2020-2021年陕晋冀地区五种破坏类型井下构造煤的深度学习图像数据集  

A dataset of deep learning images of underground tectonic coal with five destruction types in Shaanxi,Shanxi and Hebei(2020-2021)

作　　者：冯永安[1,2] 韩晓天 刘铁 吕伏 FENG Yong’an;HAN Xiaotian;LIU Tie;LÜFu(Ordos Institute,Liaoning Technical University,Ordos 017000,P.R.China;Software College,Liaoning Technical University,Huludao 125105,P.R.China;CCTEG Shenyang Engineering Co.,Ltd.,Fushun 113122,P.R.China)

机构地区：[1]辽宁工程技术大学,鄂尔多斯研究院,内蒙古鄂尔多斯017000 [2]辽宁工程技术大学,软件学院,辽宁葫芦岛125105 [3]中煤科工集团沈阳研究院有限公司,辽宁抚顺113122

出　　处：《中国科学数据（中英文网络版）》2024年第4期376-383,共8页China Scientific Data

基　　金：国家自然科学青年基金(51904144);国家自然科学基金面上项目(52274206)。

摘　　要：井下煤体破坏类型的机器视觉识别可以为煤与瓦斯突出事故的预测、井下安全的评估和智慧矿山工程的建设提供信息支撑和技术支持。目前煤矿中构造煤的煤体破坏类型判别仍以经验性地宏观物理观察辨识为主,但该方法受主观经验和环境因素影响较大。构建构造煤的图像分类机器学习模型需要高质量、均衡、真实的井下构造煤样本图像。本数据集通过人为携带高清防爆相机在多个不同矿井环境下进行拍摄采集,提供了可用于深度学习的井下原生状态破坏煤图像集。数据集大小约为1.55GB,包含不同矿井环境、光照、角度下的非破坏煤、破坏煤、强烈破坏煤、粉碎煤和全粉煤,共计1031张图像,平均每类206张,能够满足深度学习模型的基础训练需要。在数据采集过程中,采取了严格的数据质量控制措施,对图像的清晰度、真实性等进行了评估,确保了数据的可靠性。本数据集针对构造煤破坏类型进行构建,为下一阶段基于机器学习的煤体破坏类型识别研究提供必要的数据基础和研究素材。Machine vision recognition of underground coal destruction types can provide crucial information and technical support for predicting coal and gas herniation accidents,evaluating underground safety,and advancing the construction of intelligent mine engineering.At present,the discrimination of coal destruction types of tectonic coals in mines still relies on empirical macro-physical observations,which are heavily influenced by subjective experience and environmental conditions.To develop an effective machine learning model for classifying tectonic coal destruction types,it is essential to use highquality,balanced,and realistic sample images of underground tectonic coal.Based on the photos taken by human-carried high-definition explosion-proof cameras in several different mine environments,this dataset provides a set of downhole native-state destructive coal images suitable for deep learning.The dataset size is about 1.55 G,including images of Non-destructive coal,Destructive coal,Strongly damaged coal,Pulverized coal and Fully pulverized coal under different mine conditions,illumination and angles.A total of 1,031 images,with an average of 206 images per class,can meet the basic training needs of deep learning models.During the data collection process,strict data quality control measures were taken,and the clarity and authenticity of the images were evaluated to ensure the reliability of the data.Designed for recognizing tectonic coal destruction types,this dataset is expected to provide essential resources for the future machine learning-based research on the recognition of coal destruction types.

关 键 词：构造煤 煤体破坏类型 井下环境 深度学习 识别 训练 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] TP391.41[自动化与计算机技术—控制科学与工程] TD163.1[矿业工程—矿山地质测量]