军事知识服务平台
基于频域图像分析的矿山微震信号分类方法研究
Study on classification method of mine microseismic signals based on frequency-domain image analysis
摘要 针对矿山微震分类传统方法存在特征提取依赖经验、分类准确率有限等问题,提出一种基于频域图像识别的矿山微震信号分类新方法.首先,模拟了 3类典型矿山微震信号(Ⅰ类岩石微震、Ⅱ类岩石微震和爆破震动),并分析了不同类别信号的时域特征;然后,采用短时傅里叶变换(STFT)、连续小波变换(CWT)和格拉姆角场(Gramian Angular Field,GAF)方法将一维微震信号转换为频域图像;其次,在此基础上设计了基于频域图像的深度学习分类模型,包括融合注意力机制的残差网络(ResNet-CBAM);最后,通过F1、精确度等评价指标对模型在模拟数据上的分类性能进行评估.结果表明,基于频域图像的深度学习模型能够有效提取微震信号的差异性特征,实现岩石微震与爆破震动事件的高准确率分类.
Abstract:
Addressing the issues of traditional methods for mine microseismic classification,such as dependence on experience for fea-ture extraction and limited classification accuracy,a new method for mine microseismic signal classification based on frequency domain image recognition was proposed.Firstly,three types of typical mine microseismic signals(Type-Ⅰ rock microseisms,Type-Ⅱ rock micro-seisms,and blast-induced vibrations)were simulated,and their respective time-domain characteristics were analyzed;subsequently,one-dimensional microseismic signals were transformed into frequency-domain images by using Short-Time Fourier Transform(STFT),Contin-uous Wavelet Transform(CWT),and Gramian Angular Field(GAF);secondly,based on this,a deep learning classification model based on frequency-domain images was developed,specifically a residual network integrated with the attention mechanism(ResNet-CBAM);final-ly,the classification performance of the model on simulated data was evaluated using evaluation metrics such as F1-score and accuracy.The results showed that the deep learning model based on frequency domain images could effectively extract the differential features of mi-croseismic signals,achieving high classification accuracy between rock microseismic events and blast-induced vibration events.
刊 名 能源与环保
年,卷(期): 2026, 48(7)
分类号: TD326
机标分类号: TP391.41 TN911.73 R734.2
在线出版日期: 2026年10月19日
基金项目: 2024507:应急管理部信息研究院科技创新发展项目,