Fresnel zone pickup model, training method and diffracted wave imaging method based on machine learning

Abstract

The invention provides a Fresnel zone pickup model, a training method and a diffracted wave imaging method based on machine learning. The training method comprises the following steps: acquiring seismic wave training data; a Fresnel band label is marked for the horizontal slice image set of the initial 3D dip angle gather, and a real label set is obtained; inputting the horizontal slice image set of the initial 3D dip angle gather and the real label set into a pre-constructed Fresnel zone pickup model for iterative training to obtain a prediction result of a Fresnel zone; during each training, updating parameters of a Fresnel zone pickup model based on a real label of the Fresnel zone and a loss value of a prediction result; and a preset iteration condition is reached. According to the diffracted wave imaging method, diffracted wave imaging is carried out by using a trained Fresnel band pickup model. According to the method, the neural network is introduced on the basis of the 3D dip angle gather to pick the Fresnel zone, so that the precision of the picked Fresnel zone is higher, and the precision and efficiency of diffracted wave imaging are improved.

Description

Technical Field

[0001] The present invention relates to the field of geological survey technology, and in particular to a Fresnel zone picking model, a training method and a diffraction wave imaging method based on machine learning. Background Art

[0002] In geological exploration, the diffraction wave field is generated by underground small-scale discontinuities, so the diffraction wave contains a lot of high-resolution information about underground micro-inhomogeneous geological bodies. Therefore, the use of diffraction wave imaging has a natural advantage in identifying small-scale changes in micro-structures in geological bodies, and can break through the resolution limit of the main frequency wavelength in reflection wave imaging, which is of great significance for improving the ability to identify abnormalities in tiny geological bodies in oil and gas exploration.

[0003] The key to diffraction wave imaging is how to separate the diffraction wave from the reflection wave field, or how to effectively suppress the reflection wave. Since the main energy of reflection wave imaging comes from the corresponding reflection Fresnel band, the reflection Fresnel band is related to the imaging depth, offset distance and formation dip, and shows different forms in different data domains, especially in the area where diffractors exist, it becomes more complicated. Therefore, picking the Fresnel band is a repetitive and very time-consuming task. The existing technology uses manual picking, which is inefficient and leads to low imaging efficiency. At the same time, the results of manual picking depend largely on the personal experience of the implementer, resulting in low picking accuracy, which in turn affects the imaging accuracy of the diffraction wave. Summary of the invention

[0004] In view of this, an object of the present invention is to provide a Fresnel zone picking model, a training method and a diffraction wave imaging method based on machine learning to improve the efficiency and accuracy of diffraction wave imaging.

[0005] In a first aspect, a Fresnel zone picking model is provided, including an encoder and a decoder; The encoder includes an input unit and a plurality of feature extraction units; each feature extraction unit includes a first convolution module, an activation function and a second convolution module; wherein the activation function is a Tanh activation function; the convolution kernel size of the second convolution module is 3*3 and the step size is 2; The decoder includes feature reconstruction units and output units, the number of which is the same as that of the feature extraction units, and each feature reconstruction unit includes a deconvolution module.

[0006] In a second aspect, a training method for a Fresnel zone picking model is provided, the method comprising: Acquire seismic wave training data of the target work area; the seismic wave training data includes a horizontal slice image set of the initial 3D dip gather corresponding to all imaging points on the target imaging line in the seismic wave imaging space; Annotate the Fresnel band labels for the horizontal slice image set of the initial 3D dip gather to obtain the true label set; The horizontal slice image set and the real label set of the initial 3D dip gather are input into the pre-built Fresnel zone picking model for iterative training to obtain the prediction results of the Fresnel zone; During each training, the parameters of the Fresnel band picking model are updated based on the true label of the Fresnel band and the loss value of the predicted result; until the preset iteration condition is reached.

[0007] Optionally, the Fresnel band labels are annotated for the horizontal slice image set of the initial 3D dip gather, and the obtained real label set includes: Pick up a Fresnel zone for each horizontal slice in the horizontal slice image set of the initial 3D dip gather; Cut off the picked Fresnel zone; The true label set is made based on the horizontal slice image set after cutting off the Fresnel zone.

