Amplitude spectrum driving-based artificial intelligence high-resolution processing method and system

Abstract

The embodiment of the invention provides an artificial intelligence high-resolution processing method and system based on amplitude spectrum driving, and belongs to the technical field of seismic exploration. The method comprises the following steps: transforming seismic data into a frequency domain, and determining an effective frequency band range corresponding to the seismic data based on a transformed frequency domain result; constructing an amplitude spectrum training sample set; training an amplitude spectrum driven artificial intelligence model based on the amplitude spectrum training sample set, and processing an amplitude spectrum corresponding to the to-be-processed seismic data by using the trained artificial intelligence model to obtain a seismic amplitude spectrum; and based on the effective frequency band range, reconstructing the seismic amplitude spectrum to obtain corresponding seismic data. According to the scheme of the invention, on the premise of keeping the seismic data phase information unchanged, the amplitude spectrum in the effective frequency band is subjected to targeted optimization, so that the seismic section is improved in the aspects of detail discernibility and frequency band consistency.

Description

Technical Field

[0001] This invention relates to the field of seismic exploration technology, and more specifically to an amplitude spectrum-driven artificial intelligence high-resolution processing method and an amplitude spectrum-driven artificial intelligence high-resolution processing system. Background Technology

[0002] High-resolution seismic data processing is a crucial step in the seismic exploration process, directly impacting the accuracy of seismic interpretation and the reliability of subsequent geological understanding and resource development decisions. As seismic exploration targets deeper and more complex tectonic regions, the requirements for seismic data resolution and signal-to-noise ratio are constantly increasing, making high-resolution seismic data processing technology an important research direction in the field of geophysical exploration.

[0003] Existing high-resolution seismic data processing methods can be mainly divided into two categories: conventional processing methods and deep learning-based processing methods.

[0004] Among conventional processing methods, inverse Q-filtering and deconvolution are widely used techniques. Inverse Q-filtering is based on the physical mechanism of seismic wave energy attenuation with frequency during propagation through strata. It improves resolution by inverting the Q-value of the stratum quality factor and performing energy compensation. However, this method is highly sensitive to the accuracy of Q-value inversion. In cases of complex strata and strong noise interference, the Q-value inversion error is significant, easily leading to compensation distortion and limiting its applicability. Deconvolution mainly relies on mathematical algorithms to compress or truncate the wavelet components in the seismic signal. Its effectiveness depends on a high signal-to-noise ratio, stable wavelet characteristics, and a clear reflection coefficient distribution. In low signal-to-noise ratio or complex seismic data processing, it is prone to noise amplification and unstable resolution improvement.

[0005] In recent years, with the development of deep learning technology, high-resolution seismic data processing methods based on neural networks have gradually become a research hotspot. Related methods include various technical routes such as generative adversarial networks, temporal neural networks, convolutional neural networks and their improved structures, and diffusion probability models. Compared with conventional methods, deep learning methods can achieve higher resolution and signal-to-noise ratio in some scenarios. However, in practical engineering applications, these methods still face significant technical bottlenecks: on the one hand, due to limitations in acquisition conditions and stratigraphic characteristics, seismic data generally suffers from limited effective bandwidth or dynamic range, and existing deep learning methods have not established effective constraint mechanisms to address this problem, making it difficult to overcome bandwidth limitations; on the other hand, the high cost and limited quantity of high-quality training samples lead to overfitting of models, insufficient generalization ability, and unstable processing results under different geological conditions and noise levels.

[0006] Therefore, how to achieve both high signal-to-noise ratio and high resolution in seismic data processing under conditions of limited effective bandwidth and limited training samples remains a key technical problem that needs to be solved by existing technologies. Summary of the Invention

[0007] The purpose of this invention is to provide an artificial intelligence high-resolution processing method and system based on amplitude spectrum driving, so as to at least solve the problem that the high-resolution processing effect of deep learning is limited under the conditions of limited effective frequency band and limited training samples of existing seismic data.

[0008] To achieve the above objectives, a first aspect of the present invention provides an artificial intelligence high-resolution processing method based on amplitude spectrum driving, the method comprising: acquiring seismic data to be processed; transforming the seismic data to the frequency domain; and determining the effective frequency band range corresponding to the seismic data based on the transformed frequency domain result; constructing an amplitude spectrum training sample set based on the effective frequency band range, the amplitude spectrum training sample set including low-resolution seismic amplitude spectrum training samples generated within the same effective frequency band range and corresponding high-resolution seismic amplitude spectrum training samples; training an amplitude spectrum-driven artificial intelligence model based on the amplitude spectrum training sample set; and using the trained artificial intelligence model to process the amplitude spectrum corresponding to the seismic data to be processed to obtain a seismic amplitude spectrum; and reconstructing the seismic amplitude spectrum based on the effective frequency band range to obtain the corresponding seismic data.

[0009] Optionally, acquiring seismic data to be processed, transforming the seismic data to the frequency domain, and determining the effective frequency band range corresponding to the seismic data based on the transformed frequency domain result includes: acquiring seismic time-domain data to be processed, performing frequency domain transformation on the seismic time-domain data to obtain the corresponding frequency domain representation result; extracting the amplitude spectrum of the seismic data based on the frequency domain representation result; performing continuity judgment on the extracted amplitude spectrum of the seismic data, identifying continuous frequency intervals that satisfy preset energy determination rules as candidate effective frequency intervals; performing continuity judgment on the candidate effective frequency intervals, and determining the effective frequency band range corresponding to the seismic data based on the judgment result.

[0010] Optionally, frequency energy distribution analysis is performed on the amplitude spectrum to identify continuous frequency intervals that meet preset energy determination rules as candidate effective frequency intervals. This includes: obtaining the amplitude energy distribution corresponding to each frequency position in the amplitude spectrum; determining the amplitude energy distribution based on the amplitude spectrum dynamic range parameter of the amplitude spectrum to identify frequency points where the amplitude energy is located within the amplitude spectrum dynamic range; merging adjacent frequency points that meet the amplitude spectrum dynamic range determination conditions into a continuous frequency interval in the amplitude spectrum; and determining the continuous frequency interval as the candidate effective frequency interval.

