深度卷积神经网络目标检测算法在煤矿断层检测上的应用  

Deep convolutional neural network target detection algorithm for coal mine fault detection

作　　者：张春翔 唐烨锈 邹冠贵[1,2] 曾义文 樊卓 Zhang Chunxiang;Tang Yexiu;Zou Guangui;Zeng Yiwen;Fan Zhuo(College of Geosciences and Surveying Engineering,China University of Mining and Technology-Beijing,Beijing 100083,China;State Key Laboratory of Coal Resources and Safe Mining,Beijing 100083,China;School of Mechanics and Civil Engineering,China University of Mining and Technology-Beijing,Beijing 100083,China)

机构地区：[1]中国矿业大学(北京)地球科学与测绘工程学院,北京100083 [2]煤炭资源与安全开采国家重点实验室,北京100083 [3]中国矿业大学(北京)力学与土木工程学院,北京100083

出　　处：《矿业科学学报》2023年第6期733-743,共11页Journal of Mining Science and Technology

基　　金：国家重点研发计划(2018YFC0807803);国家自然科学基金(2022Z01010029);中国矿业大学(北京)大学生创新训练(202202050)。

摘　　要：断层解释技术在煤矿安全开采领域起着至关重要的作用。随着神经网络技术的发展,许多基于神经网络算法的智能化地震资料解释处理方案被提出。首先通过对比不同的深度卷积神经网络目标检测算法,选择适合于识别断层的Faster R-CNN目标检测算法;其次建立具有多种地质特征的地震正演模型,分别对AlexNet、残差网络ResNet50和ResNet101三种特征提取网络进行测试,优选得出ResNet101特征提取网络在断层检测方面具有更加优异的表现;最后基于优选的ResNet101特征提取网络和Faster R-CNN目标检测算法构建断层检测模型,对实际地震数据进行断层检测。结果表明:基于深度卷积神经网络的目标检测算法在断层检测上具有很好的泛化能力,提高了断层的解释效率,在实际应用上具有巨大潜力。Fault interpretation plays an important role in the field of coal mine safety.The development of neural network gives rise to many intelligent seismic data interpretation and processing schemes based on neural network algorithm.This study①selected the Faster R-CNN target detection algorithm more suitable for fault recognition by comparing different deep convolutional neural network target detection algorithms.②tested AlexNet,residual network ResNet50 and ResNet101 feature extraction networks through seismic forward modeling with various geological characteristics.It is found that ResNet101 fea-ture extraction network has better performance in fault detection.③constructed a fault detection model based on the preferred ResNet101 feature extraction network and Faster R-CNN target detection algo-rithm,and detected the actual seismic data.Results show that the object detection algorithm based on deep convolutional neural network shows satisfactory generalization ability in fault detection.It could im-prove the fault interpretation efficiency,and has potential in application.

关 键 词：深度卷积神经网络 断层检测 地震正演模型 目标检测 

分 类 号：P315.2[天文地球—地震学]