[0008] Optionally, the Fresnel zone is elliptical on the horizontal slice; picking the Fresnel zone for each horizontal slice in the horizontal slice image set of the initial 3D dip gather comprises: For each horizontal slice in the horizontal slice image set of the initial 3D dip gather, 3D dip gather data corresponding to the horizontal slice is obtained; the 3D dip gather data includes the two-way travel time of the seismic wave, the x-direction dip angle and the y-direction dip angle corresponding to the seismic trace, and the main frequency of the seismic wave; The radius of the Fresnel zone in two directions is calculated based on the 3D dip gather data and the Fresnel zone picking calculation formula, and the range of the Fresnel zone is determined based on the radius in two directions and the stable phase point; wherein, the Fresnel zone picking calculation formula is:

[0009] in, For Fresnel Radius in direction; For Fresnel Radius in direction; is the round-trip travel time of seismic waves; is the main frequency of seismic waves; is the stable phase point of the 3D dip gather at the imaging point on the corresponding horizontal slice, where: They correspond to the inclination angles of the stratum where the imaging point is located along the x-direction and the y-direction respectively.

[0010] Optionally, during iterative training, each training process includes: An input unit based on the Fresnel zone picking model receives a horizontal slice image set and a true label set of an initial 3D dip gather; Extracting the Fresnel zone features of the horizontal slice image set of the initial 3D dip gather based on the first convolution module, and detecting the edge of the Fresnel zone in each horizontal slice based on the Fresnel zone features; wherein the Fresnel zone features at least include the shape, position, and the relative relationship between the energy strength inside and outside the Fresnel zone; Increase the complexity of the extracted Fresnel band features based on the Tanh activation function; The second convolution module based on the convolution kernel size of 3*3 and the stride of 2 reduces the spatial resolution of each slice in the slice image set; After multiple feature extractions, the spatial resolution of each horizontal slice in the slice image set is gradually restored through the deconvolution module, and the horizontal slices with the same spatial resolution in the feature extraction and feature reconstruction are fused; The prediction result of the finally detected Fresnel zone is outputted through the output unit.

[0011] Optionally, the expression for the output unit is:

[0012] Where pro represents the probability value of a single pixel in the output being classified as a Fresnel zone area, represented by a value from 0 to 1; m represents the parameter that controls the degree of energy attenuation outside the Fresnel zone. In a third aspect, a diffraction wave imaging method based on machine learning is provided, the method comprising: Acquire seismic wave detection data of the target work area, the seismic wave detection data includes a horizontal slice image set of 3D dip gathers corresponding to all imaging points on all imaging lines in the seismic wave imaging space; Input the horizontal slice image set of the 3D dip gather into the pre-trained Fresnel zone picking model to pick up the Fresnel zone on the horizontal slice corresponding to each imaging point; Cut out the picked Fresnel zone in the horizontal slice image set of 3D dip gather; The horizontal slice image set of the 3D dip gather after cutting off the Fresnel zone is superimposed to obtain the diffraction wave imaging profile.

[0013] In a fourth aspect, a diffraction wave imaging device based on machine learning is provided, the device comprising: An acquisition unit is used to acquire seismic wave detection data of the target work area, wherein the seismic wave detection data includes a horizontal slice image set of a 3D dip gather corresponding to all imaging points on all imaging lines in the seismic wave imaging space; A picking unit, used for inputting a horizontal slice image set of a 3D dip gather into a pre-trained Fresnel zone picking model to pick up a Fresnel zone on a horizontal slice corresponding to each imaging point; A cutting unit is used to cut out the picked Fresnel zone in the horizontal slice image set of the 3D dip gather; The stacking unit is used to stack the horizontal slice image set of the 3D dip gather after the Fresnel zone is cut off to obtain the diffraction wave imaging profile.

[0014] In a fifth aspect, an electronic device is provided, comprising a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus; Memory, used to store computer programs; The processor is used to implement the method steps described in any one of the second aspect or the third aspect when executing the program stored in the memory.

[0015] In a sixth aspect, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method steps described in any one of the second aspect or the third aspect are implemented.

[0016] The embodiment of the present invention provides a Fresnel zone picking model, a training method and a diffraction wave imaging method based on machine learning. The training is performed based on an improved Fresnel zone picking model. The training method obtains seismic wave training data; the horizontal slice image set of the initial 3D dip gather is labeled with Fresnel zone labels to obtain a real label set; the horizontal slice image set of the initial 3D dip gather and the real label set are input into a pre-built Fresnel zone picking model for iterative training to obtain the prediction result of the Fresnel zone; each time the training is performed, the parameters of the Fresnel zone picking model are updated based on the real label of the Fresnel zone and the loss value of the prediction result; until the preset iteration condition is reached. The trained Fresnel zone picking model is used to perform diffraction wave imaging. The present invention introduces a neural network to pick up Fresnel zones on the basis of 3D dip gathers, and the horizontal feature of the Fresnel zones is an ellipse. Compared with the traditional method of extracting Fresnel zone boundaries based on 2D dip gathers in mutually perpendicular X and Y directions to form a rectangle, this feature is more consistent with the situation of Fresnel zones in the dip domain, so that the accuracy of the picked Fresnel zones is higher, thereby improving the accuracy of diffraction wave imaging, and separating reflected waves and performing diffraction wave imaging through the trained Fresnel zone picking model is more efficient.