[0011] Optionally, based on the effective frequency band range, an amplitude spectrum training sample set is constructed, including: within the effective frequency band range, generating a corresponding low-resolution seismic data amplitude spectrum using a low-resolution seismic wavelet and using it as a low-resolution seismic amplitude spectrum training sample; within the same effective frequency band range as the low-resolution seismic amplitude spectrum training sample, generating a corresponding high-resolution seismic data amplitude spectrum using a high-resolution seismic wavelet and using it as a high-resolution seismic amplitude spectrum training sample; outside the effective frequency band range, maintaining the background noise energy distribution of the low-resolution seismic amplitude spectrum training sample and the high-resolution seismic amplitude spectrum training sample consistent; and combining the low-resolution seismic amplitude spectrum training sample and the high-resolution seismic amplitude spectrum training sample to form the amplitude spectrum training sample set.

[0012] Optionally, within the same effective frequency band as the low-resolution seismic amplitude spectrum training sample, a corresponding high-resolution seismic data amplitude spectrum is generated using a high-resolution seismic wavelet, and used as a high-resolution seismic amplitude spectrum training sample. This includes: within the effective frequency band, filtering the amplitude energy at corresponding frequency positions in the low-resolution seismic wavelet amplitude spectrum based on a preset amplitude spectrum threshold parameter; performing weighted enhancement processing on frequency components that meet the amplitude spectrum threshold parameter conditions to increase the amplitude energy of each frequency component within the effective frequency band; outside the effective frequency band, keeping the amplitude energy of corresponding frequency components in the seismic wavelet amplitude spectrum unchanged; and generating a corresponding high-resolution seismic amplitude spectrum based on the enhanced amplitude energy distribution, which is then used as the high-resolution seismic amplitude spectrum training sample.

[0013] Optionally, training the amplitude spectrum-driven artificial intelligence model based on the amplitude spectrum training sample set includes: using low-resolution seismic amplitude spectrum training samples from the amplitude spectrum training sample set as model input data, and using the corresponding high-resolution seismic amplitude spectrum training samples as model training labels; constructing an amplitude spectrum mapping relationship based on the input data and the training labels, and updating the parameters of the amplitude spectrum-driven artificial intelligence model; repeating the parameter update process of the amplitude spectrum mapping relationship within a preset training round to obtain a trained amplitude spectrum-driven artificial intelligence model.

[0014] Optionally, based on the input data and the training labels, an amplitude spectrum mapping relationship is constructed, and the parameters of the amplitude spectrum-driven artificial intelligence model are updated, including: inputting the low-resolution seismic amplitude spectrum corresponding to the input data into the amplitude spectrum-driven artificial intelligence model to obtain the corresponding predicted amplitude spectrum output result; comparing the predicted amplitude spectrum output result with the high-resolution seismic amplitude spectrum corresponding to the training labels to obtain an amplitude spectrum deviation result used to characterize the difference between the two; and performing an update operation on the trainable parameters in the amplitude spectrum-driven artificial intelligence model based on the amplitude spectrum deviation result to gradually correct the amplitude spectrum mapping relationship.

[0015] Optionally, based on the effective frequency band range, the seismic amplitude spectrum is reconstructed to obtain corresponding seismic data, including: within the effective frequency band range, determining an amplitude compensation relationship to describe amplitude changes based on the high-resolution seismic amplitude spectrum; outside the effective frequency band range, keeping the amplitude of the corresponding frequency component in the original seismic amplitude spectrum unchanged; applying the amplitude compensation relationship to the seismic amplitude spectrum to obtain a target seismic amplitude spectrum with amplitude update completed within the effective frequency band range; and performing an inverse frequency domain transform on the target seismic amplitude spectrum based on the phase information corresponding to the original seismic data to obtain the corresponding seismic data.

[0016] A second aspect of the present invention provides an amplitude spectrum-driven artificial intelligence high-resolution processing system, the system comprising: an acquisition unit for acquiring seismic data to be processed, transforming the seismic data to the frequency domain, and determining the effective frequency band range corresponding to the seismic data based on the transformed frequency domain result; a sample construction unit for constructing an amplitude spectrum training sample set based on the effective frequency band range, the amplitude spectrum training sample set including low-resolution seismic amplitude spectrum training samples generated within the same effective frequency band range and corresponding high-resolution seismic amplitude spectrum training samples; a training unit for training an amplitude spectrum-driven artificial intelligence model based on the amplitude spectrum training sample set, and using the trained artificial intelligence model to process the amplitude spectrum corresponding to the seismic data to be processed to obtain a seismic amplitude spectrum; and a reconstruction unit for reconstructing the seismic amplitude spectrum based on the effective frequency band range to obtain the corresponding seismic data.

[0017] On the other hand, the present invention provides a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the above-described amplitude spectrum-driven artificial intelligence high-resolution processing method.

[0018] Through the above technical solution, this invention, when processing seismic data at high resolution, first clarifies the effective frequency band range of the seismic data in the frequency domain. This effective frequency band range serves as a unified constraint throughout data construction, model training, and result reconstruction, ensuring that subsequent processing remains confined to the frequency range carrying the true physical information and avoiding blind enhancement of ineffective frequency bands. Based on this, by constructing a training sample set with a one-to-one correspondence between low-resolution and high-resolution amplitude spectra within the same effective frequency band, the mapping relationship learned by the artificial intelligence model reflects only the amplitude variation characteristics within the effective frequency band, reducing dependence on large-scale, high-quality samples and alleviating the training instability problem under limited sample conditions. Furthermore, the trained model is used to process the amplitude spectrum of the seismic data to be processed, and amplitude spectrum reconstruction is completed within the effective frequency band. This achieves an improvement in seismic data resolution while maintaining the phase information of the original seismic data. Therefore, in engineering scenarios where the effective frequency band of seismic data is limited, this invention can obtain high-resolution seismic processing results with stronger physical consistency and higher stability, improving the applicability of deep learning methods in complex seismic data.

[0019] Other features and advantages of the embodiments of the present invention will be described in detail in the following detailed description section. Attached Figure Description

[0020] The accompanying drawings are provided to further illustrate embodiments of the present invention and form part of the specification. They are used together with the following detailed description to explain the embodiments of the present invention, but do not constitute a limitation thereof. In the drawings:

[0021] Figure 1 This is a flowchart of the steps of an amplitude spectrum-driven artificial intelligence high-resolution processing method for a photovoltaic power grid system provided by one embodiment of the present invention;

[0022] Figure 2 This is a diagram of an amplitude spectrum-driven deep learning high-resolution seismic processing network structure provided by one embodiment of the present invention.

