基于小波变换与迁移学习的岩石岩性及含水状态超声波检测  

Ultrasonic detection of rock lithology and water-bearing status based on wavelet transform and transfer learning

作　　者：张胜[1] 黄宁 陈梓浓 凌同华[2] 张亮 ZHANG Sheng;HUANG Ning;CHEN ZiNong;LING TongHua;ZHANG Liang(School of Civil Engineering,Hunan City University,Yiyang 413000,China;School of Civil Engineering,Changsha University of Science and Technology,Changsha 410114,China)

机构地区：[1]湖南城市学院土木工程学院,益阳413000 [2]长沙理工大学土木工程学院,长沙410114

出　　处：《地球物理学进展》2024年第6期2407-2419,共13页Progress in Geophysics

基　　金：湖南省教育厅科学研究项目(22A0562和23C0664);湖南省交通运输厅科技项目(202319)联合资助。

摘　　要：在智能隧道设计与建设中,迅速而准确地识别隧道围岩的岩石岩性与含水状态至关重要.本文针对现有识别方法智能化程度低、主观性强以及识别周期长等问题,提出了一种基于超声波检测信号时频图像与轻量级卷积神经网络的识别方法.该方法通过分析岩石超声波检测信号与小波基波形的相似性,选取最优小波基,利用连续小波变换构建岩石超声波信号时频图像数据集;再通过预训练和迁移学习建立岩石岩性及含水状态的识别模型.与GoogLeNet、AlexNet以及SqueezeNet等模型相比,本方法在岩石超声波检测信号的识别上更为精确,具有高准确率、稳定性好以及低计算存储需求等优点,适合应用于移动和嵌入式设备,为复杂环境下的岩石岩性及含水状态快速在线识别提供了新方案.In the realm of intelligent tunnel design and construction, it is crucial to determine the lithology and water content of surrounding rock rapidly and precisely. Based on ultrasonic detection signal time-frequency image and lightweight convolutional neural network, this paper introduces an innovative method that addresses limitations of current recognition methods, such as low automation, high subjectivity, and lengthy recognition processes. Utilizing time-frequency images from ultrasonic detection signals, coupled with lightweight convolutional neural networks, this method effectively analyzes rock ultrasonic detection signals against wavelet basis waveforms. It selects the optimal wavelet basis and employs continuous wavelet transform to create comprehensive time-frequency image datasets. The method further develops a reliable model for identifying rock lithology and water content using pre-training and transfer learning techniques. Compared to established models like GoogLeNet, AlexNet, and SqueezeNet, this proposed approach demonstrates superior precision in recognizing ultrasonic signals from various rock types. Its high accuracy, stability, and low computational demands make it particularly suitable for mobile and embedded devices, offering a novel and efficient solution for the rapid online identification of rock properties in complex environments.

关 键 词：岩石 超声波检测 小波变换 迁移学习 含水状态 

分 类 号：P631[天文地球—地质矿产勘探]