Soil water content is one of the most important factors affecting the safety and stability of buildings or structures, especially in roadbeds, slopes, earth dams and foundations. Accurate assessments of soil water content can ensure the quality of construction, reduce construction costs and prevent accidents, among other benefits. In this study, ground penetrating radar (GPR) was used to detect and evaluate changes in soil water content. The GPR signal is usually nonstationary and nonlinear; however, traditional Fourier theory is typically suitable for periodic stationary signals, and cannot reflect the law of the frequency and energy of the GPR signal changing with time. Wavelet transform has good time-frequency localization characteristics, and therefore represents a new method for analyzing and processing GPR signals. According to the time-frequency characteristics of GPR signals, in this paper, a new biorthogonal wavelet basis which was highly matched with the GPR waveform was constructed using the lifting framework of wavelet theory. Subsequently, an evaluation method, namely, the wavelet packet-based energy analysis (WPEA) method, was proposed. The method was utilized to calculate the wavelet packet-based energy indexes (WPEI) of the GPR single-channel signals for clay samples with water contents ranging from 10% to 24%. The research results showed that there was a highly correlated linear relationship between the WPEI and the soil water contents, and the relationship between the two was fitted with a linear fitting function. The feasibility of the method was verified by comparing our results with those obtained using classical wavelet bases to perform the wavelet packet transform. The large-area, continuous scanning measurement method of GPR was shown to be suitable for evaluations of soil water contents in roadbeds, slopes, earth dams, and foundations.

The hard rocks in the stratum can pose safety risks and hinder the progress of urban underground tunnel construction using shield and jacking methods, thereby reducing construction efficiency and increasing construction costs. This paper utilizes wavelet scale energy spectrum, wavelet packet theory and statistical methods to conduct research on the detection of special geological formations such as hard rocks and voids, as well as the analysis of their signal time-frequency characteristics based on the ground-penetrating radar (GPR) technique. On the basis of calibrating the permittivity of different types of rock blocks, we established a forward model for detecting hard rocks and voids, and the simulated signals were analyzed in the time and frequency domains. Subsequently, laboratory experiments were conducted to perform GPR tests on different types of hard rocks in natural and water-saturated states and voids, to explore the time-frequency characteristics, frequency band energy variations, and statistical patterns of typical single-trace signals. The results show that the granite detection signal contains more low-frequency components, the sandstone detection signal contains more medium-low frequency components, while the limestone detection signal contains more medium-high frequency components in their natural state; the signal from the karst cave has relatively more low-frequency components than the signal from the empty cavity. The geometric shape of the rock has no influence on the dominant frequency and time-frequency distribution of its reflection signal. Generally, rocks with higher rebound values (hardness) also exhibit larger variance and standard deviation in frequency band energy. The research has important theoretical significance and practical value for the measurement and assessment of special geological features such as hard rocks and voids in urban underground trenchless construction.

Advanced geological prediction of tunnels has become an indispensable task to ensure the safety and effectiveness of tunnel construction before excavation in karst areas. Geological disasters caused by unfavorable geological conditions, such as karst caves, faults, and broken zones ahead of a tunnel face, are highly sudden and destructive. Determining how to predict the spatial location and geometric size of unfavorable geological bodies accurately is a challenging problem. In order to facilitate a three-dimensional quantitative analysis of the filler material ahead of the tunnel face, a biorthogonal wavelet with short support, linear phase, and highly matching waveform of ground penetrating radar (GPR) wavelet is constructed by lifting a simple and general initial filter on the basis of lifting wavelet theory. A method for a time-energy density analysis of wavelet transforms (TEDAWT) is proposed in accordance with the biorthogonal wavelet. Fifteen longitudinal and horizontal survey lines are used to detect void fillers of different heights. Then, static correction, DC bias, gain, band-pass filtering, and offset processing are performed in the original GPR profile to enhance reflected signals and converge diffraction signals. A slice map of GPR profile is generated in accordance with the relative position of longitudinal and horizontal survey lines in space. The wavelet transform analysis of a single-channel signal of each survey line is performed by adopting the TEDAWT method because of the similar rule of the single-channel signal of GPR on the waveform overlay and the ability of the constructed wavelet basis to highlight the time-frequency characteristics of GPR signals. The characteristic value points of the first and second interfaces of the void fillers can be clearly determined, and the three-dimensional spatial position and geometric sizes of different void fillers can be obtained. Therefore, the three-dimensional visualization of GPR data is realized. Results show that the TEDAWT method has a good practical application effect in the quantitative identification of void fillers, which provides a basis for the interpretation of advanced geological prediction data of tunnels and for the construction decision.