[0023] Figure 3 This is a flowchart illustrating the effective frequency band-driven amplitude spectrum optimization of seismic data provided by one embodiment of the present invention;

[0024] Figure 4 This is a schematic diagram of a low-resolution seismic data profile in the field provided by one embodiment of the present invention;

[0025] Figure 5 This is a comparison chart of the results of different high-resolution seismic processing methods provided in one embodiment of the present invention;

[0026] Figure 6This is an execution flowchart of an artificial intelligence high-resolution processing method based on amplitude spectrum driving provided by one embodiment of the present invention;

[0027] Figure 7 This is a system structure diagram of an amplitude spectrum-driven artificial intelligence high-resolution processing system provided in one embodiment of the present invention. Detailed Implementation

[0028] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0029] like Figure 1 As shown, embodiments of the present invention provide an artificial intelligence high-resolution processing method based on amplitude spectrum driving, the method comprising:

[0030] Step S1: Obtain the seismic data to be processed, transform the seismic data to the frequency domain, and determine the effective frequency band range corresponding to the seismic data based on the transformed frequency domain result.

[0031] Specifically, the process involves acquiring the seismic time-domain data to be processed, performing a frequency domain transformation on the seismic time-domain data to obtain the corresponding frequency domain representation result, extracting the amplitude spectrum of the seismic data based on the frequency domain representation result, performing frequency energy distribution analysis on the amplitude spectrum, identifying continuous frequency intervals that meet preset energy judgment rules as candidate effective frequency intervals, judging the continuity of the candidate effective frequency intervals, and determining the effective frequency band range corresponding to the seismic data based on the judgment result.

[0032] Furthermore, frequency energy distribution analysis is performed on the amplitude spectrum to identify continuous frequency intervals that meet preset energy determination rules as candidate effective frequency intervals. This includes: obtaining the amplitude energy distribution corresponding to each frequency position in the amplitude spectrum; determining the amplitude energy distribution based on the amplitude spectrum dynamic range parameter of the amplitude spectrum to identify frequency points where the amplitude energy is located within the amplitude spectrum dynamic range; merging adjacent frequency points that meet the amplitude spectrum dynamic range determination conditions into a continuous frequency interval in the amplitude spectrum; and determining the continuous frequency interval as the candidate effective frequency interval.

[0033] In this embodiment of the invention, the effectiveness of high-resolution processing depends on which frequency components in the frequency domain truly carry valid information. Therefore, before any subsequent amplitude spectrum modeling or learning, the effective frequency band range of the seismic data needs to be clearly defined. This step does not pursue complex spectral reconstruction, but emphasizes the objective identification of the frequency characteristics of existing data, providing a unified frequency reference for subsequent processing.

[0034] In the specific execution process, the seismic time-domain data to be processed is first acquired. This seismic time-domain data can be obtained from a conventional seismic acquisition process, without any additional restrictions on the acquisition method. Subsequently, a frequency domain transformation is performed on the seismic time-domain data to obtain the corresponding frequency domain representation. This frequency domain representation is used to characterize the energy distribution of the seismic data at different frequency locations, providing a basic data source for amplitude spectrum analysis. The frequency domain transformation itself is not limited to a specific implementation form; its purpose is to obtain a stable and analyzable frequency expression.

[0035] Based on the frequency domain representation results described above, the amplitude spectrum of the seismic data is further extracted. The amplitude spectrum reflects the energy distribution corresponding to different frequency components, and it can intuitively characterize the energy concentration areas of the seismic data in the frequency domain. Compared with directly judging signal characteristics in the time domain, the amplitude spectrum is more conducive to identifying the differences between effective information and background components in the frequency dimension.

[0036] After obtaining the amplitude spectrum, frequency energy distribution analysis is performed. This analysis focuses on the overall distribution of amplitude energy along the frequency axis, rather than isolated variations at a single frequency point. Using preset energy determination rules, the energy levels at each frequency position in the amplitude spectrum are assessed, and a set of frequency points that meet the determination criteria is selected. These energy determination rules are used to exclude frequency components with excessively low energy or insufficient stability, preventing noise-dominated frequencies from being included in subsequent processing.

[0037] Based on this, adjacent frequency points that satisfy the energy determination rules are merged to form a continuous frequency range. The continuity determination ensures the integrity of the identified frequency range in the frequency domain, avoiding interference from scattered frequency points in the effective frequency band determination. Through this merging process, several candidate effective frequency ranges are obtained, which characterize the frequency range where amplitude energy is relatively concentrated.

[0038] Furthermore, in the frequency energy distribution analysis process, the amplitude spectrum dynamic range parameter is introduced as one of the judgment criteria. The amplitude spectrum dynamic range parameter is used to define a reasonable range of amplitude energy variation, providing a clear scale reference for the energy determination process. In the amplitude spectrum, the amplitude energy distribution corresponding to each frequency position is first obtained. Then, based on the amplitude spectrum dynamic range parameter, the amplitude energy distribution is determined, and frequency points where the amplitude energy falls within the dynamic range are identified.

[0039] For frequency points that meet the criteria for determining the dynamic range of the amplitude spectrum, an adjacency determination is performed again, merging adjacent frequency points on the frequency axis into a continuous frequency interval. This step further stabilizes the boundaries of the candidate frequency intervals and reduces the impact of local anomalies on the determination of the effective frequency band. Finally, the obtained continuous frequency intervals are determined as the candidate effective frequency intervals corresponding to the seismic data.

[0040] After completing the above-described determination process, the candidate effective frequency intervals are uniformly confirmed, and the effective frequency band range corresponding to the seismic data is determined accordingly. This effective frequency band range serves as the fundamental constraint for subsequent amplitude spectrum sample construction, model training, and frequency domain reconstruction processes, and is maintained throughout the entire high-resolution processing workflow.

[0041] In one specific implementation, let the real number... Represents the Nyquist frequency; a real number. ( () represents the dominant frequency of the zero-phase seismic wavelet; real number ( () represents frequency; real number ( , and ) indicates the dominant frequency is The normalized amplitude spectrum of the zero-phase seismic wavelet; real number ( () represents the dynamic range of the amplitude spectrum of the low-resolution zero-phase seismic wavelet generation. This represents the effective band driving amplitude spectrum of the low-resolution zero-phase seismic wavelet, which is composed of... , and Earth - Construction of the filter amplitude attenuation function; real number and ( ) respectively represent The minimum and maximum frequencies within the effective frequency band; then, The effective frequency band is This is represented as:

[0042] (1)

[0043] (2)

[0044] in, and ( All of them are constants greater than zero; and ( All of them are constants greater than zero; Generate a random number between 0 and 1; ( () represents the formation quality factor; Take the maximum value; ( (This refers to the propagation time.) It is important to note here that... Beyond the effective frequency band This is called background noise.

