Intelligent noise removal methods and devices, electronic devices, storage media, and software products based on multi-feature deep fusion

Abstract

This application provides an intelligent noise removal method and apparatus based on multi-feature deep fusion, relating to the field of petroleum geophysical exploration data processing technology. The method includes: designing an observation system according to a pre-defined geological task; dividing seismic gather data into N regions and selecting low signal-to-noise ratio (SNR) shots; identifying data above the first arrival time window as the first type of noise data; selecting labeled shots and standardizing and quality-evaluating the labeled shot data; adding noise data from a noise database to the labeled shots according to a set weight ratio, forming noisy shot gather data through data fusion technology, adding gain, and standardizing; constructing a deep neural network model; generating training samples based on the labeled shot data through data augmentation and sample partitioning techniques; and removing noise through feature analysis and learning of the input data using the model to obtain high-quality data. The modular process and flexible parameter configuration of this application facilitate integration with existing processing systems, demonstrating strong engineering practicality and promotional value.

Description

Technical Field

[0001] This application relates to the field of petroleum geophysical exploration data processing technology, and in particular to an intelligent noise removal method and device based on multi-feature deep fusion. Background Technology

[0002] In the field of geological exploration, the accurate acquisition and processing of seismic data is crucial for gaining a deeper understanding of underground geological structures and identifying potential resources. However, in the actual seismic data acquisition process, various types of noise interference are inevitable, which seriously affects the quality of the data and the accuracy of subsequent analysis results.

[0003] Traditional seismic data acquisition methods often lack precise consideration of complex geological conditions in the design of observation systems, leading to limitations in the acquired data. For example, the settings for receiver spacing and the number of receivers may not adequately cover the target area, resulting in data that cannot fully reflect subsurface geological information. Furthermore, early acquisition equipment had limited accuracy and poor stability, making it difficult to maintain data consistency and accuracy over long periods of acquisition. This resulted in large errors and missing data, posing significant challenges to subsequent data processing and analysis.

[0004] In the data processing stage, traditional methods for dividing seismic gather data regions are relatively simple and lack comprehensive analysis of geographical information and geological conditions, making it difficult to achieve efficient and accurate data sorting. When setting the first arrival time window, the lack of sufficient integration of seismic wave propagation theory and actual geological conditions leads to inaccurate selection of low signal-to-noise ratio (SNR) shots. Furthermore, the algorithm used to determine the SNR is too simplistic and cannot accurately distinguish the SNR of different data types, resulting in some low SNR data not being effectively identified and processed.

[0005] Traditional methods have several shortcomings in constructing noise and label libraries. For example, the extraction and processing of noise data are not precise enough, resulting in a noise library that lacks representativeness and cannot fully simulate noise in real seismic data. Furthermore, the selection of label shots does not adequately consider factors such as the complexity of geological structures, seismic wave propagation paths, and the signal-to-noise ratio of the data, leading to low-quality data in the label library and an inability to provide high-quality samples for model training.

[0006] In terms of model building and training, early deep neural network models had simple architectures and were unable to effectively learn and extract features from complex seismic data. The optimization algorithms for these models were also relatively outdated, unable to dynamically adjust according to the characteristics of the data and changes during the training process, resulting in slow convergence speed and low accuracy.

[0007] In summary, existing seismic data processing technologies have many shortcomings when facing complex geological conditions and diverse noise interference. This invention aims to overcome these problems by employing intelligent noise removal technology based on multi-feature deep fusion. From observation system design, data acquisition, and processing to model building and application, this comprehensively improves the quality and processing efficiency of seismic data, providing more reliable data support for geological exploration. Summary of the Invention

[0008] This application discloses an intelligent noise removal method and apparatus based on multi-feature deep fusion.

[0009] In a first aspect, this application discloses an intelligent noise removal method based on multi-feature deep fusion, characterized in that the method includes: Based on the pre-set geological task, the observation system is designed with the excitation point location as the coordinate center, the shot line as the vertical axis, and the receiver line as the horizontal axis. The seismic gather data is evenly divided into N regions in the rectangular coordinate system, and low signal-to-noise ratio shots with a signal-to-noise ratio lower than the first set value are selected. For each low signal-to-noise ratio shot, data above the first arrival time window is identified as the first type of noise data; the noise data is then shifted downward to construct the second type of noise data with the same trace length as the original data; the first type of noise data and the second type of noise data are integrated and stored to form a noise database; Based on the complexity of the geological structure and the signal-to-noise ratio of the data, shot collection data with a signal-to-noise ratio higher than the second set value are selected as tagged shot data in different areas of the exploration area, and the tagged shot data are standardized and quality evaluated. The noise data in the noise library is added to the tagged gun according to a set ratio weight, and the noisy gun set data is formed by data fusion technology. Gain is added to the noisy gun set data and it is then standardized. Using multi-channel features fused from multi-attribute data as input, including data with gain and standardized data, a deep neural network model is constructed with seismic trace data after noise probability masking as output. Through a pre-defined network structure and parameter initialization method, complex seismic data is effectively learned and features are extracted. Training samples are generated based on labeled shot data using data augmentation and sample partitioning techniques. A hybrid loss function, including mean squared error, structural similarity loss, and sparse regularization term, is used to optimize the network parameters. During training, a stochastic gradient descent optimization algorithm combined with a learning rate adjustment strategy is employed to ensure rapid convergence and stable training of the model. By analyzing and learning the features of the input data through the model to remove noise, high-quality data after denoising is obtained.