Geological investigation and surrounding rock classification ahead of tunnel faces are among the most critical fundamental tasks in tunnel construction. However, current methods for obtaining rock mass parameters and conducting grade evaluation are plagued by significant subjectivity, prolonged cycles, and insufficient intelligence. To address these issues, this study takes the Shitang Tunnel, Zaoxi Tunnel, Jiangjunshan Tunnel, Kuiba Tunnel, Dapingwei Tunnel as engineering cases, integrating geological survey data from tunnel faces and ground penetrating radar (GPR) advance detection data. A dynamic rock mass classification method during tunnel construction is then proposed based on deep learning and digital image processing techniques. The proposed method employs MobileNetV3 to intelligently identify key features in tunnel face images, including rock lithology, rock mass structure type, fracture development degree, weathering degree, and water content, achieving an identification accuracy exceeding 95%. Additionally, MobileNet-single shot multibox detector is utilized for rapid intelligent interpretation of GPR images, with an identification precision of 96.57%. By inputting the recognition results into the basic quality classification formula, real-time dynamic determination of rock mass grades is achieved, yielding a prediction accuracy of 85.7%. The results demonstrate that this method significantly enhances the intelligence level and operational efficiency of tunnel rock mass classification, providing a novel technical approach for dynamic design and construction optimization in tunnel engineering.

Ground-penetrating radar (GPR) technology is a crucial means for detecting voids in tunnel linings, yet manual interpretation of its images suffers from inefficiencies and significant subjective judgment biases. Initially, the synchrosqueezing wavelet transform (SWT) is employed for time–frequency analysis of GPR signals, followed by the introduction of the ResNet34 network augmented with a CBAM attention module. An intelligent recognition model is established using the enhanced ResNet34 network, aiming to improve the accuracy and efficiency of identifying voids with different filling media behind the initial tunnel support. The results indicate that SWT can more effectively highlight the time–frequency characteristics of radar signals compared to traditional methods such as short-time Fourier transform, generalized S-transform, and continuous wavelet transform. The method combining SWT with an improved residual neural network achieves recognition accuracies of 100.00%, 82.61%, 87.18%, 100.00%, and 94.59% for air, dry sand, crushed stone, water, and saturated sand, respectively, and possesses the highest precision evaluation value. Traditional CNN models are inferior to the improved ResNet34 network in terms of recognition accuracy and generalization ability. This paper proposes a tunnel lining void identification method combining SWT with an enhanced ResNet34 deep learning network, which enhances the accuracy of tunnel lining void detection and reduces the subjectivity and complexity of manual interpretation. This method provides new technical support for tunnel lining quality inspection and safety evaluation.

The finite difference time domain (FDTD) method was used to solve the Maxwell’s equation to obtain the reverse time migration (RTM) of ground penetrating radar (GPR) signals, namely, the FDTD-RTM. In order to ensure that the correct numerical solution of iterative calculation was achieved, it was necessary to obtain high-resolution signals, which greatly limits the applicability of FDTD-RTM in engineering. Based on the characteristics of wavelet multi-resolution analysis, this study proposed a super-resolution signal reconstruction method to improve signal resolution, with the view to completely solving the problem of FDTD-RTM limitation caused by insufficient signal sampling using GPR. The results of electromagnetic simulation showed that the signals reconstructed by the above method were highly similar to the signals sampled with the same resolution. On this basis, reverse time migration electromagnetic simulation and physical model tests were designed. The results of both experiments showed that the under-sampled GPR signals could achieve FDTD-RTM following super-resolution reconstruction, and that the migration imaging results of the target were basically consistent with the design scheme. The signal super-resolution reconstruction method based on the wavelet theory was thus shown to successfully achieve the overall application of FDTD-RTM in GPR signal analysis.