[0045] Furthermore, let the real number... ( ) is used to represent ( and Amplitude spectrum threshold parameter; real number For the reason , and Constructed high-resolution zero-phase seismic wavelet effective frequency band The initial amplitude spectrum of the driving force can then be expressed as:

[0046] (3)

[0047] This enhances the energy of each internal frequency component. The effective frequency band, while maintaining The energy of background noise outside the effective frequency band. Therefore, The corresponding seismic wavelet resolution is higher than The corresponding seismic wavelet resolution. Accordingly, let Indicates amplitude spectrum and The amplitude spectrum compensation function between them; then:

[0048] (4)

[0049] set up The resolution identification factor, a complex number, represents the resolution identification factor of seismic data (or seismic wavelet). ( () indicates earthquake data Given the spectrum of a zero-phase seismic wavelet, according to the definition of resolution, the Fourier transform theorem, and Parseval's theorem, it can be expressed as:

[0050] (5)

[0051] in, It is important to note that... The higher the value, the higher the seismic resolution. Table 1 shows a comparison of the seismic wavelet resolution identification factors for low-resolution and high-resolution seismic wavelets, which are calculated from Equation (5) and the corresponding amplitude spectrum.

[0052] Table 1 Comparison of seismic resolution identification factors between low-resolution and high-resolution seismic wavelets

[0053] / Seismic resolution identification factor Low-resolution seismic wavelet 0.1653 High-resolution seismic wavelet 0.2214

[0054] EQRG (Effective-band and Quadratic-spectrum driven Reflection coefficient Generation) aims to extract the reflection coefficient amplitude spectrum using in-situ or synthetic seismic data. This represents the amplitude spectrum of the fringe reflection coefficient. express The data augmentation results are as follows:

[0055] (6)

[0056] Where m is a constant, It is worth noting that the reflection coefficients generated by EQRG are derived from in-situ or synthetic seismic data, which can provide a large amount of reflection coefficient data and generate a large training dataset. Sufficient training samples ultimately enhance the network's generalization performance.

[0057] Step S2: Based on the effective frequency band range, construct an amplitude spectrum training sample set, which includes low-resolution seismic amplitude spectrum training samples generated within the same effective frequency band range and their corresponding high-resolution seismic amplitude spectrum training samples.

[0058] Specifically, based on the effective frequency band, an amplitude spectrum training sample set is constructed, including: generating a low-resolution seismic amplitude spectrum corresponding to a wavelet of a low-resolution earthquake within the effective frequency band, as a low-resolution seismic amplitude spectrum training sample; performing energy enhancement processing on the seismic wavelet amplitude spectrum within the same effective frequency band as the low-resolution seismic amplitude spectrum training sample to generate a high-resolution seismic amplitude spectrum, as a high-resolution seismic amplitude spectrum training sample corresponding one-to-one with the low-resolution seismic amplitude spectrum training sample; maintaining the background energy distribution of the low-resolution seismic amplitude spectrum training sample and the high-resolution seismic amplitude spectrum training sample consistent outside the effective frequency band; and combining the low-resolution seismic amplitude spectrum training sample and the high-resolution seismic amplitude spectrum training sample to form the amplitude spectrum training sample set.

[0059] Furthermore, within the same effective frequency band as the low-resolution seismic amplitude spectrum training samples, energy enhancement processing is performed on the seismic wavelet amplitude spectrum to generate a high-resolution seismic amplitude spectrum, which serves as a high-resolution seismic amplitude spectrum training sample corresponding one-to-one with the low-resolution seismic amplitude spectrum training samples. This includes: within the effective frequency band, filtering the amplitude energy at corresponding frequency positions in the low-resolution seismic wavelet amplitude spectrum based on a preset amplitude spectrum threshold parameter; performing weighted enhancement processing on frequency components that meet the amplitude spectrum threshold parameter conditions to increase the amplitude energy of each frequency component within the effective frequency band; outside the effective frequency band, keeping the amplitude energy of corresponding frequency components in the seismic wavelet amplitude spectrum unchanged; and generating a corresponding high-resolution seismic amplitude spectrum based on the enhanced amplitude energy distribution, which serves as the high-resolution seismic amplitude spectrum training sample.

[0060] In this embodiment of the invention, an amplitude spectrum training sample set for training an amplitude spectrum-driven artificial intelligence model is constructed based on the aforementioned determined effective frequency band range. The amplitude spectrum training sample set is generated within the same effective frequency band range to ensure consistency in frequency coverage and spectral structure of the training samples, avoiding the introduction of additional interference factors due to frequency band inconsistencies. During the construction of this sample set, low-resolution seismic amplitude spectrum training samples and high-resolution seismic amplitude spectrum training samples are generated in pairs, maintaining a strict correspondence in frequency position and frequency band range.

[0061] Specifically, within the effective frequency band, a low-resolution seismic amplitude spectrum is generated based on the seismic wavelet corresponding to the low-resolution seismic data, and this spectrum is used as a training sample for the low-resolution seismic amplitude spectrum. This low-resolution amplitude spectrum retains the frequency distribution characteristics and energy structure of the original seismic data within the effective frequency band, without adjusting its frequency positions; it serves only as the baseline input for constructing subsequent sample pairs. The low-resolution seismic amplitude spectrum training samples obtained in this way can accurately reflect the spectral characteristics of actual seismic data within the effective frequency band.

[0062] Within the same effective frequency band as the low-resolution seismic amplitude spectrum training samples, energy enhancement processing is performed on the seismic wavelet amplitude spectrum to generate a high-resolution seismic amplitude spectrum, which serves as a one-to-one high-resolution seismic amplitude spectrum training sample corresponding to the low-resolution seismic amplitude spectrum training samples. The energy enhancement processing only applies to frequency components within the effective frequency band, and the enhancement method primarily involves amplitude energy adjustment without introducing new frequency components, thereby ensuring the consistency of the amplitude spectrum in frequency structure before and after enhancement.

[0063] Furthermore, within the effective frequency band, a preset amplitude spectrum threshold parameter is introduced to filter the amplitude energy at corresponding frequency positions in the low-resolution seismic wavelet amplitude spectrum, performing enhancement operations only on frequency components that meet the threshold conditions. For frequency components that meet the amplitude spectrum threshold parameter conditions, their amplitude energy is weighted and enhanced according to a preset weighting rule to improve the energy level of the main frequency components within the effective frequency band. This weighted enhancement process enhances the resolution information within the effective frequency band while maintaining the continuity of the overall amplitude spectrum shape.

