An automatic extraction method of Rayleigh wave dispersion curve based on improved U-net neural network

Abstract

The present invention discloses a Rayleigh wave dispersion curve automatic extraction method based on an improved U-net neural network, including constructing a dispersion curve training set; constructing a dispersion curve test set; cutting the image data in the dispersion curve test set and the dispersion curve training set, and binarizing the theoretical dispersion curve in the dispersion curve training set; geometrically transforming the image data in the dispersion curve training set to obtain an expanded dispersion curve training set; constructing an improved U-net neural network model, and using the expanded dispersion curve training set for model training; and extracting multi-order Rayleigh wave dispersion curves from the dispersion curve test set using the trained improved U-net neural network model. The present invention makes full use of the image segmentation capability of deep learning, can quickly and accurately classify the dispersion energy map of field surface wave data, improves the efficiency of extracting Rayleigh wave dispersion curves, and further improves the efficiency of surface wave exploration.

Description

Technical Field

[0001] The invention relates to the technical field of shallow shear wave velocity structure imaging, and in particular to an automatic extraction method for Rayleigh wave dispersion curve based on an improved U-net neural network. Background Art

[0002] In the wave field generated by earthquake excitation, the energy of Rayleigh waves accounts for 67% of the total energy of the wave field. During data acquisition, observation data with a higher signal-to-noise ratio can be obtained. On the other hand, the Rayleigh wave data processing process is simple and direct, and the obtained shear wave velocity profile has a higher resolution. Although conventional multi-channel analysis of surface waves (MASW) has been widely used, its dispersion curve extraction requires manual extraction, and the accuracy and efficiency of the extraction depend on the experience and knowledge background of the interpreter. Therefore, manual extraction of dispersion curves is only suitable for experienced geophysical engineers to extract the dispersion curve of surface wave data. When faced with massive surface wave data, manual extraction of dispersion curves is inefficient and the accuracy of dispersion curves is greatly affected by human factors. There are large differences in the processing results of geophysical engineers with different experiences. Summary of the invention

[0003] In view of the above-mentioned deficiencies in the prior art, the present invention provides a method for automatically extracting Rayleigh wave dispersion curves based on an improved U-net neural network.

[0004] In order to achieve the above-mentioned object of the invention, the technical solution adopted by the present invention is:

[0005] A method for automatically extracting Rayleigh wave dispersion curves based on an improved U-net neural network comprises the following steps:

[0006] S1. Using surface wave field numerical simulation and surface wave dispersion energy imaging, obtain the dispersion energy map of the theoretical model, and divide it into training data and test data; calculate the theoretical dispersion curve of the theoretical model, and construct a dispersion curve training set in combination with the divided training data;

[0007] S2, collecting field seismic surface wave data, and performing dispersion energy imaging to obtain a dispersion energy map, and constructing a dispersion curve test set in combination with the divided test data;

[0008] S3, cutting the image data in the dispersion curve test set and the dispersion curve training set, and binarizing the theoretical dispersion curve in the dispersion curve training set;

[0009] S4, performing geometric transformation on the image data in the dispersion curve training set to obtain an expanded dispersion curve training set;

[0010] S5. Build an improved U-net neural network model and use the expanded dispersion curve training set to train the model;

[0011] S6. Use the trained improved U-net neural network model to extract multi-order Rayleigh wave dispersion curves from the dispersion curve test set.

[0012] Optionally, the dispersion curve test set includes a fundamental-order dispersion energy diagram, a noisy fundamental-order dispersion energy diagram, and a multi-order dispersion energy diagram.

[0013] Optionally, the dispersion curve training set includes a fundamental-order dispersion energy diagram, a noisy fundamental-order dispersion energy diagram, and a multi-order dispersion energy diagram.

[0014] Optionally, step S4 specifically includes:

[0015] The image data in the dispersion curve training set are subjected to multi-angle right-angle rotation transformation, horizontal mirror flip transformation and vertical mirror flip transformation respectively, and the image data after rotation transformation and flip transformation are merged to obtain the expanded dispersion curve training set.

[0016] Optionally, the improved U-net neural network model specifically includes:

[0017] A U-net neural network, a convolutional layer arranged at the end of the first number of encoding layers after the encoder of the U-net neural network, and a selection layer and a coordinate conversion layer arranged at the end of the decoder of the U-net neural network.