[0017] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0019] Figure 1 A schematic diagram of the network structure of a Fresnel zone picking model provided by an embodiment of the present invention is shown; Figure 2 A flow chart of a method for training a Fresnel zone picking model provided by an embodiment of the present invention is shown; Figure 3 A schematic diagram showing the comparison results before and after the Fresnel zone is cut off in a three-dimensional space provided by an embodiment of the present invention; Figure 4 A schematic diagram showing changes in each layer of the Fresnel zone picking model of the horizontal slice image set provided by an embodiment of the present invention is shown; Figure 5 A schematic diagram showing a comparison of output results before and after the improvement of the output unit provided in an embodiment of the present invention is shown; Figure 6 A schematic flow chart of a diffraction wave imaging method based on machine learning provided in an embodiment of the present invention is shown; Figure 7 A schematic diagram showing a comparison of a horizontal slice corresponding to an imaging point provided by an embodiment of the present invention before and after the Fresnel zone is cut off; Figure 8 A schematic diagram of a diffraction wave imaging section provided by an embodiment of the present invention is shown; Fig. 9 A schematic diagram of a reflection wave imaging section in the prior art is shown; Fig.10 A schematic structural diagram of a diffraction wave imaging device based on machine learning is shown; Fig.11 A schematic structural diagram of an electronic device provided by an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0020] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present invention.

[0021] The principle of using seismic waves for geological exploration is: first, one or more seismic sources and multiple detectors are set up above the ground, and the seismic waves generated by the multiple seismic sources are transmitted to the ground in sequence or simultaneously. After being reflected by the geological surfaces of each layer on the ground, the seismic waves are reflected back to the ground and received by all or some of the detectors. Pre-stack migration processing is performed based on parameters such as the amplitude and travel time of the detected reflected waves to obtain information on complex underground structures.

[0022] The existing technology mainly uses reflected waves for imaging, but the resolution limit of the reflected wave imaging method is one-fourth of the wavelength of the main frequency of the seismic wave. Even if various frequency domain protection or expansion methods are adopted, its resolution still cannot meet the needs of actual production.

[0023] Therefore, the embodiment of the present invention uses diffraction wave imaging, and the diffraction wave field is generated by the underground small-scale discontinuity. Therefore, the diffraction wave contains a large amount of high-resolution information about the underground micro-inhomogeneous geological body. Therefore, the use of diffraction waves has a natural advantage in identifying the small-scale changes of the above-mentioned micro-structures, and can break through the limitation of the main frequency wavelength on the resolution in the reflection wave imaging, which is of great significance for improving the ability to identify the abnormalities of tiny geological bodies in oil and gas exploration.

[0024] The key to diffraction wave imaging is to separate the diffraction wave from the reflected wave. The energy of the reflected wave is mainly concentrated in the Fresnel zone area. Therefore, by picking up the Fresnel zone, the area of ​​the reflected wave can be determined and then separated.

[0025] The existing technology uses manual picking of Fresnel zones, which is inefficient and leads to low imaging efficiency. At the same time, the manual picking results largely rely on the personal experience of the implementer, resulting in low picking accuracy, which in turn affects the imaging accuracy of the diffraction wave.

[0026] Based on this, an embodiment of the present invention provides a Fresnel zone picking model and a training method thereof, which are described below through embodiments.

[0027] The embodiment of the present invention provides a Fresnel zone picking model, such as Figure 1 As shown, it includes an encoder and a decoder; The encoder includes an input unit 101 and multiple feature extraction units 102; each feature extraction unit 102 includes a first convolution module, an activation function and a second convolution module; wherein the activation function is a Tanh activation function; the convolution kernel size of the second convolution module is 3*3 and the step size is 2.

[0028] The Fresnel band picking model of the embodiment of the present invention is an improvement based on the traditional U-net network structure; in the traditional U-net network, the activation function of the feature extraction unit adopts the LeakyReLU activation function, and this activation function is not applicable to seismic wave detection data, because there are some negative features in the seismic wave detection data, and the use of the LeakyReLU activation function will cause the negative features to disappear. Therefore, the embodiment of the present invention adopts the Tanh activation function. The Tanh function is continuous and differentiable over the entire real number domain, and maps the input to between (-1, 1). This means that it can better process and retain the negative information in the seismic wave detection data, and will not make the negative features disappear or become less significant like LeakyReLU. In addition, the Tanh function provides a smoother nonlinear transformation than LeakyReLU, which helps the model learn more complex patterns and features, especially when processing signal data with a rich dynamic range.