[0010] Optionally, the step of selecting a low signal-to-noise ratio (SNR) gun with a SNR lower than a first set value includes: The signal-to-noise ratio of the data is determined by calculating the mean of the energy ratios of the data before and after the first arrival wave; If the mean ratio of the energy of the data before and after the first arrival wave in the source data is less than 1.0, it is identified as low signal-to-noise ratio data. If the average ratio of the energy of the data before and after the first arrival wave in well shot data is less than 1.5, it is identified as low signal-to-noise ratio data.

[0011] Optionally, the step of integrating and storing the first type of noise data and the second type of noise data to form a noise library includes: Noise data is constructed by using the same number of sampling points per channel as the length of each channel in the shot gather data and the same sampling interval.

[0012] Optionally, the step of selecting shot gather data with a signal-to-noise ratio higher than a second set value as tagged shots in different areas of the exploration zone based on the complexity of the geological structure and the signal-to-noise ratio of the data includes: The number of seed shots is 0.5% of the total number of shots, and the higher the signal-to-noise ratio, the fewer seed shots are selected, and the lower the signal-to-noise ratio, the more seed shots are selected.

[0013] Optionally, the step of adding noise data from the noise database to the tag gun according to a set proportional weight includes: The noise data in the noise library is added to the tag shot with a weighted ratio of 1:1. If the signal-to-noise ratio of the shot data is low, the weighted ratio is increased, and if the signal-to-noise ratio of the shot data is high, the weighted ratio is decreased.

[0014] Optionally, it also includes: normalizing the noisy shot gather data.

[0015] Optionally, the step of constructing a deep neural network model using multi-channel features fused from multi-attribute data as input, wherein the multi-channel features include data with gain and normalized data, and using seismic trace data after noise probability masking as output, includes: A deep neural network model is constructed by adopting a multi-scale fusion U-Net architecture and embedding a channel attention module, using features that are fused at the feature level after parallel branch input as multi-attribute data.

[0016] Optionally, the step of optimizing network parameters using a hybrid loss function, including mean squared error, structural similarity loss, and sparse regularization term, includes: An adaptive learning rate algorithm is used to optimize the deep neural network model by dynamically adjusting the batch size and dropout rate based on the signal-to-noise ratio of the validation set.

[0017] Alternatively, network parameters can be optimized according to the expression described below:

[0018] in, , , Weighting coefficients for different loss terms, These are network parameters.

[0019] Secondly, this application discloses an intelligent noise removal device based on multi-feature deep fusion, characterized in that the device comprises: The low signal-to-noise ratio shot selection module is used to design the observation system according to the preset geological task. With the excitation point location as the coordinate center, the shot line as the vertical axis, and the receiver line as the horizontal axis, the seismic gather data is evenly divided into N regions in the rectangular coordinate system, and low signal-to-noise ratio shots with a signal-to-noise ratio lower than the first set value are selected. The noise database construction module is used to determine the data above the first arrival time window as the first type of noise data for each low signal-to-noise ratio shot; and to perform downward translation and sliding processing on the noise data to construct the second type of noise data with the same channel length as the original data; and to integrate and store the first type of noise data and the second type of noise data to form a noise database. The tag-shot construction module is used to select shot collection data with a signal-to-noise ratio higher than a second set value as tag-shot data in different areas of the exploration area based on the complexity of the geological structure and the signal-to-noise ratio of the data, and to perform standardization processing and quality assessment on the tag-shot data. The gun collection data fusion module is used to add noise data from the noise library to the tagged gun according to a set ratio weight, form noisy gun collection data through data fusion technology, and add gain and standardize the noisy gun collection data. The model building module is used to construct a deep neural network model with multi-channel features obtained by fusing multi-attribute data as input, including data with gain and standardized data, and seismic trace data after noise probability masking as output. It effectively learns and extracts features from complex seismic data through a preset network structure and parameter initialization method. Training samples are generated based on labeled shot data through data augmentation and sample partitioning techniques. A hybrid loss function, including mean squared error, structural similarity loss, and sparse regularization term, is used to optimize the network parameters. During training, a stochastic gradient descent optimization algorithm combined with a learning rate adjustment strategy is employed to ensure rapid convergence and stable training of the model. The noise removal module is used to remove noise by analyzing and learning the features of the input data through a model, resulting in high-quality denoised data.

[0020] Thirdly, this application discloses an electronic device comprising: a processor; and a memory for storing processor-executable instructions; wherein the processor is configured to perform the method as described in any of the preceding aspects.

[0021] Fourthly, this application discloses a non-transitory computer-readable storage medium in which, when the instructions in the storage medium are executed by a processor of an electronic device, enable the electronic device to perform the methods described in any of the preceding aspects.

[0022] Fifthly, this application discloses a computer program product in which, when the instructions in the computer program product are executed by a processor of an electronic device, the electronic device is enabled to perform the method described in any of the preceding aspects.

[0023] The technical solution provided in this application may include the following beneficial effects: This application, in processing seismic data, firstly designs a precise observation system based on geological tasks using advanced technology, employing high-precision equipment and real-time monitoring to ensure the accuracy and completeness of acquired data. Next, the seismic gather data is divided into regions, sorted using professional algorithms, and first-arrival time windows are set, with low signal-to-noise ratio (SNR) shots selected. For these shots, noise data is precisely extracted and optimized to construct a noise database. Simultaneously, a label database is constructed by selecting high SNR shot sets based on multiple factors, and noise is dynamically added to the labeled shots to form noisy shot set data. Then, adaptive gain and standardization transformations are applied to the noisy data. A deep neural network model employing a multi-scale fusion U-Net architecture and a self-developed channel attention module is constructed, inputting multi-attribute fusion features, and improving model performance through optimization algorithms. Training samples are generated based on labeled shots, and a hybrid loss function is used to optimize the network. Finally, the model uses noise identification and removal algorithms to denoise the input data. Through real-time monitoring and automatic parameter adjustment, intelligent and adaptive seismic data denoising is achieved, meeting the needs of different geological conditions and effectively improving seismic data quality. The modular process and flexible parameter configuration provided in this application are easy to integrate into existing processing systems, possessing strong engineering practicality and promotional value. Attached Figure Description