[0018] Optionally, the coordinate conversion layer performs coordinate conversion on the classification result of the decoder on the pixel points in the dispersion energy map, converts the position information of the dispersion points in the dispersion energy map into the frequency phase velocity domain, and then performs curve fitting to obtain the dispersion curve.

[0019] Optionally, the coordinate conversion layer performs coordinate conversion in the following manner:

[0020]

[0021]

[0022] Where (i, j) is the coordinate of the dispersion point in the dispersion energy diagram, V(i, j) is the phase velocity of the dispersion curve, F(i, j) is the frequency of the dispersion curve, V min 、V max is the maximum and minimum value of the phase velocity, F min 、F max are the maximum and minimum values ​​of the frequency.

[0023] The present invention has the following beneficial effects:

[0024] (1) Unlike the traditional manual extraction of dispersion curves, which requires experienced geologists to extract dispersion curves, the present invention automatically extracts Rayleigh wave multi-order dispersion curves through an improved U-net neural network. The neural network is used for image segmentation to divide the pixel points in the dispersion energy map into two categories: background and dispersion points. The coordinate conversion formula is then used to convert the position information of the dispersion points in the dispersion energy map into the frequency phase velocity domain. The dispersion curve can then be directly obtained through curve fitting. This method can automatically extract Rayleigh wave multi-order dispersion curves through a computer, with high extraction accuracy and strong practicality. The present invention can avoid the influence of human subjective factors on the accuracy of extracting dispersion curves. When faced with massive surface wave data, this method can quickly and accurately extract Rayleigh wave multi-order dispersion curves, and has broad application prospects in shallow velocity structure imaging. It can provide guidance for site selection for livelihood projects such as urban construction, large-scale water conservancy and hydropower facilities, railway and highway construction, save manpower and material resources and time spent, shorten the operation period, and avoid unnecessary economic losses;

[0025] (2) Different from the traditional manual extraction of dispersion curves and manual clicking of dispersion energy areas to achieve semi-automatic extraction of dispersion curves, the improved U-net neural network can automatically extract the dispersion curves of Rayleigh wave multi-order dispersion curves. This method makes full use of the image segmentation capability of deep learning and can quickly and accurately classify the dispersion energy maps of field surface wave data, thereby improving the efficiency of extracting Rayleigh wave dispersion curves and further improving the efficiency of surface wave exploration.

[0026] (3) The training set includes fundamental-order dispersion energy maps, noisy fundamental-order dispersion energy maps, and multi-order dispersion energy maps. During the training process, the improved U-net neural network learns the extraction rules of dispersion curves. The diversity of the training set improves the robustness of the improved U-net neural network and ensures the stability of the dispersion curve extraction process.

[0027] (4) This method uses the dispersion energy diagram and dispersion curve obtained during the forward simulation as the training set, which ensures the accuracy of the training set data and ultimately improves the accuracy of the improved U-net neural network in extracting the dispersion curve. At the same time, it can effectively shorten the time for obtaining the training set and improve the efficiency of extracting the dispersion curve.

[0028] (5) This method has a low requirement for the number of samples in the training set and can extract better dispersion curves with fewer training samples. As the number of training sets increases, the extraction effect becomes better and better. When the number of training sets reaches a certain number, the extraction effect is no longer significantly improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 It is a schematic diagram of a process of automatically extracting Rayleigh wave dispersion curves based on an improved U-net neural network in an embodiment of the present invention;

[0030] Figure 2 This is a schematic diagram of the original recording of surface wave data in an embodiment of the present invention;

[0031] Figure 3 A single shot data graph collected in the field in an embodiment of the present invention;

[0032] Figure 4 is a dispersion energy diagram in an embodiment of the present invention;

[0033] Figure 5 is a clipped dispersion energy diagram in an embodiment of the present invention;

[0034] Figure 6 A geometric expansion graph of training data in an embodiment of the present invention;

[0035] Figure 7 It is a geometric expansion diagram of label data in an embodiment of the present invention;

[0036] Figure 8 Schematic diagram of an improved U-net neural network in an embodiment of the present invention;

