becomes a research hot spot in recent years.It is of great significance to the development of geosciences.One of applications of artificial intelligence is to achieve automated identification and classification of rocks or minerals using the computer vision technique.However,most researches are generally focused on the classification of rock type rather than the directly precise identification of multiple and complex minerals in the rock based on its thin section images.Although the object detection technique has been applied by many scholars to identify and classify types of rocks and minerals based on images,its application objects are mostly rock hand specimen images,and it can only be used to detect a single object in the image.In the research field of identification and classification,there is a lack of good quality algorithms and datasets for identidying and classifying minerals in the rock based on the rock thin section images.In order to solve these problems,firstly,we have collected more than 3000 images of thin sections of granite under crossed polarizer microscope,have labeled more than 10000 mineral samples on those images,have enhanced the dataset by means of data augmentation and have established a dataset with good quality and diversity.Secondly,we have proposed an improved algorithm of the RDB-Yolov5x based on the Yolov5x.In this method,the dense connection method was added in the feature extraction process,and the residual dense block(RDB)was used to have replaced the traditional residual structure.Thus,the semantic and location information details of images can be effectively preserved by using this method.Experimental results showed that this method had good generalization capability,and excellent performance for identifying small sized and fuzzy characterized mineral grains in images.By using this method,we have accurately and effectively identified five kinds of targeting minerals(quartz,biotite,muscovite,plagioclase,potassium

关 键 词：人工智能识别 岩石矿物 RDB-Yolov5x 镜下图像 

分 类 号：P575[天文地球—矿物学] TP391.4[天文地球—地质学]