Least-squares reverse time migration using convolutional neural networks

Least-squares reverse time migration (LSRTM) has the potential to reconstruct a high-resolution image of subsurface reflectivity. However, the current data-domain LSRTM approach, which iteratively updates the subsurface reflectivity by minimizing the data residuals, is a computationally expensive task. To alleviate this problem and improve imaging quality, we develop a LSRTM approach using convolutional neural networks (CNNs), which is referred to as CNN-LSRTM. Specifically, the LSRTM problem can be implemented via a gradient-like iterative scheme, in which the updating component in each iteration is learned via a CNN model. In order to make the most of observation data and migration velocity model at hand, we utilize the common-source RTM image, the stacked RTM image, and the migration velocity model rather than only the stacked RTM image as the input data of CNN. We have successfully trained the constructed CNN model on the training data sets with a total of 5000 randomly layered and fault models. Based on the well-trained CNN model, we have proved that the proposed approach can efficiently recover the high-resolution reflection image for the layered, fault, and overthrust models. Through a marine field data experiment, it can determine the benefit of our constructed CNN model in terms of computational efficiency. In addition, we analyze the influence of input data of the constructed CNN model on the reconstruction quality of the reflection image.

Numerical Simulation Research on Response Characteristics of Grouting Defects of Ground Penetrating Radar for Detection of Grouting Quality behind Tunnel Wall

The effect of grouting behind tunnel wall directly affects the surrounding ground settlement and the stability of tunnel structure, so the grouting quality detection is very necessary. As an efficient and convenient shallow geophysical exploration method, ground-penetrating radar can meet the high-resolution and non-destructive requirements of grouting quality detection behind the tunnel wall, so it is widely used in engineering in recent years. Most of the existing studies have obvious regional pertinence and special geological conditions, and there are few universal studies on the characteristics of the ground penetrating radar reflection image of the grouting defect behind the tunnel wall. In view of this, this paper uses the finite difference time domain method to simulate several grouting defects behind the wall, such as voids, water-bearing anomaly, cracks, and other grouting defects. The simulation results show that the reflection image of the direct wave is characterized by a white band with strong amplitude; the interface between primary support and second lining, primary support, and surrounding rock is also banded; the circular cavity and water anomaly characteristics are all hyperbolic, the difference is that the phase of the lower part of the radar image of the cavity anomaly is 0, and there are only hyperbolic tails on both sides, and the water-bearing anomaly also has obvious hyperbolic characteristics at each interface; the reflected wave characteristics of the rectangular crack are striped and watery and the reflected wave characteristic of rectangular cracks is striped, and the abnormal range of water-bearing cracks on the radar image is larger than that of air. The research results can provide an effective theoretical reference for the engineering application of ground penetrating radar detection of grouting defects behind the tunnel wall.

Least-squares imaging of diffractions by solving a hybrid L1-L2 norm minimization problem

Traditional diffraction images without a specific migration kernel for promoting focusing abilities may cause confusion to seismic interpretation because diffraction images may show a finite-array response of diffracted/scattered waves. Because diffractors are discontinuous and sparsely distributed, a least-squares diffraction-imaging method is formulated by solving a hybrid L1-L2 norm minimization problem that imposes a sparsity constraint on diffraction images. It uses two different forward modeling operators for reflections and diffractions and L2 and L1 regularizations for penalizing the amplitudes of the reflection and diffraction images, respectively. A classic Kirchhoff diffraction demigration operator is implemented on an initial diffraction image model to synthesize diffracted/scattered waves. A Kirchhoff reflection demigration operator, formulated by considering the local reflection slopes and a cosine attenuation weighting function, is implemented on an initial reflection image to synthesize the reflected waves. A modified alternating direction approach of multipliers is developed for iteratively solving this minimization problem to create diffraction images and their separated diffractions. The depths and local reflection slopes of the reflection images are fixed during this iteration. To alleviate the energy leakage between diffractions and reflections, after performing the plane-wave destruction method on the conventional migration data, its estimated reflection image and residual image are provided as the initial reflection and diffraction images, respectively. Our method can remove steep-slope reflections, increase the focusing power of the diffractions, and eliminate noise. Two numerical experiments demonstrate its capability of separating and imaging small-scale discontinuities and inhomogeneities. The exposed geologic structures in the tunnel of field coal mining further illustrate this method’s potential in ascertaining hidden faults, edges, and collapsed columns. A safety warning should be definitely required if a mining working surface is advancing these hidden geologic disasters because an emergency of water bursting or gas leakage may happen.