[0037] Fig. 9 is a dispersion curve extracted by using an improved U-net neural network in an embodiment of the present invention;

[0038] Fig.10 This is a dispersion curve picking result diagram corresponding to the improved U-net neural network when the number of training sets is different in the embodiment of the present invention;

[0039] Fig.11 It is a comparison diagram of the dispersion curve extracted by the improved U-net neural network in the embodiment of the present invention and the theoretical dispersion curve. DETAILED DESCRIPTION

[0040] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

[0041] In order to overcome the shortcomings of the traditional multi-channel surface wave analysis technology in manually extracting dispersion curves, the present invention proposes a method for extracting Rayleigh wave dispersion curves through machine learning artificial neural networks. Specifically, the improved U-net neural network is used to treat the dispersion curve extraction process as an image classification process. By learning the relationship between the input dispersion energy map of training data and the output dispersion curve map, the pixels in the dispersion energy map are divided into two categories: "dispersion curve and background". The trained U-net model can quickly and accurately extract the multi-order dispersion curves of Rayleigh waves from the dispersion energy map.

[0042] like Figure 1 As shown, the embodiment of the present invention provides a method for automatically extracting Rayleigh wave dispersion curves based on an improved U-net neural network, comprising the following steps S1 to S6:

[0043] S1. Using surface wave field numerical simulation and surface wave dispersion energy imaging, obtain the dispersion energy map of the theoretical model, and divide it into training data and test data; calculate the theoretical dispersion curve of the theoretical model, and construct a dispersion curve training set in combination with the divided training data;

[0044] In an optional embodiment of the present invention, the present invention obtains the seismic record of the theoretical model by using the wave equation to perform surface wave field forward simulation, such as Figure 2 As shown; then the corresponding dispersion energy diagram is obtained by the joint phase shift method and the τ-p transformation method, including the fundamental order dispersion energy diagram, the noisy fundamental order dispersion energy diagram and the multi-order dispersion energy diagram. In order to improve the extraction accuracy of the subsequent improved U-net neural network model, the present invention divides the obtained dispersion energy diagram into training data and test data, one part is used to train the improved U-net neural network model, and the other part is used to test the improved U-net neural network model.

[0045] The present invention adopts the Schwab-Knopoff algorithm to calculate the theoretical dispersion curve of the theoretical model as label data; finally, the dispersion curve training set is constructed by combining the training data and the label data.

[0046] S2, collecting field seismic surface wave data, and performing dispersion energy imaging to obtain a dispersion energy map, and constructing a dispersion curve test set in combination with the divided test data;

[0047] In an optional embodiment of the present invention, the present invention first collects field seismic surface wave data, and obtains a fundamental order dispersion energy map and a multi-order dispersion energy map through dispersion energy imaging. In order to verify the extraction accuracy of the subsequent improved U-net neural network model, the present invention combines the obtained dispersion energy map with the test data divided in step S1 to construct a dispersion curve test set for subsequent extraction of Rayleigh wave dispersion curves.

[0048] like Figure 3 As shown in the figure, it is the single shot data collected in the field, where (a) is the first shot data and (b) is the second shot data. Figure 4 As shown, it is a dispersion energy map obtained by dispersion energy imaging, where (a) is the dispersion energy map corresponding to the surface wave data of the first shot, and (b) is the dispersion energy map corresponding to the surface wave data of the second shot.

[0049] S3, cutting the image data in the dispersion curve test set and the dispersion curve training set, and binarizing the theoretical dispersion curve in the dispersion curve training set;

[0050] In an optional embodiment of the present invention, in order to increase the resolution of the output image, the present invention crops the image data in the dispersion curve test set and the dispersion curve training set, that is, the image data in the dispersion curve test set and the dispersion curve training set are cropped into 512×512×3 RGB images, wherein the 3 in the third dimension represents the channel number of the image, and each channel stores 0-255 red, green and blue values, respectively, such as Figure 5 shown.

[0051] In order to improve the classification efficiency and accuracy of the improved U-net neural network model, the present invention binarizes the theoretical dispersion curve in the dispersion curve training set, that is, binarizes the label data in the dispersion curve training set to obtain a 512×512 binary image, and the pixels in the binary image are divided into two categories: 0 and 1, where 1 represents the background and 0 represents the dispersion curve.