[0064] For frequency components located outside the effective frequency band, the background energy distribution in the low-resolution seismic amplitude spectrum is kept consistent with that in the high-resolution seismic amplitude spectrum, and their amplitude energy is not adjusted. This avoids introducing unnecessary energy changes in the non-effective frequency band, ensuring that the differences between training sample pairs are mainly concentrated within the effective frequency band, thereby reducing the interference of background noise on the model training process.

[0065] Based on the amplitude energy distribution after energy enhancement processing, a corresponding high-resolution seismic amplitude spectrum is generated, serving as a training sample for the high-resolution seismic amplitude spectrum. Finally, the low-resolution seismic amplitude spectrum training samples are combined with the high-resolution seismic amplitude spectrum training samples to form the amplitude spectrum training sample set. This sample set construction method ensures that the training samples maintain consistency in frequency band, frequency position, and background energy distribution, while simultaneously establishing clear amplitude energy differences within the effective frequency band, providing a stable and controllable sample foundation for the subsequent training of amplitude spectrum-driven artificial intelligence models.

[0066] In one specific implementation, an effective frequency band-driven seismic amplitude spectrum training dataset is generated. The aim is to generate effective frequency band-driven synthetic low-resolution seismic training data using the seismic reflection coefficient amplitude spectrum and the low-resolution seismic wavelet amplitude spectrum; and to generate effective frequency band-driven synthetic high-resolution seismic training samples using the seismic reflection coefficient and the high-resolution seismic wavelet. Let... This represents training samples of low-resolution earthquake amplitude spectra. Let the training samples represent high-resolution seismic amplitude spectra. Then:

[0067] (7)

[0068] Step S3: Train the amplitude spectrum-driven artificial intelligence model based on the amplitude spectrum training sample set, and use the trained artificial intelligence model to process the amplitude spectrum corresponding to the seismic data to be processed to obtain the seismic amplitude spectrum.

[0069] Specifically, training the amplitude spectrum-driven artificial intelligence model based on the amplitude spectrum training sample set includes: using low-resolution seismic amplitude spectrum training samples from the amplitude spectrum training sample set as model input data, and using the corresponding high-resolution seismic amplitude spectrum training samples as model training labels; constructing an amplitude spectrum mapping relationship based on the input data and the training labels, and updating the parameters of the amplitude spectrum-driven artificial intelligence model; and repeating the parameter update process of the amplitude spectrum mapping relationship within a preset training round to obtain a trained amplitude spectrum-driven artificial intelligence model.

[0070] Furthermore, based on the input data and the training labels, an amplitude spectrum mapping relationship is constructed, and the parameters of the amplitude spectrum-driven artificial intelligence model are updated, including: inputting the low-resolution seismic amplitude spectrum corresponding to the input data into the amplitude spectrum-driven artificial intelligence model to obtain the corresponding predicted amplitude spectrum output result; comparing the predicted amplitude spectrum output result with the high-resolution seismic amplitude spectrum corresponding to the training labels to obtain an amplitude spectrum deviation result used to characterize the difference between the two; and performing an update operation on the trainable parameters in the amplitude spectrum-driven artificial intelligence model based on the amplitude spectrum deviation result to gradually correct the amplitude spectrum mapping relationship.

[0071] In this embodiment of the invention, an amplitude spectrum-driven artificial intelligence model is trained based on the amplitude spectrum training sample set constructed above. After training, the trained model is used to process the amplitude spectrum corresponding to the seismic data to be processed to obtain an updated seismic amplitude spectrum. This training process uses the frequency domain amplitude spectrum as the only input and output object, and the training objective is clearly defined as the mapping relationship between low-resolution amplitude spectrum and high-resolution amplitude spectrum, without involving direct adjustment of phase information or time-domain waveforms.

[0072] In the specific training process, low-resolution seismic amplitude spectrum training samples from the amplitude spectrum training sample set are used as the model's input data, and the corresponding high-resolution seismic amplitude spectrum training samples are used as the model's training labels. The input data and training labels maintain consistency in frequency location, frequency band range, and background energy distribution, differing only in amplitude distribution within the effective frequency band. This approach allows the model to focus on learning the amplitude variation patterns within the effective frequency band during the training phase, avoiding interference from ineffective frequency bands.

[0073] Based on the input data and training labels, an amplitude spectrum mapping relationship is constructed, and parameter update operations are performed on the amplitude spectrum-driven artificial intelligence model. Specifically, in each training round, the low-resolution seismic amplitude spectrum is input into the amplitude spectrum-driven artificial intelligence model to obtain the corresponding predicted amplitude spectrum output. The predicted amplitude spectrum and the high-resolution seismic amplitude spectrum corresponding to the training labels are in the same frequency coordinate system, and the two can be directly compared and analyzed.

[0074] A difference calculation is performed between the predicted amplitude spectrum output and the training labels to obtain an amplitude spectrum deviation result, which characterizes the difference in amplitude distribution between the two. This amplitude spectrum deviation result reflects the degree of deviation between the model's current output and the target amplitude spectrum at various frequency positions. Based on this amplitude spectrum deviation result, an update operation is performed on the trainable parameters in the amplitude spectrum-driven artificial intelligence model to correct the current amplitude spectrum mapping relationship.

[0075] The above prediction, comparison, and parameter update process is repeated within a preset training cycle. As the training cycle progresses, the predicted amplitude spectrum output by the model gradually approaches the high-resolution seismic amplitude spectrum corresponding to the training label, and the amplitude spectrum deviation shows a convergence trend. After completing the preset training cycle, a trained amplitude spectrum-driven artificial intelligence model is obtained.

[0076] After model training is complete, the amplitude spectrum corresponding to the seismic data to be processed is input into the amplitude spectrum-driven artificial intelligence model to obtain the seismic amplitude spectrum after model processing. The processed seismic amplitude spectrum remains consistent with the input amplitude spectrum in terms of frequency position and frequency band range, but within the effective frequency band, it exhibits amplitude distribution characteristics updated based on the training-derived mapping relationship. This seismic amplitude spectrum is used as input data for subsequent frequency domain reconstruction steps.

[0077] In one specific implementation, Figure 2 The diagram illustrates the architecture of an amplitude spectrum-driven deep learning network for high-resolution seismic processing. It takes low-resolution seismic training samples as input and corresponding high-resolution seismic training samples as labels, outputting the amplitude spectrum of high-resolution seismic data. It's important to note that each convolutional module (ConvModule) contains a convolution (Conv) operation and a rectified linear unit (ReLU) operation; the last layer in the DHN uses the Conv operation. Compared to traditional network structures (such as U-Net), the amplitude spectrum-driven deep learning network for high-resolution seismic processing is a lightweight network that can utilize fewer resources for processing parameters in high-resolution seismic data.