Locating passive sources on seismic reflection image in the presence of velocity errors

Locating passive-source positions by using microseismic events is essential for monitoring hydraulic fractures. Among all microseismic source locating approaches, time-reversal imaging (TRI) is a promising one that is based on the principle that all of the back-propagated receiver wavefields should coincide at the source position when the velocity is accurate. It can image microseismic sources by applying a focusing imaging condition to the reconstructed receiver wavefields. However, the TRI method is highly sensitive to velocity errors and it is time-consuming or even challenging to refine the velocity model when the subsurface structure is complex. Instead of updating the velocity model, we have adopted a new method to locate microseismic events on a seismic reflection image under the condition that the seismic data are colocated with surface or near-surface microseismic observations. This method does not correctly place the microseismic events in depth; rather, it makes them consistent with the seismic reflection image; thus, it is still capable of providing correct local structure information around the passive source without the need of building an accurate velocity model. We have theoretically analyzed the variation of the imaged source location when the velocity model is inaccurate. Our result shows that faster velocities cause shallower depths whereas slower velocities cause deeper depths of source locations on the image. This is similar to the behavior of the focusing depth, but it is opposite to that of the migration depth in active-source seismic reflection imaging. Then, we match the locations of microseismic events with the well-focused seismic reflection image, which is extracted by slicing the time-shift common-image gathers at the time lag where the image has the maximal energy. Finally, synthetic tests validate the effectiveness of the proposed approach.

Color Image Generation from LiDAR Reflection Data by Using Selected Connection UNET

In this paper, a modified encoder-decoder structured fully convolutional network (ED-FCN) is proposed to generate the camera-like color image from the light detection and ranging (LiDAR) reflection image. Previously, we showed the possibility to generate a color image from a heterogeneous source using the asymmetric ED-FCN. In addition, modified ED-FCNs, i.e., UNET and selected connection UNET (SC-UNET), have been successfully applied to the biomedical image segmentation and concealed-object detection for military purposes, respectively. In this paper, we apply the SC-UNET to generate a color image from a heterogeneous image. Various connections between encoder and decoder are analyzed. The LiDAR reflection image has only 5.28% valid values, i.e., its data are extremely sparse. The severe sparseness of the reflection image limits the generation performance when the UNET is applied directly to this heterogeneous image generation. In this paper, we present a methodology of network connection in SC-UNET that considers the sparseness of each level in the encoder network and the similarity between the same levels of encoder and decoder networks. The simulation results show that the proposed SC-UNET with the connection between encoder and decoder at two lowest levels yields improvements of 3.87 dB and 0.17 in peak signal-to-noise ratio and structural similarity, respectively, over the conventional asymmetric ED-FCN. The methodology presented in this paper would be a powerful tool for generating data from heterogeneous sources.

Crosshole reflection imaging with ground-penetrating radar data: Applications in near-surface sedimentary settings

Crosshole ground-penetrating radar (GPR) is applied in areas that require a very detailed subsurface characterization. Analysis of such data typically relies on tomographic inversion approaches providing an image of subsurface parameters. We have developed an approach for processing the reflected energy in crosshole GPR data and applied it on GPR data acquired in different sedimentary settings. Our approach includes muting of the first arrivals, separating the up- and the downgoing wavefield components, and backpropagating the reflected energy by a generalized Kirchhoff migration scheme. We obtain a reflection image that contains information on the location of electromagnetic property contrasts, thus outlining subsurface architecture in the interborehole plane. In combination with velocity models derived from different tomographic approaches, these images allow for a more detailed interpretation of subsurface structures without the need to acquire additional field data. In particular, a combined interpretation of the reflection image and the tomographic velocity model improves the ability to locate layer boundaries and to distinguish different subsurface units. To support our interpretations of our field data examples, we compare our crosshole reflection results with independent information, including borehole logs and surface GPR data.