[0052] S4, performing geometric transformation on the image data in the dispersion curve training set to obtain an expanded dispersion curve training set;

[0053] In an optional embodiment of the present invention, since the workload of collecting field seismic surface wave data is large and time-consuming, in order to reduce the workload and time cost of generating training data and label data, the present invention performs a geometric transformation on the image data in the dispersion curve training set, thereby increasing the number of data samples and improving the robustness of the improved U-net neural network model, thereby improving the work efficiency of model training.

[0054] The present invention performs geometric transformation on the image data in the dispersion curve training set, specifically performing multi-angle right-angle rotation transformation, horizontal mirror flip transformation and vertical mirror flip transformation on the image data in the dispersion curve training set, and merging the image data after the rotation transformation and the flip transformation to obtain the expanded dispersion curve training set.

[0055] Specifically, the present invention expands the training set by right-angle rotation transformation (90°, 180°, 270° rotation), horizontal mirroring and vertical mirroring of the image to improve the robustness of the neural network. In this embodiment, the training set includes 100 pairs of original data, including base-order data, noisy base-order data and multi-order data. The original training set is expanded to 500 pairs through sample expansion techniques such as rotation (90°, 180°, 270°), horizontal mirroring and vertical mirroring. Figure 6 As shown in the figure, it is a geometric expansion diagram of the training data; (a) and (d) are the original images, (b) and (e) are the images after the original images are rotated 90° counterclockwise, and (c) and (f) are the images of the horizontal mirror images of the original images. Figure 7 As shown, it is a geometric expansion diagram of the label data; (a) and (d) are the original images, (b) and (e) are the images after the original images are rotated 90° counterclockwise, and (c) and (f) are the images of the horizontal mirror images of the original images.

[0056] S5. Build an improved U-net neural network model and use the expanded dispersion curve training set to train the model;

[0057] In an optional embodiment of the present invention, the present invention improves the traditional U-net neural network so that the output of the improved network is the accurate coordinate information of the dispersion point in the dispersion energy map. In order to extract deeper features of the image, the present invention adds a convolution layer to the last three layers of the encoder, and uses multiple convolution layers to extract deep image features, thereby improving the extraction effect of image features; and adds a selection layer and a coordinate conversion layer after the trained U-net neural network model, and uses the selection layer to select the probability matrix of the dispersion curve classification result from the probability matrix of the background classification result output by the U-net neural network model and the probability matrix of the dispersion curve classification result by setting the selection threshold, and then uses the coordinate conversion layer to convert the position information of the dispersion point in the dispersion energy map into the frequency phase velocity domain through the coordinate conversion formula according to the selected probability matrix of the dispersion curve classification result, and then the dispersion curve can be directly obtained by curve fitting, so that the output of the improved U-net neural network is directly the dispersion curve.

[0058] like Figure 8 As shown, the improved U-net neural network model constructed by the present invention specifically includes:

[0059] A U-net neural network, a convolutional layer arranged at the end of the first number of encoding layers after the encoder of the U-net neural network, and a selection layer and a coordinate conversion layer arranged at the end of the decoder of the U-net neural network.

[0060] The present invention uses an improved U-net neural network model for image segmentation to divide the pixel points in the dispersion energy map into two categories: background and dispersion points, and then uses a coordinate conversion layer to convert the position information of the dispersion points in the dispersion energy map into the frequency phase velocity domain, and then the dispersion curve can be directly obtained through curve fitting.

[0061] When the frequency range, phase velocity range and image size of the image are determined, the pixels on the image will correspond to the frequency phase velocity one by one. Assume that the origin of the coordinates is at the upper left corner of the image, F min 、F max and V min 、V max are the upper and lower limits of frequency and phase velocity, respectively. i and j are the horizontal and vertical coordinates of the image pixel. V(i, j) and F(i, j) are the phase velocity and frequency of the dispersion curve at point (i, j), respectively. The coordinate transformation can be calculated as follows:

[0062]

[0063]

[0064] S6. Use the trained improved U-net neural network model to extract multi-order Rayleigh wave dispersion curves from the dispersion curve test set.