[0029] The feature extraction unit of the traditional U-net network structure uses maximum pooling to reduce the spatial resolution of the feature map, which will cause part of the information of the feature map to be lost. Therefore, the embodiment of the present invention uses a second convolution module with a convolution kernel size of 3*3 and a step size of 2 to replace the maximum pooling layer. Compared with the maximum pooling operation, the use of the convolution module for feature extraction can learn how to best compress spatial information instead of simply selecting the maximum value of each area, so that more original information can be retained in the feature map after dimensionality reduction, thereby improving the model performance.

[0030] The decoder includes feature reconstruction units 103 and output units 104, the number of which is the same as the number of feature extraction units, and each feature reconstruction unit 103 includes a deconvolution module.

[0031] The deconvolution module gradually restores the feature map to its original spatial resolution, and merges it with the feature map of the same spatial resolution in the feature reconstruction unit, and finally outputs it through the output unit. The output unit has also been improved, which will be described in the following embodiments and will not be repeated here.

[0032] With respect to the Fresnel zone pickup model constructed in the above embodiment, an embodiment of the present invention provides a training method for the Fresnel zone pickup model, such as Figure 2 As shown, the method comprises the following steps: Step S201: Acquire seismic wave training data of the target work area.

[0033] The seismic wave training data includes a horizontal slice image set of the initial 3D dip gather corresponding to all imaging points on the target imaging line in the seismic wave imaging space.

[0034] In an embodiment of the present invention, the seismic wave imaging space is a three-dimensional imaging space, which includes multiple imaging lines, each imaging line includes multiple CDPs (Common Depth Points), and each CDP contains a seismic trace, which includes multiple imaging points.

[0035] In this step, the target imaging lines are relatively typical imaging lines. For example, some imaging lines may pass through important geological structures, such as faults, fold axes, salt domes, etc. These areas are crucial for understanding the geological background of the entire exploration area. Therefore, these imaging lines can be determined as target imaging lines. Training with a small amount of dip gather data of typical imaging lines can improve training accuracy.

[0036] In this step, firstly, the 3D dip gather corresponding to each imaging point is obtained, and the dip gather is obtained after the initial pre-stack gather is offset; since the 3D dip gather is a three-dimensional data, and the Fresnel zone picking model processes a two-dimensional image, it is necessary to perform dimensionality reduction processing on the 3D dip gather to obtain horizontal slices.

[0037] In a specific example, the depth of any imaging point is , then the corresponding 3D dip gather result is: (1); in, is the shot point coordinate; is the coordinate of the imaging point; is the coordinate of the detection point; is the round-trip travel time of seismic waves; is the travel time of the seismic wave from the earthquake source to the imaging point; is the travel time from the imaging point to the detector; It means to obtain the first-order derivative of the seismic trace sequence. is the number of the seismic channel; is the number of seismic traces. for The inclination of the direction; for The inclination angle in the direction.

[0038] Among them, during the calculation of the dip gather of the imaging point, the dip corresponding to the seismic trace numbered j is Calculated by the following formula: (2); (3); In the formula, is the RMS velocity of the imaging point, and the other parameters refer to the meanings of the parameters in formula 1.

[0039] Finally, the dimensionality reduction process is performed on the above 3D dip gathers, that is, for a given depth The horizontal slice image set corresponding to each depth is obtained. Step S202: labeling the horizontal slice image set of the initial 3D dip gather with Fresnel band labels to obtain a true label set. If the horizontal slices of the entire imaging space are labeled, it will be a very large workload. Therefore, in order to reduce the workload and improve the picking efficiency, in a feasible implementation, the Fresnel band labels are labeled for the horizontal slice image set of the initial 3D dip gather, and the real label set includes: Step S2021: Pick a Fresnel zone for each horizontal slice in the horizontal slice image set of the initial 3D dip gather.

[0040] When labeling Fresnel zones in the training phase, the Fresnel zones are first picked up by human-computer interaction. Specifically, the 3D dip gathers are superimposed along the x-direction and y-direction, and the stable phase points are picked up by human-computer interaction on the superimposed sections in each direction. (i.e., the center point of the Fresnel zone), where They correspond to the stratum dip angles of the imaging point along the x-direction and y-direction. Then, the spatial range of the Fresnel zone is further picked using the picking result and the Fresnel zone radius calculation formula, and finally, the Fresnel zone label can be obtained by horizontally slicing it.