[0078] Here, we use Huber loss function to form the loss function of the amplitude spectrum-driven deep learning high-resolution seismic processing network. Let L be the Huber loss function, N be the total number of pixels in the seismic training samples, x(t) be the high-resolution seismic training samples or labels, and y(t) be the high-resolution seismic result obtained by the amplitude spectrum-driven deep learning high-resolution seismic processing network. Then we have:

[0079] (8)

[0080] Where δ is the weighting coefficient, usually δ=1.

[0081] Step S4: Perform continuity judgment on the candidate effective frequency intervals, and determine the effective frequency band range corresponding to the seismic data based on the judgment result.

[0082] Specifically, within the effective frequency band, an amplitude compensation relationship is determined based on the high-resolution seismic amplitude spectrum to describe amplitude changes; outside the effective frequency band, the amplitudes of the corresponding frequency components in the original seismic amplitude spectrum are kept unchanged; the amplitude compensation relationship is applied to the seismic amplitude spectrum to obtain a target seismic amplitude spectrum that has completed amplitude updates within the effective frequency band; based on the phase information corresponding to the target seismic amplitude spectrum and the original seismic data, an inverse frequency domain transformation is performed on the target seismic amplitude spectrum to obtain the corresponding seismic data.

[0083] In this embodiment of the invention, the focus of this stage is to determine the continuity of candidate effective frequency intervals and thereby determine the final effective frequency band range of the seismic data. Candidate intervals often originate from the initial screening using energy determination rules and may themselves exhibit discrete or fragmented distributions. Directly using discrete frequency points for subsequent processing can easily introduce discontinuous changes in the spectrum. Therefore, it is necessary to determine the continuity of candidate intervals and merge adjacent or gently varying frequency intervals. Through this determination process, a continuous and structurally complete effective frequency band range on the frequency axis can be obtained, providing a clear boundary for subsequent amplitude processing.

[0084] After determining the effective frequency band, amplitude processing rules must strictly distinguish between inside and outside the band. Frequency components within the effective frequency band are considered the primary information-carrying area, and their amplitude variations directly affect the resolution characteristics of the seismic data. Based on the high-resolution seismic amplitude spectrum, an amplitude compensation relationship is constructed within this band to describe the magnitude and direction of amplitude adjustments at different frequency locations. This compensation relationship uses frequency location as an index, corresponding one-to-one with the amplitude spectrum, ensuring that amplitude adjustments have a clear frequency directionality.

[0085] Outside the effective frequency band, the processing strategy remains restrained. The amplitudes of the corresponding frequency components in the original seismic amplitude spectrum remain unchanged, without introducing additional energy adjustments. This rule avoids misidentifying ineffective frequency bands as areas requiring enhancement, helping to maintain the stability of the background noise structure. Through this separation of internal and external processing, amplitude variations are confined to a physically meaningful frequency band.

[0086] The amplitude compensation relationship is applied to the seismic amplitude spectrum. The compensation operation is performed in the frequency domain, updating only the amplitudes within the effective frequency band to form the target seismic amplitude spectrum. This target amplitude spectrum remains consistent with the original amplitude spectrum in terms of frequency position and frequency band range, but exhibits the updated amplitude distribution characteristics within the effective frequency band. This process emphasizes amplitude-level reconstruction without involving frequency position rearrangement.

[0087] Based on the phase information corresponding to the target earthquake amplitude spectrum and the original earthquake data, an inverse frequency domain transform is performed on the target earthquake amplitude spectrum. The phase information remains unchanged during this process, constraining the waveform structure of the time-domain reconstruction. Through the inverse frequency domain transform, the corresponding earthquake data can be obtained, realizing the conversion from frequency domain amplitude adjustment to time-domain signal expression. This processing flow introduces controlled amplitude variations while maintaining the original phase structure, enabling the reconstructed earthquake data to exhibit clearer frequency characteristics within the effective frequency band.

[0088] In one specific implementation, such as Figure 3 Effective frequency band driven amplitude spectrum optimization of seismic data aims to obtain ideal high-resolution seismic results (instantaneous phase and energy outside the effective frequency band remain unchanged). It mainly includes the following five steps:

[0089] 1) Applying the maximum selection method to normalize the output of an amplitude spectrum-driven deep learning high-resolution seismic processing network. Reconstruction is performed to obtain the normalized amplitude spectrum. , is represented as:

[0090] (9)

[0091] 2) Calculate the compensation function, which can be expressed as:

[0092] (10)

[0093] 3) Use smoothing algorithms (e.g., mean filter, median filter, etc.) to smooth the surface. By limiting its effective frequency band, we obtain the smoothing compensation function. .

[0094] 4) Fit the amplitude spectrum compensation function using a higher-order function. The fitted amplitude spectrum compensation function is obtained. :

[0095] (11)

[0096] Where G represents a positive integer greater than or equal to 2, h i It is a constant obtained through polynomial fitting.

[0097] 5) Apply the optimized compensation function High-resolution seismic data driven by reconstructed effective frequency bands and optimized amplitude spectrum compensation functions. Let real numbers... Represents high-resolution seismic data driven by effective frequency band and optimized amplitude spectrum compensation function; complex numbers ( () represents the spectrum of high-resolution seismic data Then we have:

[0098] (12)

[0099] (13)

[0100] Equation (13) shows the two amplitude spectra and Having the same phase, therefore corresponding to two seismic data and They have the same instantaneous phase.

[0101] In one specific implementation, such as Figure 4 and Figure 5 As shown, Figure 4 As shown, Figure 4 (a) and Figure 4 (b) The inline201 and crossline423 profiles of the low-resolution seismic data in the field are presented, with a time range of 0–612 ms. It can be seen that the main reflection interfaces in the original seismic data are generally continuous, but in the shallow and middle layers and in areas with obvious tectonic changes, the in-phase axis width is large, the detailed reflection features are not clear enough, and some areas are still affected by noise.