[0065] like Fig. 9 As shown, the extracted dispersion curve and dispersion energy Figure 10 The results are consistent with each other, achieving a relatively ideal picking effect.

[0066] The test performance of the extraction method of the present invention is described below with reference to specific examples.

[0067] In this embodiment, the network training computing platform is a 64-bit Windows system, equipped with an NVIDIA GeForce GTX2080S graphics card. The U-net network is built based on tensorflow1.13.2 as the backend GPU-Keras deep learning framework. The inputs required for network training mainly include: training set, improved U-net network model, and training parameters.

[0068] During training, the dispersion curve energy map corresponding to the theoretical model synthetic seismic record is used as training data, and the dispersion curve map corresponding to the same model is used as label data. The initial learning rate is set to 0.0001, and the Batch-size is set to 2, that is, two pairs of images are input for training each time; all images are trained once, which is called an epoch. The epoch in this embodiment is set to 50. The number of training images and label images is 500 pairs, and the Batch-size is set to 2. The final number of iterations N can be calculated by the following formula:

[0069]

[0070] By using the training set to train the improved U-net neural network, a trained U-net neural network model can be obtained.

[0071] In this embodiment, when generating a data set, 500 pairs of data are obtained after sample expansion of 100 pairs of data as training sets, and another 10 pairs of data are used as test sets. The ratio of the test set to the training set is 50:1. After the training, the obtained U-net network model achieved good extraction accuracy during the test. In order to explore whether the prediction effect of the improved U-net neural network is related to the number of training set samples, the network is trained and tested using training sets of different sizes.

[0072] In this embodiment, 50 pairs, 100 pairs, 200 pairs, 500 pairs and 600 pairs of samples are used to form five training sets. Different U-net network models are obtained by training the improved U-net neural network with different training sets. The extraction results of different U-net network models are tested with the same test data. Fig.10 shown.

[0073] Depend on Fig.10 It can be seen that when the training set is 50 pairs, the trained U-net network model can extract the dispersion curve, but the extraction effect of the image edge is poor. When the number of training sets is 100 pairs and 200 pairs, the effective range of extracting the dispersion curve is significantly increased, and the position is more accurate, but the left and right ends of the image still have the problem of unsatisfactory extraction effect. When the training set is 500 pairs, the extraction effect on the left and right ends of the image is effectively improved, and the dispersion curve extraction effect is relatively ideal. When the training set is 600 pairs, the extraction effect of the dispersion curve in (f) is not significantly improved compared with Figure (e). It can be seen that the improved U-net neural network can extract better dispersion curves with fewer training samples. As the number of training sets increases, the extraction effect becomes better and better. When the number of training sets reaches a certain number, the extraction effect is no longer significantly improved.

[0074] Considering the extraction effect of the dispersion curve and the training time, this embodiment selects the U-net network model obtained when the number of training set samples is 500 pairs as the optimal network model.

[0075] This embodiment uses a trained U-net neural network model to extract dispersion curves of fundamental order, noisy fundamental order and multi-order dispersion energy diagrams, and compares them with theoretical dispersion curves calculated by the Schwab-Knopoff algorithm.

[0076] Input the fundamental order dispersion energy map, and use the trained U-net neural network model to extract the dispersion curve of the fundamental order dispersion energy map. Fig.11 (a) is a comparison of the dispersion curve extracted by the improved U-net neural network and the theoretical dispersion curve. The mean square error and maximum relative error between the two can be calculated by the following formula:

[0077]

[0078]

[0079] in and They respectively represent the phase velocity calculated theoretically at the i-th frequency and the phase velocity extracted by the improved U-net network, and m is the total number of frequency points involved in the comparison.

[0080] From the above formula, we can get: Fig.11 The mean square error between the theoretical dispersion curve and the dispersion curve extracted by the improved U-net network in (a) is 0.22, and the maximum relative error is 0.24%. The calculation results show that the improved U-net network can extract the dispersion curve more accurately.