[0041] In the embodiment of the present invention, the Fresnel zone is elliptical on the horizontal slice; therefore, as long as the major axis radius, minor axis radius and center point of the Fresnel zone are calculated, the position and shape of the Fresnel zone on the horizontal slice can be determined.

[0042] In a feasible implementation, picking a Fresnel zone for each horizontal slice in a horizontal slice image set of an initial 3D dip gather includes: Step S2021A: for each horizontal slice in the horizontal slice image set of the initial 3D dip gather, obtain the 3D dip gather data corresponding to the horizontal slice; the 3D dip gather data includes the two-way travel time of the seismic wave, the x-direction dip angle and the y-direction dip angle corresponding to the seismic trace, and the main frequency of the seismic wave; Step S2021B: Calculate the radius of the Fresnel zone in two directions based on the 3D dip gather data and the Fresnel zone picking calculation formula, and determine the range of the Fresnel zone based on the radius in the two directions and the position of the stable phase point; wherein the Fresnel zone picking calculation formula is: (4); (5); in, For Fresnel Radius in direction; For Fresnel Radius in direction; is the round-trip travel time of seismic waves; is the main frequency of seismic waves; is the stable phase point of the 3D dip gather at the imaging point on the corresponding horizontal slice, where: They correspond to the inclination angles of the stratum where the imaging point is located along the x-direction and the y-direction respectively.

[0043] Step S2022: cutting off the picked Fresnel zone.

[0044] In this step, the excision is performed manually.

[0045] Step S2023: Generate a true label set based on the horizontal slice image set after the Fresnel zone is cut off.

[0046] In one example, for example, the area of ​​the cut-off Fresnel zone is labeled as 0, and the area outside the Fresnel zone is labeled as 1, so as to form a true label set.

[0047] Figure 3 2 shows a schematic diagram of the before and after comparison of the Fresnel zone area removed in three-dimensional space, i.e., in multiple continuous horizontal slices, with the left side showing before removal and the right side showing after removal.

[0048] Since the energy of the reflected wave is mainly in the Fresnel zone, the reflected wave is suppressed by cutting off the Fresnel zone, thus achieving the separation of the reflected wave and the diffracted wave.

[0049] Step S203: input the horizontal slice image set and the true label set of the initial 3D dip gather into the pre-built Fresnel zone picking model for iterative training to obtain the prediction result of the Fresnel zone.

[0050] Figure 4A schematic diagram of the changes of the horizontal slice image set of the initial 3D dip gather through each layer of the Fresnel zone picking model is given in FIG. 1 . In this step, the process of each training is described based on the above pre-built Fresnel zone picking model: Step S2031: The input unit based on the Fresnel zone picking model receives the horizontal slice image set and the real label set of the initial 3D dip gather.

[0051] Step S2032: extracting the Fresnel zone features of the horizontal slice image set of the initial 3D dip gather based on the first convolution module, and detecting the edge of the Fresnel zone in each horizontal slice based on the Fresnel zone features.

[0052] In one example, the convolution kernel size of the first convolution module is 3*3 and the step size is 1.

[0053] The Fresnel zone picking model of the embodiment of the present invention has multi-scale feature extraction capabilities, such as Figure 4 As shown in the figure, the shallow network mainly extracts local detail features, while the deep network with a larger receptive field extracts more abstract global features. During the network training process, it can not only fully obtain the macroscopic features of the Fresnel band, such as shape, position, and the relative relationship between the internal and external energy strengths; it can also identify the detailed morphology of the reflected waves, such as frequency, amplitude, and the morphology of the event axis. Therefore, the network not only has the ability to determine the position and shape of the Fresnel band, but also can effectively detect the remaining reflected wave energy outside the Fresnel band.

[0054] Step S2033: increasing the complexity of the extracted Fresnel zone features based on the Tanh activation function.

[0055] Step S2034: reducing the spatial resolution of each horizontal slice in the slice image set based on a second convolution module with a convolution kernel size of 3*3 and a step size of 2.

[0056] Step S2035: After multiple feature extractions, the spatial resolution of each horizontal slice in the slice image set is gradually restored through the deconvolution module, and the horizontal slices with the same spatial resolution in the feature extraction and feature reconstruction are fused.

[0057] In this step, more original information can be retained and richer detail features can be extracted through splicing and fusion.

[0058] Step S2036: Outputting the prediction result of the finally detected Fresnel zone through the output unit.

[0059] In the embodiment of the present invention, since the Fresnel model can effectively detect the remaining reflected wave energy outside the Fresnel band, the output unit is improved in order to fully utilize the network to suppress the remaining reflected waves.