[0024] Figure 1 A flowchart of an intelligent noise removal method based on multi-feature deep fusion provided in this application; Figure 2 A schematic diagram of an observation system designed for a geological mission, provided for this application; Figure 3 A schematic diagram of the 18th permutation in the single-shot seismic records acquired for this application; Figure 4 A schematic diagram of the deep network model provided in this application; Figure 5For the purposes of this application Figure 4 A schematic diagram of the structure breakdown of a deep network model; Figure 6 A schematic diagram of the noise-removed data from the single-shot seismic records acquired in this application; Figure 7 A structural diagram of the intelligent noise removal device based on multi-feature deep fusion provided in this application; Figure 8 A block diagram of an electronic device provided in this application; Figure 9 A block diagram of another electronic device provided in this application. Detailed Implementation

[0025] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0026] To address the problem of poor data quality caused by noise interference in seismic data processing, this application provides an intelligent noise removal method and apparatus based on multi-feature deep fusion.

[0027] The intelligent noise removal method based on multi-feature deep fusion proposed in this application will be described below.

[0028] Example 1 Reference Figure 1 The above is a flowchart of an intelligent noise removal method based on multi-feature deep fusion provided in this application. This method can be applied to electronic devices, and specifically includes the following steps: Step S101: Design an observation system based on the preset geological task. With the excitation point location as the coordinate center, the shot line as the vertical axis, and the receiver line as the horizontal axis, divide the seismic gather data into N regions in a rectangular coordinate system, and select low signal-to-noise ratio shots with a signal-to-noise ratio lower than the first set value.

[0029] In one scenario, low signal-to-noise ratio (SNR) shots can be identified as follows: the SNR of the data is determined by calculating the average ratio of the energy of the data before and after the first arrival wave; if the average ratio of the energy of the data before and after the first arrival wave in the source data is less than 1.0, it is identified as low SNR data; if the average ratio of the energy of the data before and after the first arrival wave in the well shot data is less than 1.5, it is identified as low SNR data.

[0030] Step S102: For each low signal-to-noise ratio shot, determine the data above the first arrival time window as the first type of noise data; and perform downward translation and sliding processing on the noise data to construct the second type of noise data with the same channel length as the original data; integrate and store the first type of noise data and the second type of noise data to form a noise library.

[0031] In one scenario, the first type of noise data and the second type of noise data can be integrated and stored to form a noise database as follows: noise data constructed in a manner where the number of sampling points per channel is the same as the length of each channel of the shot gather data and the sampling interval is the same.

[0032] Step S103: Based on the complexity of the geological structure and the signal-to-noise ratio of the data, select shot gather data with a signal-to-noise ratio higher than the second set value in different areas of the exploration area as tagged shot data, and perform standardization processing and quality assessment on the tagged shot data.

[0033] In one scenario, the step of selecting shot gather data with a signal-to-noise ratio higher than a second set value as tagged shots in different areas of the exploration zone based on the complexity of the geological structure and the signal-to-noise ratio of the data includes: the number of tagged shots is 0.5% of the total number of shots, and the higher the signal-to-noise ratio, the fewer seed shots are selected, and the lower the signal-to-noise ratio, the more seed shots are selected.

[0034] Step S104: Add the noise data in the noise library to the tagged gun according to the set proportional weight, form the noisy gun set data through data fusion technology, and add gain and standardize the noisy gun set data.

[0035] In one scenario, noise data from the noise database can be added to the tag-and-spot system according to a set weighted ratio as follows: the noise data from the noise database is added to the tag-and-spot system at a 1:1 weighted ratio, and if the signal-to-noise ratio (SNR) of the data is low, the weighted ratio is increased; if the SNR of the data is high, the weighted ratio is decreased. It should be noted that the 1:1 weighted ratio is a preferred embodiment of this application and should not be construed as a limitation of the application. Those skilled in the art can set the weighted ratio reasonably based on the inverse relationship between the SNR and the weighted ratio, depending on the specific circumstances.

[0036] Furthermore, in addition to gaining and standardizing the shot gather data, normalized data processing can also be performed on the noisy shot gather data.

[0037] Step S105: Using the multi-channel features obtained from the fusion of multi-attribute data as input, wherein the multi-channel features include data with gain and standardized data, a deep neural network model is constructed with the seismic trace data after noise probability masking as output. Through a preset network structure and parameter initialization method, the complex seismic data is effectively learned and features are extracted. Based on labeled shot data, training samples are generated through data augmentation and sample partitioning techniques. A hybrid loss function, including mean squared error, structural similarity loss, and sparse regularization term, is used to optimize the network parameters. During training, a stochastic gradient descent optimization algorithm is used, combined with a learning rate adjustment strategy, to ensure rapid convergence and stable training of the model.

[0038] In one scenario, a deep neural network model can be constructed as follows: a multi-scale fusion U-Net architecture is adopted and a channel attention module is embedded. The features that are input to parallel branches and then fused at the feature level are used as multi-attribute data to construct a deep neural network model.

[0039] In one scenario, an adaptive learning rate algorithm can be used to dynamically adjust the batch size and dropout rate based on the validation set signal-to-noise ratio to optimize the deep neural network model. Specifically, network parameters are optimized according to the following expression:

[0040] in, , , Weighting coefficients for different loss terms, These are network parameters.