[0102] like Figure 5 As shown, Figure 5 (a)–(d) are based on Figure 4 This refers to the high-resolution seismic processing results obtained from low-resolution seismic data. Among them, Figure 5 (a) is the result obtained by the effective frequency band driven finite sample deep learning high-resolution seismic processing method without effective frequency band driven seismic data amplitude spectrum optimization; Figure 5 (b) is the result obtained by an amplitude spectrum-driven deep learning high-resolution seismic processing network that optimizes the amplitude spectrum of seismic data without effective frequency bands. Figure 5 (c) and Figure 5 (d) respectively correspond to Figure 4 (b), where Figure 5 (c) shows the results obtained by the effective band-driven finite-sample deep learning high-resolution seismic processing method without effective band-driven amplitude spectrum optimization of seismic data. Figure 5 (d) shows the results obtained by introducing an amplitude spectrum-driven deep learning high-resolution seismic processing network.

[0103] Through the Figure 4 (a) Figure 5 (a) and Figure 5 (b) A comparison shows that, compared to the original low-resolution seismic data, Figure 5 (a) and Figure 5 (b) Both reveal more detailed reflection information, as indicated by the yellow arrows, with improved continuity of the phase axis and interlayer detail; meanwhile, Figure 5 (b) Superior in terms of reflective interface sharpness and background noise suppression Figure 5 (a) The area indicated by the green arrow shows higher resolution and signal-to-noise ratio.

[0104] Furthermore, on Figure 4 (b) Figure 5 (c) and Figure 5 (d) Comparative analysis shows that the high-resolution seismic data obtained after introducing the amplitude spectrum driving mechanism are more stable in characterizing mid-to-deep reflection structures. Figure 5 (c) In comparison, Figure 5 (d) The area indicated by the green arrow shows a narrower phase axis and a more continuous reflection pattern, while the local area marked by the yellow arrow shows more identifiable details, indicating that the amplitude spectrum-driven deep learning high-resolution seismic processing network has better adaptability under complex seismic data conditions.

[0105] In one specific implementation, such as Figure 6 This paper presents a workflow for generating and processing high-resolution training samples of seismic amplitude spectra based on effective frequency bands. First, the seismic time-domain data to be processed is acquired, and a frequency domain transformation is performed on the seismic time-domain data to obtain the corresponding frequency domain representation. The amplitude spectrum of the seismic data is extracted in the frequency domain, and based on the frequency energy distribution characteristics of the amplitude spectrum, continuous frequency intervals that satisfy preset energy determination rules are identified, thereby determining the effective frequency band range corresponding to the seismic data. This effective frequency band range serves as a unified frequency constraint for subsequent sample construction and model training.

[0106] After determining the effective frequency band, the amplitude spectrum training sample generation stage begins. Within the effective frequency band, a corresponding low-resolution seismic amplitude spectrum is generated based on the low-resolution seismic wavelet, serving as the low-resolution amplitude spectrum training sample. Simultaneously, within the same effective frequency band, energy enhancement processing is performed on the low-resolution seismic wavelet amplitude spectrum to generate a corresponding high-resolution seismic amplitude spectrum, serving as the high-resolution amplitude spectrum training sample. Outside the effective frequency band, the background energy distribution in both the low-resolution and high-resolution amplitude spectra is kept consistent, thus forming an effective frequency band-driven amplitude spectrum training sample set.

[0107] The amplitude spectrum training sample set is input into an amplitude spectrum-driven artificial intelligence model for training. Low-resolution seismic amplitude spectra are used as model input, and the corresponding high-resolution seismic amplitude spectra are used as training labels to learn the amplitude spectrum mapping relationship within the effective frequency band. After model training is complete, the amplitude spectrum corresponding to the seismic data to be processed is input into the trained artificial intelligence model to obtain an updated seismic amplitude spectrum. Within the effective frequency band, an amplitude compensation relationship is constructed based on the updated seismic amplitude spectrum, and an inverse frequency domain transformation is performed in conjunction with the phase information of the original seismic data to obtain the corresponding high-resolution seismic data.

[0108] like Figure 7 As shown, this invention provides an amplitude spectrum-driven artificial intelligence high-resolution processing system. The system includes: an acquisition unit for acquiring seismic data to be processed, transforming the seismic data to the frequency domain, and determining the effective frequency band range corresponding to the seismic data based on the transformed frequency domain result; a sample construction unit for constructing an amplitude spectrum training sample set based on the effective frequency band range, the amplitude spectrum training sample set including low-resolution seismic amplitude spectrum training samples generated within the same effective frequency band range and their corresponding high-resolution seismic amplitude spectrum training samples; a training unit for training an amplitude spectrum-driven artificial intelligence model based on the amplitude spectrum training sample set, and using the trained artificial intelligence model to process the amplitude spectrum corresponding to the seismic data to be processed to obtain the seismic amplitude spectrum; and a reconstruction unit for reconstructing the seismic amplitude spectrum based on the effective frequency band range to obtain the corresponding seismic data.

[0109] The present invention also provides a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the above-described amplitude spectrum-driven artificial intelligence high-resolution processing method.

[0110] Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. This program is stored in a storage medium and includes several instructions to cause a microcontroller, chip, or processor to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0111] The optional embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the embodiments of the present invention are not limited to the specific details described above. Within the scope of the technical concept of the embodiments of the present invention, various simple modifications can be made to the technical solutions of the embodiments of the present invention, and these simple modifications all fall within the protection scope of the embodiments of the present invention. It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. To avoid unnecessary repetition, the embodiments of the present invention will not further describe the various possible combinations.

[0112] Furthermore, various different embodiments of the present invention can be combined in any way, as long as they do not violate the spirit of the embodiments of the present invention, they should also be regarded as the content disclosed by the embodiments of the present invention.

Claims

1. An artificial intelligence high-resolution processing method based on amplitude spectrum driving, characterized in that, The method comprises: acquiring seismic data to be processed, transforming the seismic data to a frequency domain, and determining an effective frequency band range corresponding to the seismic data based on a transformed frequency domain result; based on the effective frequency band range, constructing an amplitude spectrum training sample set, the amplitude spectrum training sample set comprising low-resolution seismic amplitude spectrum training samples and high-resolution seismic amplitude spectrum training samples corresponding thereto generated within the same effective frequency band range; training an artificial intelligence model driven by an amplitude spectrum based on the amplitude spectrum training sample set, and processing an amplitude spectrum corresponding to the seismic data to be processed using the trained artificial intelligence model to obtain a seismic amplitude spectrum; based on the effective frequency band range, reconstructing the seismic amplitude spectrum to obtain corresponding seismic data.