[0060] The output unit of the traditional U-net network model is a simple binary classification output. This output is prone to high-frequency truncation noise, especially at the inner and outer edges of the Fresnel band, if it is only classified as 0 or 1, resulting in unclear edges.

[0061] The expression of the improved output unit is: (6); In the formula, Indicates the probability value of a single pixel in the output being classified as a Fresnel zone area, ranging from 0 to 1; Represents the parameter that controls the degree of energy attenuation outside the Fresnel zone.

[0062] This is a raw probability value (ranging from 0 to 1) into another form of probability value, where It is a control parameter used to adjust the degree of attenuation of energy outside the Fresnel zone.

[0063] This conversion method makes the transition near the inner and outer boundaries of the Fresnel zone smoother. The degree of this smoothing can be controlled to achieve better edge handling, which is important for improving the quality of model predictions, especially when edges or details require finer processing.

[0064] Specifically, the parameters Controls the speed at which the energy outside the Fresnel zone decays. When it is larger, the formula is The impact of the value is more significant, resulting in a sharper change in the output value, especially in close to 0 or 1. On the contrary, a smaller A higher value will cause the output value to change more gradually, thus achieving a smoothing effect at the boundaries and reducing the appearance of high-frequency truncation noise.

[0065] In some cases, appropriate selection A higher value can enhance the contrast between the inside and outside of the Fresnel zone, which helps improve the accuracy of the model in identifying the target area.

[0066] like Figure 5 As shown in the figure, it is a schematic diagram comparing the output results before and after the output unit is improved; the left side is before the improvement, and the right side is after the improvement. It can be clearly observed that the edge of the Fresnel zone after the improvement is smoother.

[0067] Step S204: During each training, the parameters of the Fresnel band picking model are updated based on the real label of the Fresnel band and the loss value of the prediction result; until the preset iteration condition is reached.

[0068] In this step, the loss function may be a cross entropy loss function. The preset iteration condition may be reaching a preset number of iterations, or the loss value reaching a preset threshold.

[0069] Based on the above embodiments, the present invention provides a diffraction wave imaging method based on machine learning, such as Figure 6 As shown, the method comprises the following steps: Step S601: Acquire seismic wave detection data of the target work area, wherein the seismic wave detection data includes a horizontal slice image set of a 3D dip gather corresponding to all imaging points on all imaging lines in the seismic wave imaging space.

[0070] It should be noted that the target work area is the same as the target work area during training.

[0071] Step S602: input the horizontal slice image set of the 3D dip gather into the pre-trained Fresnel zone picking model to pick the Fresnel zone on the horizontal slice corresponding to each imaging point.

[0072] In this step, the specific processing process of the model refers to the above example and will not be repeated here.

[0073] The model outputs the weight coefficient value used for smoothing the Fresnel band, that is, the probability distribution of the Fresnel band.

[0074] Step S603: cutting out the picked Fresnel zone in the horizontal slice image set of the 3D dip gather.

[0075] In one example, the excision method can multiply the output result of the Fresnel band picking model with the input to obtain a horizontal slice with the Fresnel band excised. Because the area of ​​the output Fresnel band is set to 0, after multiplication with the original horizontal slice, the corresponding area of ​​the original horizontal slice is also set to 0, thus achieving excision.

[0076] In this step, the Fresnel zone area can be cut out by multiplying the output result with the input horizontal slice. Figure 7 Shown is a schematic diagram of the comparison before and after horizontal slice resection, the left side is before resection, and the right side is after resection.

[0077] Step S604: superimpose the horizontal slice image set of the 3D dip gather after the Fresnel zone is cut off to obtain the diffraction wave imaging section.

[0078] In this step, if Figure 8 FIG. 1 is a schematic diagram of an imaging cross section of a diffraction wave obtained by the method of an embodiment of the present invention. In order to make the effect of the present invention more prominent, it is compared with a conventional imaging cross section of a reflection wave. Fig. 9Shown is the reflection wave imaging section. Figure 8 and Fig. 9 By comparing the actual data imaging profiles of the same imaging area, it can be seen that the diffraction wave event axes in the deep learning diffraction wave imaging results correspond well to the disconnection positions of the reflection wave event axes in the reflection wave imaging results, indicating that the machine learning diffraction wave imaging method based on 3D formation dip gathers proposed in the present invention can effectively suppress the reflection wave energy in seismic data and realize diffraction wave imaging; it can cope with the situation of complex underground structures and reflect the spatial position relationship of underground diffractors, and has higher accuracy than the reflection wave imaging method.