[0041] Step S106: Remove noise by analyzing and learning the features of the input data through the model to obtain high-quality denoised data.

[0042] As can be seen from the above, high-quality seismic data is crucial for analyzing geological structures in geological exploration, but raw data is often affected by various types of noise. By utilizing advanced geological modeling and data analysis techniques to precisely design observation systems, coupled with high-precision calibration and real-time monitoring equipment, the accuracy and integrity of acquired data are ensured, laying a solid foundation for subsequent processing. Professional algorithms are used to perform regional division, sorting, time window setting, and signal-to-noise ratio analysis on the data, accurately constructing noise and label libraries to simulate real noise characteristics. Advanced data fusion and transformation algorithms, as well as deep neural network models, are employed to process and learn from noisy data, achieving effective feature extraction from complex seismic data. By optimizing algorithms and dynamically adjusting parameters, model performance is improved. Finally, noise identification and removal algorithms are used for precise denoising, achieving intelligent and adaptive seismic data denoising to meet the application needs under different geological conditions and provide high-quality data support for geological exploration.

[0043] To facilitate a comprehensive understanding of this application, the intelligent noise removal method based on multi-feature deep fusion provided in this application will be described in detail below.

[0044] Step (1): Design a precise observation system based on the specific geological task.

[0045] The receiver line spacing was set to L meters, and the number of receiver lines was S. Seismic data was acquired by exciting and collecting single-shot data, ensuring m samples per shot and n sampling points per sample. During data acquisition, the accuracy and stability of the acquisition equipment were strictly controlled to ensure the accuracy and completeness of the acquired data.

[0046] Step (2): Using the excitation point as the coordinate center, the shot line as the vertical axis, and the receiver line as the horizontal axis, the seismic gather data is uniformly divided into N regions in a rectangular coordinate system. Within each region, the data is sorted in ascending order of shot-receiver distance, and the time window for the first arrival wave is set. At the same time, at different locations in the work area, based on the actual geological conditions and through signal-to-noise ratio analysis of a single shot, shots with low signal-to-noise ratios are selected.

[0047] In this step, selecting shots with low signal-to-noise ratios (SNR) is mainly determined by calculating the average ratio of the energy of the data before and after the first arrival. For source data, if the average ratio of the energy of the data before and after the first arrival is less than 1.0, it is considered low SNR data; for well shot data, if the average ratio of the energy of the data before and after the first arrival is less than 1.5, it is considered low SNR data.

[0048] Step (3): For each shot with a low signal-to-noise ratio, select the data above the first arrival time window as noise data.

[0049] These noise data are then shifted downwards to construct noise data with the same trace length as the original data. This noise data is then integrated and stored to form a comprehensive and representative noise database. In one scenario, the number of sampling points per trace in the constructed noise data is the same as the length of each trace in the shot gather data, and the sampling interval is also the same.

[0050] Step (4): In different areas of the exploration zone, considering factors such as the complexity of the geological structure and the signal-to-noise ratio of the data, carefully select shot gather data with high signal-to-noise ratio as tagged shots. Standardize and evaluate the quality of these tagged shot data to ensure the reliability and consistency of the data, and then construct a high-quality tag library.

[0051] Generally, 0.5% of the total number of seed shots are selected. Furthermore, the higher the signal-to-noise ratio, the fewer seed shots are selected, and the lower the signal-to-noise ratio, the more seed shots are selected.

[0052] Step (5): The noise data produced in step (3) is added to the labeled shot data constructed in step (4) according to a certain weight ratio. Through data fusion technology, noisy shot gather data is formed. During the addition process, the intensity and distribution of noise are strictly controlled to simulate the diversity and complexity of noise in real seismic data.

[0053] It should be noted that the noise data produced is added to the tag shot constructed in step (4) according to a certain weight ratio. The weight ratio here is generally 1:1. If the signal-to-noise ratio of the shot set data is low, this ratio should be larger, and if the signal-to-noise ratio of the shot set data is high, this ratio should be smaller.

[0054] Step (6): The noisy shot gather data prepared in Step (5) is subjected to gain addition and standardization operations in sequence. During the gain addition process, an adaptive gain algorithm is used based on the dynamic range and signal characteristics of the data to ensure effective signal enhancement and reasonable noise suppression. Simultaneously, the seismic data is standardized to form both gain-added and standardized data. Furthermore, in addition to gain-added and standardized data, normalized data can also be used.

[0055] Step (7): Construct a deep neural network model. The input to this model is the multi-channel features after the fusion of multi-attribute data, including data with gain and standardized data. Through the designed network structure and parameter initialization method, the model is ensured to effectively learn and extract features from complex seismic data. The output of the model is the seismic trace data after noise probability masking. By optimizing the network structure and activation function, the output accuracy and reliability of the model are improved.

[0056] In one scenario, the deep neural network employs a multi-scale fusion U-Net architecture and embeds a channel attention module, where multi-attribute data is input as parallel branches and then fused at the feature level.

[0057] In one scenario, the optimization of the deep neural network model employs an adaptive learning rate algorithm, dynamically adjusting the batch size and dropout rate based on the validation set signal-to-noise ratio.

[0058] Step (8): Based on the labeled cannon data, a rich variety of training samples are generated through data augmentation and sample partitioning techniques. A hybrid loss function, including mean squared error, structural similarity loss, and sparse regularization term, is used to optimize the network parameters. During training, optimization algorithms such as stochastic gradient descent are used in conjunction with learning rate adjustment strategies to ensure rapid convergence and stable training of the model.

[0059] Specifically, the hybrid loss function is as follows:

[0060] in, , , Weighting coefficients for different loss terms, These are network parameters.

