基于卷积神经网络的碳酸盐岩生物化石显微图像识别

Microscopic recognition of micro fossils in carbonate rocks based on convolutional neural network

余晓露 叶恺 杜崇娇 宫晗凝 马中良

YU Xiaolu YE Kai DU Chongjiao GONG Hanning MA Zhongliang

中国石化 油气成藏重点实验室, 江苏 无锡 214126 中国石化 石油勘探开发研究院 无锡石油地质研究所, 江苏 无锡 214126

SINOPEC Key Laboratory of Petroleum Accumulation Mechanisms, Wuxi, Jiangsu 214126, China Wuxi Research Institute of Petroleum Geology, SINOPEC, Wuxi, Jiangsu 214126, China

碳酸盐岩薄片中的生物化石识别对判断沉积环境研究具有重要的意义，但传统的人工鉴定方法对经验要求高，受主观影响较大。该文提出一种基于ResNet卷积神经网络的碳酸盐岩生物化石显微图像识别方法，通过图像预处理、设计模型、训练模型等步骤，实现了薄片图像中生物化石的智能识别，识别准确率为86%；并同时提出进阶YOLO（You Only Look Once）目标检测模型，可实现薄片图像中生物化石所在区域的检测和识别，识别准确率为85%。该方法验证了使用数字图像处理和深度学习方法对碳酸盐岩生物化石显微图像进行智能识别的可行性，可作为传统人工鉴定方法的有益补充，具有一定的实际应用价值。

The identification of microfossils in carbonate rocks with thin-section observation is of great significance for the study of sedimentary environment, but the traditional method by manual identification is highly experience required and is greatly affected by subjective factors.In this paper, a method for microscopic recognition of carbonate rocks based on ResNet convolutional neural network was introduced. Through image preprocessing, model design, model training etc., the intelligent recognition of fossils of organisms within thin section images were realized, and the recognition accuracy showed to be 86%.Meanwhile, an advanced YOLO(You Look Only Once) object detection model was proposed, which could realize the detection and recognition of the area where the organism locates in thin section image, and the recognition accuracy appeared to be 85%.This method verified the feasibility of using digital image processing algorithm and deep learning method to intelligently identify biological microscopic images of carbonate rocks.It can be regarded as a useful supplement to traditional manual identification methods and has certain practical application value.

中国石化优秀青年科技创新项目“岩石（矿物）自动化鉴定分析仪” P19028;国家自然科学基金 42072156

https://doi.org/10.11781/sysydz202105880