[0081] Input noisy fundamental order dispersion energy map and multi-order dispersion energy map, Fig.11 (b) and (c) are comparisons of the dispersion curves extracted by the improved U-net neural network and the theoretical dispersion curves. As shown in (b), for the noisy fundamental-order dispersion energy map, the dispersion curve extracted by the improved U-net network is almost completely consistent with the theoretical dispersion curve, with a corresponding mean square error of 0.27 and a maximum relative error of 0.57%. As shown in (c), the multi-order dispersion curves extracted by the improved U-net network are basically consistent with the theoretical multi-order dispersion curves. Among them, the mean square error of the fundamental-order mode dispersion curve is 3.3, and the maximum relative error is 2.9%; the mean square error of the first-order dispersion curve is 4.0, and the maximum relative error is 0.9%; the mean square error of the second-order dispersion curve is 1.2, and the maximum relative error is 0.6%.

[0082] From the above, it can be seen that the improved U-net neural network used in this embodiment to extract Rayleigh wave multi-order dispersion curves has a good effect and is highly applicable.

[0083] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0084] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0085] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0086] The present invention uses specific embodiments to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

[0087] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific variations and combinations that do not deviate from the essence of the present invention based on the technical revelations disclosed by the present invention, and these variations and combinations are still within the protection scope of the present invention.

Claims

1. A method for automatically extracting Rayleigh wave dispersion curves based on an improved U-net neural network, characterized in that: The following steps are involved: S1. Using surface wave field numerical simulation and surface wave dispersion energy imaging, obtain the dispersion energy map of the theoretical model, and divide it into training data and test data; calculate the theoretical dispersion curve of the theoretical model, and construct a dispersion curve training set in combination with the divided training data; S2, collecting field seismic surface wave data, and performing dispersion energy imaging to obtain a dispersion energy map, and constructing a dispersion curve test set in combination with the divided test data; S3, cutting the image data in the dispersion curve test set and the dispersion curve training set, and binarizing the theoretical dispersion curve in the dispersion curve training set; S4, performing geometric transformation on the image data in the dispersion curve training set to obtain an expanded dispersion curve training set; S5, constructing an improved U-net neural network model, and using the expanded dispersion curve training set to perform model training; the improved U-net neural network model specifically includes a U-net neural network, a convolutional layer arranged at the end of the first number of encoding layers of the encoder of the U-net neural network, and a selection layer and a coordinate conversion layer arranged at the end of the decoder of the U-net neural network; S6. Use the trained improved U-net neural network model to extract multi-order Rayleigh wave dispersion curves from the dispersion curve test set.

2. The method for automatically extracting Rayleigh wave dispersion curves based on an improved U-net neural network according to claim 1, characterized in that: The dispersion curve test set includes a fundamental-order dispersion energy diagram, a noisy fundamental-order dispersion energy diagram, and a multi-order dispersion energy diagram.

3. The method for automatically extracting Rayleigh wave dispersion curves based on an improved U-net neural network according to claim 1, characterized in that: The dispersion curve training set includes a fundamental order dispersion energy diagram, a noisy fundamental order dispersion energy diagram and a multi-order dispersion energy diagram.

4. The method for automatically extracting Rayleigh wave dispersion curves based on an improved U-net neural network according to claim 1, characterized in that: Step S4 specifically includes: The image data in the dispersion curve training set are subjected to multi-angle right-angle rotation transformation, horizontal mirror flip transformation and vertical mirror flip transformation respectively, and the image data after rotation transformation and flip transformation are merged to obtain the expanded dispersion curve training set.

5. The method for automatically extracting Rayleigh wave dispersion curves based on an improved U-net neural network according to claim 1, characterized in that: The coordinate conversion layer performs coordinate conversion on the classification results of the decoder on the pixel points in the dispersion energy map, converts the position information of the dispersion points in the dispersion energy map into the frequency phase velocity domain, and then performs curve fitting to obtain the dispersion curve.

6. The method for automatically extracting Rayleigh wave dispersion curves based on an improved U-net neural network according to claim 5, characterized in that: The coordinate conversion layer performs coordinate conversion in the following manner: Where (i, j) is the coordinate of the dispersion point in the dispersion energy diagram, V(i, j) is the phase velocity of the dispersion curve, F(i, j) is the frequency of the dispersion curve, V min 、V max is the maximum and minimum value of the phase velocity, F min 、F max are the maximum and minimum values ​​of the frequency.

Images