[0061] Step (9): Through feature analysis and learning of the input data by the model, noise is accurately removed to obtain high-quality denoised data. During the application process, the performance of the model and the denoising effect of the data are monitored in real time, and parameters are adjusted and optimized according to the actual situation.

[0062] As can be seen, the main technology of this application revolves around seismic data denoising, integrating various professional technologies and innovative algorithms to achieve high-precision noise removal. In the data acquisition phase, advanced geological modeling and data analysis technologies are used to precisely design the observation system. High-precision calibration equipment and a real-time monitoring system ensure the stability of the acquisition equipment and guarantee the accuracy and completeness of the data. During data processing, geographic information processing algorithms are used to divide the data area. First-arrival time windows are set using seismic wave propagation theory and simulation software. A self-developed signal-to-noise ratio (SNR) analysis algorithm is used to select low SNR shots to construct a noise library. Multiple tools and models are used to select high SNR shot sets to construct a label library. Noise is added according to the dynamic adjustment ratio to form noisy shot set data. Then, algorithms such as adaptive gain and normalization are used to transform the noisy data. In terms of model construction, a deep neural network with a multi-scale fusion U-Net architecture and an embedded autonomous channel attention module is adopted. Multi-attribute fusion features are input for learning, and the model is optimized using algorithms such as adaptive learning rate. Finally, model and noise identification and removal algorithms are used for denoising. Real-time monitoring and automatic parameter adjustment are employed to meet the denoising needs of seismic data under different geological conditions.

[0063] Please see Figures 2 to 6 ,in, Figure 4 This demonstrates the overall network architecture for AI-based denoising of seismic data, belonging to an encoder-decoder structure. Specifically, the input is the raw, noisy seismic data, which enters the encoding stage after a series of convolutional operations (such as convolutional layers with different numbers of channels). Multiple convolutional blocks and attention modules (such as the "Convolutional Block Attention Module") extract features from both noise and effective signals, while skip connections preserve shallow details. In the decoding stage, transposed convolutions are used for upsampling to gradually restore the data dimensions, ultimately outputting the denoised seismic data. Figure 4 The overall network architecture shown integrates attention mechanisms and skip connections, effectively preserving the valuable features of seismic data (such as geological information like reflected waves) while suppressing noise.

[0064] Figure 5 right Figure 4 The internal structure of the attention module is explained in detail. Input features are initially encoded through operations such as convolution and activation functions to extract spatial feature information. Features are then weighted along both the channel and spatial dimensions to strengthen features related to effective signals and suppress noise-related features, thereby improving the effectiveness of denoising.

[0065] The following section provides an overall description of the intelligent noise removal method based on multi-feature deep fusion provided in this application, using a specific example.

[0066] S1. Based on the geological task, the designed observation system is 8980-20-40-20-8980, with a receiver line spacing of 200m, a channel spacing of 40m, and 42 receiver lines, each line receiving 450 channels. The observation system is as follows: Figure 2 As shown. Single-shot seismic data were excited and acquired, with 4000 sampling points per trace and a sampling interval of 1 ms. The 18th sequence of the acquired single-shot seismic records is shown below. Figure 3 As shown.

[0067] S2. Using the excitation point as the coordinate center, the shot line as the vertical axis, and the receiver line as the horizontal axis, the seismic gather data is uniformly divided into two regions in a rectangular coordinate system. Within each region, the data is sorted in ascending order of shot-receiver distance, and a time window for the first arrival wave is set. Simultaneously, at different locations within the work area, based on actual geological conditions and through signal-to-noise ratio (SNR) analysis of individual shots, shots with low SNR are selected.

[0068] S3. For each shot with a low signal-to-noise ratio, select the data above the first arrival time window as noise data. Perform a downward shifting and sliding process on these noise data to construct noise data with the same trace length as the original data. Integrate and store these noise data to form a comprehensive and representative noise database.

[0069] S4. In different areas of the exploration zone, considering factors such as the complexity of the geological structure and the signal-to-noise ratio of the data, shot gather data with high signal-to-noise ratio are carefully selected as tagged shots. These tagged shot data are then standardized and quality-assessed to ensure data reliability and consistency, thereby constructing a high-quality tag library.

[0070] S5. Add the noise data generated in S3 to the labeled shot data constructed in S4 according to a certain weighting ratio. Use data fusion technology to form noisy shot gather data. During the addition process, strictly control the intensity and distribution of noise to simulate the diversity and complexity of noise in real seismic data. Add the generated noise data to the labeled shot data constructed in S4 according to a certain weighting ratio.

[0071] S6. For the noisy shot gather data prepared in S5, perform gain addition and normalization operations sequentially. During gain addition, an adaptive gain algorithm is used based on the data's dynamic range and signal characteristics to ensure effective signal enhancement and reasonable noise suppression. Simultaneously, the seismic data is normalized to generate both gain-added and normalized data.

[0072] S7. Construct a deep neural network model. The input to this model is multi-channel features fused from multi-attribute data, including data with gain and standardized data. Through the designed network structure and parameter initialization method, ensure the model's effective learning and feature extraction from complex seismic data. The model's output is seismic trace data after noise probability masking. By optimizing the network structure and activation function, improve the model's output accuracy and reliability.

[0073] Figure 3 To integrate a channel attention mechanism into the U-Net network, which functions similarly to performing principal component analysis (PCA) on the channel feature maps, this mechanism can accurately extract the main components that best reflect noise characteristics, thereby significantly improving the quality of noise removal and optimizing seismic data processing.