[0079] Based on the same inventive concept, a diffraction wave imaging device based on machine learning is provided, such as Fig.10 As shown, the device comprises: The acquisition unit 1001 is used to acquire seismic wave detection data of the target work area, wherein the seismic wave detection data includes a horizontal slice image set of a 3D dip gather corresponding to all imaging points on all imaging lines in the seismic wave imaging space. The picking unit 1002 is used to input the horizontal slice image set of the 3D dip gather into the pre-trained Fresnel zone picking model to pick the Fresnel zone on the horizontal slice corresponding to each imaging point.

[0080] The cutting unit 1003 is used to cut out the picked Fresnel zone in the horizontal slice image set of the 3D dip gather.

[0081] The stacking unit 1004 is used to stack the horizontal slice image set of the 3D dip gather after the Fresnel zone is cut off to obtain the diffraction wave imaging profile.

[0082] Based on the same technical concept, an embodiment of the present invention further provides an electronic device, such as Fig.11 As shown, it includes a processor 1101 , a communication interface 1102 , a memory 1103 and a communication bus 1104 , wherein the processor 1101 , the communication interface 1102 , and the memory 1103 communicate with each other via the communication bus 1104 .

[0083] Memory 1103, used for storing computer programs; The processor 1101 is used to implement the steps of the training method of the Fresnel zone picking model and the diffraction wave imaging method based on machine learning when executing the program stored in the memory 1103.

[0084] The communication bus mentioned in the above electronic device can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The communication bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0085] The communication interface is used for communication between the above electronic device and other devices.

[0086] The memory may include a random access memory (RAM) or a non-volatile memory (NVM), such as at least one disk memory. Optionally, the memory may also be at least one storage device located away from the aforementioned processor.

[0087] The above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0088] The computer program product of the Fresnel zone picking model training method and the diffraction wave imaging method based on machine learning provided in the embodiments of the present invention includes a computer-readable storage medium storing program code, and the instructions included in the program code can be used to execute the method described in the previous method embodiment. The specific implementation can be found in the method embodiment, which will not be repeated here.

[0089] The device of the diffraction wave imaging method based on machine learning provided in the embodiment of the present invention can be specific hardware on the device or software or firmware installed on the device. The device provided in the embodiment of the present invention has the same implementation principle and technical effects as the aforementioned method embodiment. For the sake of brief description, the parts not mentioned in the device embodiment can refer to the corresponding contents in the aforementioned method embodiment. It can be clearly understood by technicians in the relevant field that for the convenience and simplicity of description, the specific working processes of the systems, devices and units described above can all refer to the corresponding processes in the aforementioned method embodiment, and will not be repeated here.

[0090] In the embodiments provided by the present invention, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interfaces, and the indirect coupling or communication connection of devices or units can be electrical, mechanical or other forms.

[0091] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0092] In addition, each functional unit in the embodiment provided by the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0093] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program codes.

[0094] It should be noted that similar numbers and letters represent similar items in the following figures. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. In addition, the terms "first", "second", "third", etc. are only used to distinguish the description and are not to be understood as indicating or implying relative importance.

[0095] Finally, it should be noted that the above-described embodiments are only specific implementations of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them. The protection scope of the present invention is not limited thereto. Although the present invention is described in detail with reference to the above-mentioned embodiments, ordinary technicians in the field should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the above-mentioned embodiments within the technical scope disclosed by the present invention, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention. They should all be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A Fresnel zone picking model, characterized in that: Includes encoder and decoder; The encoder includes an input unit and a plurality of feature extraction units; each of the feature extraction units includes a first convolution module, an activation function and a second convolution module; wherein the activation function is a Tanh activation function; the convolution kernel size of the second convolution module is 3*3 and the step size is 2; The decoder includes feature reconstruction units and output units having the same number as the feature extraction units, and each of the feature reconstruction units includes a deconvolution module.

2. A training method for the Fresnel zone picking model according to claim 1, characterized in that: The method comprises: Acquire seismic wave training data of the target work area; the seismic wave training data includes a horizontal slice image set of an initial 3D dip gather corresponding to all imaging points on the target imaging line in the seismic wave imaging space; Annotating Fresnel band labels for a horizontal slice image set of the initial 3D dip gather to obtain a true label set; Inputting the horizontal slice image set and the true label set of the initial 3D dip gather into a pre-built Fresnel zone picking model for iterative training to obtain the prediction result of the Fresnel zone; During each training, the parameters of the Fresnel band picking model are updated based on the real label of the Fresnel band and the loss value of the prediction result; until the preset iteration condition is reached.

