基于深度学习的煤岩识别GPR信号去噪方法研究
Research on Denoising Method of GPR Signal for Coal-Rock Identification Based on Deep Learning
摘要: 煤岩识别研究可以定位煤层和岩层的分界面，确定截割位置，从而指导采煤机自动调高，实现更高水平的智能化无人开采。探地雷达是一种无损的探测手段，可以获取地下岩层的信息，识别岩性。但是探地雷达的分辨率受到信号频率和波长的限制，不同设备测量得到的数据存在差异，且信号容易被噪声干扰。本研究使用了一种U-Net结构的神经网络算法处理无线天线探地雷达数据，使其信号质量接近有线天线测得的信号水平，并使用改进的U-Net网络对信号进行去噪处理，实现了无线到有线转换和去噪后的图像。结果表明，经过U-Net处理，无线信号质量得到了较大的提升，雷达信号图像的平均峰值信噪比(PSNR)提高了约4 dB、平均结构相似性指数(SSIM)达到了0.935，证明了降低数据差异的GBR数据具有更好的去噪效果。
Abstract: The research on coal-rock identification can locate the interface between coal and rock seams and determine the cut-off position, thus guiding the coal miner to automatically adjust the height and realize a higher level of intelligent unmanned mining. Ground-penetrating radar (GPR) is a non-destructive means of detection, which can obtain the information about underground rock layers and identify the lithology. However, the resolution of GPR is limited by signal frequency and wavelength, the data obtained from different equipment measurements differ, and the signal is easily interfered by noise. In this study, a neural network algorithm with a U-Net structure is used to process the wireless antenna GPR data, so that the signal quality is close to the level of the signal measured by a wired antenna, and the signal is denoised using an improved U-Net network to realize the wireless-to-wired conversion and denoised image. The results show that after U-Net processing, the quality of the wireless signal is greatly improved, the average peak signal-to-noise ratio (PSNR) of the radar signal image is improved by about 4 dB and the average structural similarity index (SSIM) reaches 0.935, which proves that the GBR data with reduced data discrepancy has a better denoising effect.
文章引用：蒲明松, 姜绪超. 基于深度学习的煤岩识别GPR信号去噪方法研究[J]. 软件工程与应用, 2024, 13(4): 488-500. https://doi.org/10.12677/sea.2024.134051

参考文献

[1] 张强, 张润鑫, 刘峻铭, 等. 煤矿智能化开采煤岩识别技术综述[J]. 煤炭科学技术, 2022, 50(2): 1-26.
[2] 任望, 姚振岸, 陈龙凤, 等. 浅析探地雷达技术的发展历程[J]. 江西科学, 2024, 42(1): 100-107.
[3] Nekkanti, A. and Schmidt, O.T. (2021) Frequency-Time Analysis, Low-Rank Reconstruction and Denoising of Turbulent Flows Using SPOD. Journal of Fluid Mechanics, 926, A26. [Google Scholar] [CrossRef] 
[4] 刘国宏, 郭文明. 改进的中值滤波去噪算法应用分析[J]. 计算机工程与应用, 2010, 46(10): 187-189.
[5] 崔玲丽, 刘银行, 王鑫. 基于改进奇异值分解的滚动轴承微弱故障特征提取方法[J]. 机械工程学报, 2022, 58(17): 156-169.
[6] 陈晓曦, 王延杰, 刘恋. 小波阈值去噪法的深入研究[J]. 激光与红外, 2012, 42(1): 105-110.
[7] 孙志军, 薛磊, 许阳明, 等. 深度学习研究综述[J]. 计算机应用研究, 2012, 29(8): 2806-2810.
[8] 陈景霞, 林文涛, 龙旻翔, 等. 基于UNet 3+生成对抗网络的视频异常检测[J]. 计算机工程与设计, 2024, 45(3): 777-784.
[9] Chua, L.O. (1998) CNN: A Paradigm for Complexity. World Scientific. [Google Scholar] [CrossRef] 
[10] 杜渺勇, 于祥雨, 周浩, 等. 基于自编码卷积神经网络的图像去噪算法[J]. 杭州师范大学学报(自然科学版), 2021, 20(1): 95-101.
[11] 钟建军, 宋健, 由长喜, 等. 基于信噪比评价的阈值优选小波去噪法[J]. 清华大学学报(自然科学版), 2014, 54(2): 259-263.
[12] 蒲秋梅, 殷帅, 李正茂, 等. U型卷积网络在乳腺医学图像分割中的研究综述[J]. 计算机科学与探索, 2024, 18(6): 1383-1403.
[13] Huang, H., Lin, L., Tong, R., Hu, H., Zhang, Q., Iwamoto, Y., et al. (2020). UNet 3+: A Full-Scale Connected UNet for Medical Image Segmentation. 2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Barcelona, 4-8 May 2020, 1055-1059.[CrossRef] 
[14] 杜松杰, 刘宏哲, 潘卫国, 等. 基于深度学习的低光照图像增强研究综述[C]//中国计算机用户协会网络应用分会2023年第二十七届网络新技术与应用年会. 中国计算机用户协会网络应用分会2023年第二十七届网络新技术与应用年会论文集. 2023: 34-37.
[15] 刘利平, 乔乐乐, 蒋柳成. 图像去噪方法概述[J]. 计算机科学与探索, 2021, 15(8): 1418-1431.
[16] Palubinskas, G. (2016) Image Similarity/Distance Measures: What Is Really behind MSE and SSIM? International Journal of Image and Data Fusion, 8, 32-53. [Google Scholar] [CrossRef] 

为你推荐