[0074] S8. Based on labeled cannon data, rich and diverse training samples are generated through data augmentation and sample partitioning techniques. A hybrid loss function, including mean squared error, structural similarity loss, and sparsity regularization term, is used to optimize the network parameters. During training, optimization algorithms such as stochastic gradient descent are employed, combined with learning rate adjustment strategies, to ensure rapid convergence and stable training of the model.

[0075] S9. By analyzing and learning the features of the input data through the model, noise is accurately removed, resulting in high-quality denoised data. During application, the model's performance and the denoising effect of the data are monitored in real time, and parameters are adjusted and optimized according to the actual situation.

[0076] Figure 6 for Figure 3 The data shown is a single-shot seismic record after noise removal. It can be seen that this technique can completely preserve the effective signal after denoising, greatly improving the quality of seismic data.

[0077] It should be noted that, for the sake of simplicity, the method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, because according to this application, some steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also understand that the embodiments described in the specification are all optional embodiments, and the actions involved are not necessarily required by this application.

[0078] Example 2 Reference Figure 7 This is a structural diagram of an intelligent noise removal device based on multi-feature deep fusion provided in this application. The device includes: The low signal-to-noise ratio shot selection module 210 is used to design an observation system based on a preset geological task. With the excitation point location as the coordinate center, the shot line as the vertical axis, and the receiver line as the horizontal axis, the seismic gather data is uniformly divided into N regions in a rectangular coordinate system, and low signal-to-noise ratio shots with a signal-to-noise ratio lower than the first set value are selected. The noise database construction module 220 is used to determine the data above the first arrival time window as the first type of noise data for each low signal-to-noise ratio shot; and to perform downward translation and sliding processing on the noise data to construct the second type of noise data with the same channel length as the original data; and to integrate and store the first type of noise data and the second type of noise data to form a noise database. The tag-shot construction module 230 is used to select shot collection data with a signal-to-noise ratio higher than a second set value as tag-shot data in different areas of the exploration area according to the complexity of the geological structure and the signal-to-noise ratio of the data, and to perform standardization processing and quality assessment on the tag-shot data. The gun collection data fusion module 240 is used to add noise data from the noise library to the tag gun according to a set ratio weight, form noisy gun collection data through data fusion technology, and add gain and standardize the noisy gun collection data. The model building module 250 is used to construct a deep neural network model with multi-channel features obtained by fusing multi-attribute data as input, wherein the multi-channel features include data with gain and standardized data, and seismic trace data after noise probability masking as output. It effectively learns and extracts features from complex seismic data through a preset network structure and parameter initialization method. Based on labeled shot data, training samples are generated through data augmentation and sample partitioning techniques. A hybrid loss function, including mean squared error, structural similarity loss and sparsity regularization term, is used to optimize the network parameters. During training, a stochastic gradient descent optimization algorithm is used in combination with a learning rate adjustment strategy to ensure rapid convergence and stable training of the model. The noise removal module 260 is used to remove noise by analyzing and learning the features of the input data through a model, so as to obtain high-quality denoised data.

[0079] As described above, this application first utilizes advanced technology to precisely design an observation system based on geological tasks, ensuring the reliability of acquired data through calibration and real-time monitoring. Next, professional algorithms are used to regionally divide and sort the seismic gather data. First-arrival time windows are set using seismic wave theory and simulation software. Then, a self-developed signal-to-noise ratio (SNR) analysis algorithm is used to select low SNR shots and construct a noise database. Simultaneously, considering multiple tools and models and various factors, high SNR shot sets are selected to construct a label database, and noise is dynamically added to the labeled shots according to the SNR ratio to form noisy data. Afterwards, adaptive gain and standardization transformations are applied to the noisy data, and a deep neural network model using a multi-scale fusion U-Net architecture and an autonomous channel attention module is constructed, with multi-attribute fusion features input for learning. The model is trained by optimizing algorithms and dynamically adjusting parameters. Finally, the model and algorithm are used for precise denoising. With the help of real-time monitoring and automatic parameter adjustment, intelligent and adaptive seismic data denoising is achieved, meeting the application needs of different geological conditions.

[0080] As the device embodiment is basically similar to the method embodiment, the description is relatively simple, and relevant parts can be found in the description of the method embodiment.

[0081] Example 3 Optionally, this application also provides an electronic device, including: a processor, a memory, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, it implements the various processes of the above method embodiments and achieves the same technical effect. To avoid repetition, it will not be described again here.

[0082] This application also provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the various processes of the above-described method embodiments and achieves the same technical effects. To avoid repetition, it will not be described again here. The computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, etc.

[0083] Figure 8 This application provides a block diagram of an electronic device 800. For example, the electronic device 800 may be a mobile phone, computer, digital broadcasting terminal, messaging device, game console, tablet device, medical device, fitness equipment, personal digital assistant, etc.

[0084] Reference Figure 8The electronic device 800 may include one or more of the following components: a processing component 802, a memory 804, a power supply component 806, a multimedia component 808, an audio component 810, an input / output (I / O) interface 812, a sensor component 814, and a communication component 816.

[0085] Processing component 802 typically controls the overall operation of electronic device 800, such as operations associated with display, telephone calls, data communication, camera operation, and recording operations. Processing component 802 may include one or more processors 820 to execute instructions to complete all or part of the steps of the methods described above. Furthermore, processing component 802 may include one or more modules to facilitate interaction between processing component 802 and other components. For example, processing component 802 may include a multimedia module to facilitate interaction between multimedia component 808 and processing component 802.

[0086] Memory 804 is configured to store various types of data to support the operation of device 800. Examples of this data include instructions for any application or method operating on electronic device 800, contact data, phonebook data, messages, images, videos, etc. Memory 804 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0087] Power supply component 806 provides power to various components of electronic device 800. Power supply component 806 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power to electronic device 800.