3. The method according to claim 2, characterized in that The Fresnel band labels are annotated on the horizontal slice image set of the initial 3D dip gather to obtain a real label set including: Picking a Fresnel zone for each horizontal slice in the horizontal slice image set of the initial 3D dip gather; Cut off the picked Fresnel zone; The true label set is made based on the horizontal slice image set after cutting off the Fresnel zone.

4. The method according to claim 2, characterized in that: The Fresnel zone is elliptical on the horizontal slice; and picking the Fresnel zone for each horizontal slice in the horizontal slice image set of the initial 3D dip gather comprises: For each horizontal slice in the horizontal slice image set of the initial 3D dip gather, 3D dip gather data corresponding to the horizontal slice is obtained; the 3D dip gather data includes the two-way travel time of the seismic wave, the x-direction dip angle and the y-direction dip angle corresponding to the seismic trace, and the main frequency of the seismic wave; The radii of the Fresnel zone in two directions are calculated based on the 3D dip gather data and the Fresnel zone picking calculation formula, and the range of the Fresnel zone is determined based on the radii in the two directions and the stable phase point; wherein the Fresnel zone picking calculation formula is: in, For Fresnel Radius in direction; For Fresnel Radius in direction; is the round-trip travel time of seismic waves; is the main frequency of seismic waves; is the stable phase point of the 3D dip gather at the imaging point on the corresponding horizontal slice, where: They correspond to the inclination angles of the stratum where the imaging point is located along the x-direction and the y-direction respectively.

5. The method according to claim 2, characterized in that: The process of each training in iterative training includes: An input unit based on the Fresnel zone picking model receives a horizontal slice image set and a true label set of the initial 3D dip gather; Extracting Fresnel zone features of the horizontal slice image set of the initial 3D dip gather based on the first convolution module, and detecting the edge of the Fresnel zone in each slice based on the Fresnel zone features; wherein the Fresnel zone features at least include shape, position, and relative relationship between internal and external energy strength of the Fresnel zone; Increasing the complexity of the extracted Fresnel band features based on the Tanh activation function; The second convolution module based on the convolution kernel size of 3*3 and the stride of 2 reduces the spatial resolution of each horizontal slice in the slice image set; After multiple feature extractions, the spatial resolution of each horizontal slice in the slice image set is gradually restored through the deconvolution module, and the horizontal slices with the same spatial resolution in feature extraction and feature reconstruction are fused; The prediction result of the finally detected Fresnel zone is outputted through the output unit.

6. The method according to claim 5, characterized in that The expression of the output unit is: Where pro represents the probability value of a single pixel in the output being classified as a Fresnel zone area, represented by a value from 0 to 1; m represents the parameter that controls the degree of energy attenuation outside the Fresnel zone.

7. A diffraction wave imaging method based on machine learning, characterized in that: The method comprises: Acquire seismic wave detection data of the target work area, wherein the seismic wave detection data includes a horizontal slice image set of a 3D dip gather corresponding to all imaging points on all imaging lines in the seismic wave imaging space; Inputting the horizontal slice image set of the 3D dip gather into a pre-trained Fresnel zone picking model to pick up the Fresnel zone on the horizontal slice corresponding to each imaging point; wherein the Fresnel zone picking model is trained by using any one of the methods described in claims 2-6; Cutting out the picked Fresnel zone in the horizontal slice image set of the 3D dip gather; The horizontal slice image set of the 3D dip gather after cutting off the Fresnel zone is superimposed to obtain the diffraction wave imaging profile.

8. A diffraction wave imaging device based on machine learning, characterized in that: The device comprises: An acquisition unit, used for acquiring seismic wave detection data of a target work area, wherein the seismic wave detection data includes a horizontal slice image set of a 3D dip gather corresponding to all imaging points on all imaging lines in a seismic wave imaging space; A picking unit, used for inputting the horizontal slice image set of the 3D dip gather into a pre-trained Fresnel zone picking model to pick up the Fresnel zone on the horizontal slice corresponding to each imaging point; wherein the Fresnel zone picking model is trained by the method described in any one of claims 2 to 6; A cutting unit, used for cutting out the picked Fresnel zone in the horizontal slice image set of the 3D dip gather; The stacking unit is used to stack the horizontal slice image set of the 3D dip gather after the Fresnel zone is cut off to obtain the diffraction wave imaging profile.

9. An electronic device, characterized in that: It includes a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus; Memory, used to store computer programs; A processor, for implementing the steps of the Fresnel zone picking model training method described in any one of claims 2 to 6 and the diffraction wave imaging method based on machine learning described in claim 7 when executing the program stored in the memory.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by the processor, the steps of the training method of the Fresnel zone picking model described in any one of claims 2 to 6 and the diffraction wave imaging method based on machine learning described in claim 7 are implemented.

Images