2. The artificial intelligence high-resolution processing method based on amplitude spectrum driving according to claim 1, characterized in that, acquiring seismic data to be processed, transforming the seismic data to a frequency domain, and determining an effective frequency band range corresponding to the seismic data based on a transformed frequency domain result, comprising: acquiring seismic time domain data to be processed, and performing frequency domain transformation on the seismic time domain data to obtain a corresponding frequency domain representation result; extracting an amplitude spectrum of the seismic data based on the frequency domain representation result; performing continuity judgment on the candidate effective frequency interval, and determining an effective frequency band range corresponding to the seismic data based on the judgment result.

3. The artificial intelligence high resolution processing method based on amplitude spectrum driving according to claim 2, characterized in that, performing frequency energy distribution analysis on the amplitude spectrum, identifying a continuous frequency interval satisfying a preset energy judgment rule as a candidate effective frequency interval, comprising: in the amplitude spectrum, acquiring an amplitude energy distribution corresponding to each frequency position; based on an amplitude spectrum dynamic range parameter of the amplitude spectrum, judging the amplitude energy distribution to identify a frequency interval in which amplitude energy is located within the amplitude spectrum dynamic range; in the amplitude spectrum, merging adjacent frequency points satisfying the amplitude spectrum dynamic range judgment condition into a continuous frequency interval; determining the continuous frequency interval as the candidate effective frequency interval. 4.The artificial intelligence high-resolution processing method based on amplitude spectrum driving according to claim 1, wherein, based on the effective frequency band range, constructing an amplitude spectrum training sample set, comprising: within the effective frequency band range, generating a corresponding low-resolution seismic data amplitude spectrum through a low-resolution seismic wavelet, and taking it as a low-resolution seismic amplitude spectrum training sample; within the same effective frequency band range as the low-resolution seismic amplitude spectrum training sample, generating a corresponding high-resolution seismic data amplitude spectrum through a high-resolution seismic wavelet, and taking it as a high-resolution seismic amplitude spectrum training sample; outside the effective frequency band range, keeping the background noise energy distribution of the low-resolution seismic amplitude spectrum training sample consistent with that of the high-resolution seismic amplitude spectrum training sample; combining the low-resolution seismic amplitude spectrum training sample and the high-resolution seismic amplitude spectrum training sample to form the amplitude spectrum training sample set.

5. The artificial intelligence high resolution processing method based on amplitude spectrum driving according to claim 4, characterized in that, within the same effective frequency band range as the low-resolution seismic amplitude spectrum training sample, generating a corresponding high-resolution seismic data amplitude spectrum through a high-resolution seismic wavelet, and taking it as a high-resolution seismic amplitude spectrum training sample, comprising: within the effective frequency band range, based on a preset amplitude spectrum threshold parameter, screening the amplitude energy of the corresponding frequency position in the low-resolution seismic wavelet amplitude spectrum; performing a weighted enhancement process on the frequency components satisfying the amplitude spectrum threshold parameter condition to increase the amplitude energy of each frequency component in the effective frequency band range; outside the effective frequency band range, keeping the amplitude energy of the corresponding frequency component in the seismic wavelet amplitude spectrum unchanged; based on the enhanced amplitude energy distribution, generating a corresponding high-resolution seismic amplitude spectrum as the high-resolution seismic amplitude spectrum training sample.

6. The artificial intelligence high-resolution processing method based on amplitude spectrum driving according to claim 4, characterized in that, training an amplitude spectrum driven artificial intelligence model based on the amplitude spectrum training sample set, including: taking the low-resolution seismic amplitude spectrum training sample in the amplitude spectrum training sample set as model input data, and taking the corresponding high-resolution seismic amplitude spectrum training sample as model training label; based on the input data and the training label, constructing an amplitude spectrum mapping relationship, and updating the parameters of the amplitude spectrum driven artificial intelligence model; repeating the parameter updating process of the amplitude spectrum mapping relationship within a preset training round to obtain the trained amplitude spectrum driven artificial intelligence model.

7. The artificial intelligence high resolution processing method based on amplitude spectrum driving according to claim 6, characterized in that, based on the input data and the training label, constructing an amplitude spectrum mapping relationship, and updating the parameters of the amplitude spectrum driven artificial intelligence model, including: inputting the low-resolution seismic amplitude spectrum corresponding to the input data into the amplitude spectrum driven artificial intelligence model to obtain a corresponding predicted amplitude spectrum output result; comparing the predicted amplitude spectrum output result with the high-resolution seismic amplitude spectrum corresponding to the training label to obtain an amplitude spectrum deviation result for characterizing the difference between them; based on the amplitude spectrum deviation result, performing an update operation on the trainable parameters in the amplitude spectrum driven artificial intelligence model to gradually correct the amplitude spectrum mapping relationship. 8.The artificial intelligence high-resolution processing method based on amplitude spectrum driving according to claim 1, wherein, based on the effective frequency band range, reconstructing the seismic amplitude spectrum to obtain corresponding seismic data, including: within the effective frequency band range, determining an amplitude compensation relationship for describing amplitude changes based on the high-resolution seismic amplitude spectrum; outside the effective frequency band range, keeping the amplitude of the corresponding frequency component in the original seismic amplitude spectrum unchanged; applying the amplitude compensation relationship to the seismic amplitude spectrum to obtain a target seismic amplitude spectrum with amplitude updated in the effective frequency band range; based on the target seismic amplitude spectrum and the phase information corresponding to the original seismic data, performing inverse frequency domain transformation on the target seismic amplitude spectrum to obtain corresponding seismic data.

9. An artificial intelligence high resolution processing system based on amplitude spectrum driving, characterized by, The system includes: a collection unit configured to obtain to-be-processed seismic data, transform the seismic data to the frequency domain, and determine an effective frequency band range corresponding to the seismic data based on the transformed frequency domain result; a sample construction unit configured to construct an amplitude spectrum training sample set based on the effective frequency band range, the amplitude spectrum training sample set including a low-resolution seismic amplitude spectrum training sample and a corresponding high-resolution seismic amplitude spectrum training sample generated in the same effective frequency band range; a training unit configured to train an amplitude spectrum driven artificial intelligence model based on the amplitude spectrum training sample set, and process the amplitude spectrum corresponding to the to-be-processed seismic data using the trained artificial intelligence model to obtain a seismic amplitude spectrum; A reconstruction unit is configured to reconstruct the seismic amplitude spectrum based on the effective frequency band range to obtain corresponding seismic data.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores instructions, which, when executed on a computer, cause the computer to perform the artificial intelligence high-resolution processing method based on the amplitude spectrum driving according to any one of claims 1-8.

Images