[0088] Multimedia component 808 includes a screen that provides an output interface between the electronic device 800 and the user. In some embodiments, the screen may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen may be implemented as a touchscreen to receive input signals from the user. The touch panel includes one or more touch sensors to sense touches, swipes, and gestures on the touch panel. The touch sensors may sense not only the boundaries of the touch or swipe action but also the duration and pressure associated with the touch or swipe operation. In some embodiments, multimedia component 808 includes a front-facing camera and / or a rear-facing camera. When the device 800 is in an operating mode, such as a shooting mode or a video mode, the front-facing camera and / or the rear-facing camera may receive external multimedia data. Each front-facing camera and rear-facing camera may be a fixed optical lens system or have focal length and optical zoom capabilities.

[0089] Audio component 810 is configured to output and / or input audio signals. For example, audio component 810 includes a microphone (MIC) configured to receive external audio signals when electronic device 800 is in an operating mode, such as call mode, recording mode, and voice recognition mode. The received audio signals may be further stored in memory 804 or transmitted via communication component 816. In some embodiments, audio component 810 also includes a speaker for outputting audio signals.

[0090] I / O interface 812 provides an interface between processing component 802 and peripheral interface modules, such as keyboards, click wheels, buttons, etc. These buttons may include, but are not limited to, home buttons, volume buttons, power buttons, and lock buttons.

[0091] Sensor assembly 814 includes one or more sensors for providing state assessments of various aspects of electronic device 800. For example, sensor assembly 814 may detect the on / off state of device 800, the relative positioning of components such as the display and keypad of electronic device 800, changes in position of electronic device 800 or a component of electronic device 800, the presence or absence of user contact with electronic device 800, orientation or acceleration / deceleration of electronic device 800, and temperature changes of electronic device 800. Sensor assembly 814 may include a proximity sensor configured to detect the presence of nearby objects without any physical contact. Sensor assembly 814 may also include a light sensor, such as a CMOS or CCD image sensor, for use in imaging applications. In some embodiments, sensor assembly 814 may also include an accelerometer, gyroscope, magnetometer, pressure sensor, or temperature sensor.

[0092] Communication component 816 is configured to facilitate wired or wireless communication between electronic device 800 and other devices. Electronic device 800 can access wireless networks based on communication standards, such as WiFi, carrier networks (such as 2G, 3G, 4G, or 5G), or combinations thereof. In one exemplary embodiment, communication component 816 receives broadcast signals or broadcast operation information from an external broadcast management system via a broadcast channel. In one exemplary embodiment, communication component 816 also includes a near-field communication (NFC) module to facilitate short-range communication. For example, the NFC module may be implemented based on radio frequency identification (RFID) technology, Infrared Data Association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology, and other technologies.

[0093] In an exemplary embodiment, the electronic device 800 may be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the methods described above.

[0094] In an exemplary embodiment, a non-transitory computer-readable storage medium including instructions is also provided, such as a memory 804 including instructions, which can be executed by a processor 820 of an electronic device 800 to perform the above-described method. For example, the non-transitory computer-readable storage medium may be a ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, and optical data storage device, etc.

[0095] Example 4 Figure 9 A block diagram of another electronic device 1900 provided for this application. For example, electronic device 1900 may be provided as a server.

[0096] Reference Figure 9 The electronic device 1900 includes a processing component 1922, which further includes one or more processors, and memory resources represented by memory 1932 for storing instructions, such as application programs, that can be executed by the processing component 1922. The application programs stored in memory 1932 may include one or more modules, each corresponding to a set of instructions. Furthermore, the processing component 1922 is configured to execute instructions to perform the methods described above.

[0097] Electronic device 1900 may also include a power supply component 1926 configured to perform power management of electronic device 1900, a wired or wireless network interface 1950 configured to connect electronic device 1900 to a network, and an input / output (I / O) interface 1958. Electronic device 1900 can operate on an operating system stored in memory 1932, such as Windows Server™, Mac OS X™, Unix™, Linux™, FreeBSD™, or similar.

[0098] Example 5 Fifthly, this application discloses a computer program product in which, when the instructions in the computer program product are executed by a processor of an electronic device, the electronic device is enabled to perform the method described in any of the preceding aspects.

[0099] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0100] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0101] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

[0102] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed in this application can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0103] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0104] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0105] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0106] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0107] If the aforementioned functions are implemented as 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 this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, ROM, RAM, magnetic disks, or optical disks.

[0108] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A smart noise removal method based on multi-feature deep fusion, characterized in that, The method includes: Based on the pre-set geological task, the observation system is designed with the excitation point location as the coordinate center, the shot line as the vertical axis, and the receiver line as the horizontal axis. The seismic gather data is evenly divided into N regions in the rectangular coordinate system, and low signal-to-noise ratio shots with a signal-to-noise ratio lower than the first set value are selected. For each low signal-to-noise ratio shot, data above the first arrival time window is identified as the first type of noise data; the noise data is then shifted downward to construct the second type of noise data with the same trace length as the original data; the first type of noise data and the second type of noise data are integrated and stored to form a noise database; Based on the complexity of the geological structure and the signal-to-noise ratio of the data, shot collection data with a signal-to-noise ratio higher than the second set value are selected as tagged shot data in different areas of the exploration area, and the tagged shot data are standardized and quality evaluated. The noise data in the noise library is added to the tagged gun according to a set ratio weight, and the noisy gun set data is formed by data fusion technology. Gain is added to the noisy gun set data and it is then standardized. Using multi-channel features fused from multi-attribute data as input, including data with gain and standardized data, a deep neural network model is constructed with seismic trace data after noise probability masking as output. Through a pre-defined network structure and parameter initialization method, complex seismic data is effectively learned and features are extracted. Training samples are generated based on labeled shot data using data augmentation and sample partitioning techniques. A hybrid loss function, including mean squared error, structural similarity loss, and sparse regularization term, is used to optimize the network parameters. During training, a stochastic gradient descent optimization algorithm combined with a learning rate adjustment strategy is employed to ensure rapid convergence and stable training of the model. By analyzing and learning the features of the input data through the model to remove noise, high-quality data after denoising is obtained.

