基于YOLOv11的岩心识别与编录后处理算法实现-《计算机技术与发展》

文章信息/Info

Title:: Implementation of Core Recognition and Catalog Post-processing Algorithm Based on YOLOv11

Author(s):: JIN Wei-xi1; 2; 1. China Railway First Survey and Design Institute Group Co. ,Ltd. ,Xi'an 710043,China; 2. China Railway Construction Corporation,Beijing 100089,China

Keywords:: rock core recognition; YOLOv11; instance segmentation; data augmentation; post-processing algorithm

摘要:: 传统岩心分析存在人工观察与记录效率低且主观性强的问题,对此提出一种基于YOLOv11的岩心识别与编录后处理算法。该算法结合高精度的实例分割网络和尺度标定机制,采用YOLOv11n-seg作为主干网络进行岩心图像的实例分割任务。针对样本不平衡问题,设计了动态数据增强策略,通过几何变换、光度畸变增强和过采样技术,使数据总量扩充至原数据集的2.34倍,有效缓解了类别分布不均衡问题。后处理流程包括标定区域提取、尺度转换、岩心区域分割、掩码分析、岩性分类与深度累加等步骤,实现了从原始岩心图像到结构化地质数据的自动转换。提出基于长度的识别准确率评估指标,更符合岩心识别应用场景。实验结果表明,在8个测试钻孔上的平均识别准确率达到87.68%,最高达到97.43%。该算法显著提高了岩心分析的效率与精度,为地质勘探工作提供了科学可靠的数据支持。

Abstract:: Traditional rock core analysis faces problems of low efficiency and high subjectivity in manual observation and recording. A rock core recognition and post-processing algorithm based on YOLOv11 is proposed to address these issues. The algorithm combines high-precision instance segmentation networks with scale calibration mechanisms,utilizing YOLOv11 n-seg as the backbone network for rock core image instance segmentation tasks. Dynamic data augmentation strategies are designed to tackle sample imbalance problems.These strategies employ geometric transformations,photometric distortion enhancement,and oversampling techniques. The total data volume is expanded to 2.34 times the original dataset,effectively alleviating class distribution imbalance issues. Post-processing workflow includes calibration area extraction,scale conversion,rock core area segmentation,mask analysis,lithology classification,and depth accumulation. This workflow achieves automatic conversion from original rock core images to structured geological data. Length-based recognition accuracy evaluation metrics are proposed to better suit rock core recognition application scenarios. Experimental results demonstrate average recognition accuracy reaching 87.68% across eight test boreholes,with the highest accuracy at 97.43%. The proposed algorithm significantly improves efficiency and precision of rock core analysis. Scientific and reliable data support is provided for geological exploration work.