2. The intelligent noise removal method based on multi-feature deep fusion according to claim 1, characterized in that, The step of selecting a low signal-to-noise ratio (SNR) gun with a SNR lower than a first preset value includes: The signal-to-noise ratio of the data is determined by calculating the mean of the energy ratios of the data before and after the first arrival wave; If the mean ratio of the energy of the data before and after the first arrival wave in the source data is less than 1.0, it is identified as low signal-to-noise ratio data. If the average ratio of the energy of the data before and after the first arrival wave in well shot data is less than 1.5, it is identified as low signal-to-noise ratio data.

3. The intelligent noise removal method based on multi-feature deep fusion according to claim 1, characterized in that, The step of integrating and storing the first type of noise data and the second type of noise data to form a noise library includes: Noise data is constructed by using the same number of sampling points per channel as the length of each channel in the shot gather data and the same sampling interval.

4. The intelligent noise removal method based on multi-feature deep fusion according to claim 1, characterized in that, The step of selecting shot gather data with a signal-to-noise ratio higher than a second set value as tagged shots in different areas of the exploration zone based on the complexity of the geological structure and the signal-to-noise ratio of the data includes: The number of seed shots is 0.5% of the total number of shots, and the higher the signal-to-noise ratio, the fewer seed shots are selected, and the lower the signal-to-noise ratio, the more seed shots are selected.

5. The intelligent noise removal method based on multi-feature deep fusion according to claim 1, characterized in that, The step of adding noise data from the noise database to the tag cannon according to a set weighted ratio includes: The noise data in the noise library is added to the tag shot with a weighted ratio of 1:

1. If the signal-to-noise ratio of the shot data is low, the weighted ratio is increased, and if the signal-to-noise ratio of the shot data is high, the weighted ratio is decreased.

6. The intelligent noise removal method based on multi-feature deep fusion according to claim 1, characterized in that, Also includes: The noisy shot gather data is subjected to normalized data processing.

7. The intelligent noise removal method based on multi-feature deep fusion according to claim 1, characterized in that, The step of constructing a deep neural network model using multi-channel features fused from multi-attribute data as input, wherein the multi-channel features include data with gain and standardized data, and using seismic trace data after noise probability masking as output, includes: A deep neural network model is constructed by adopting a multi-scale fusion U-Net architecture and embedding a channel attention module, using features that are fused at the feature level after parallel branch input as multi-attribute data.

8. The intelligent noise removal method based on multi-feature deep fusion according to claim 1, characterized in that, The step of optimizing network parameters using a hybrid loss function, including mean squared error, structural similarity loss, and sparsity regularization term, includes: An adaptive learning rate algorithm is used to optimize the deep neural network model by dynamically adjusting the batch size and dropout rate based on the signal-to-noise ratio of the validation set.

9. The intelligent noise removal method based on multi-feature deep fusion according to claim 8, characterized in that, Optimize network parameters according to the expression described below: in, , , Weighting coefficients for different loss terms, These are network parameters.

10. An intelligent noise removal device based on multi-feature deep fusion, characterized in that, The device includes: The low signal-to-noise ratio shot selection module is used to design the observation system according to the preset geological task. With the excitation point location as the coordinate center, the shot line as the vertical axis, and the receiver line as the horizontal axis, the seismic gather data is evenly divided into N regions in the rectangular coordinate system, and low signal-to-noise ratio shots with a signal-to-noise ratio lower than the first set value are selected. The noise database construction module is used to determine the data above the first arrival time window as the first type of noise data for each low signal-to-noise ratio shot; and to perform downward translation and sliding processing on the noise data to construct the second type of noise data with the same channel length as the original data; and to integrate and store the first type of noise data and the second type of noise data to form a noise database. The tag-shot construction module is used to select shot collection data with a signal-to-noise ratio higher than a second set value as tag-shot data in different areas of the exploration area based on the complexity of the geological structure and the signal-to-noise ratio of the data, and to perform standardization processing and quality assessment on the tag-shot data. The gun collection data fusion module is used to add noise data from the noise library to the tagged gun according to a set ratio weight, form noisy gun collection data through data fusion technology, and add gain and standardize the noisy gun collection data. The model building module is used to construct a deep neural network model with multi-channel features obtained by fusing multi-attribute data as input, including data with gain and standardized data, and seismic trace data after noise probability masking as output. It effectively learns and extracts features from complex seismic data through a preset network structure and parameter initialization method. Training samples are generated based on labeled shot data through data augmentation and sample partitioning techniques. A hybrid loss function, including mean squared error, structural similarity loss, and sparse regularization term, is used to optimize the network parameters. During training, a stochastic gradient descent optimization algorithm combined with a learning rate adjustment strategy is employed to ensure rapid convergence and stable training of the model. The noise removal module is used to remove noise by analyzing and learning the features of the input data through a model, resulting in high-quality denoised data.

11. An electronic device, characterized in that, include: A processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the method as described in any one of claims 1 to 9.

12. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method as described in any one of claims 1 to 9.

13. A computer program product, characterized in that, When the instructions in the computer program product are executed by the processor of the electronic device, the electronic device implements the method as described in any one of claims 1 to 9.

Images