Airborne magnetic method data denoising, model generation method, device, medium and program product

Abstract

The application discloses an aviation magnetic method data denoising and model generation method, equipment, medium and program product. The method comprises the following steps: acquiring original two-dimensional magnetic method data collected by an aviation magnetometer, and performing normalization processing on the original two-dimensional magnetic method data to obtain normalized magnetic method data; then, the normalized magnetic method data is cropped into image blocks of a preset size, and a preset number of to-be-covered points are randomly generated in each image block; for each to-be-covered point, the original pixel of the to-be-covered point is replaced by the pixels in the neighborhood of the to-be-covered point to obtain a covered image block, and a covering mark image for marking the position of the to-be-covered point is generated; the covered image block is used as input, and a loss calculation position is limited based on the covering mark image; a lightweight U-net denoising model is trained to obtain a trained lightweight U-net denoising model. The application can improve the denoising processing quality and processing efficiency of aviation magnetic method measurement data.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method, device, medium, and program product for denoising and generating models from airborne magnetic data. Background Technology

[0002] Airborne magnetics is a crucial technique in geophysical exploration, using aircraft equipped with magnetometers to conduct large-scale, high-resolution measurements of the Earth's surface and subsurface magnetic fields. However, due to the complex flight environment and the high sensitivity of the equipment, raw measurement data is often affected by random noise (such as environmental electromagnetic interference and instrument background noise) and system noise (such as flight path changes and instrument drift), leading to degraded data quality and impacting subsequent geological interpretation. Traditional denoising methods, such as linear filtering, empirical mode decomposition, and manually labeled deep learning, all suffer from drawbacks including strong reliance on noise models, high labor costs, slow speed, and limited effectiveness.

[0003] Existing technologies face the following technical challenges in the field of airborne magnetic survey data: 1. Large data scale and high real-time requirements: Airborne magnetic surveys cover a large area and generate massive amounts of data (a single image can reach thousands × thousands), placing high demands on model inference speed and memory consumption. In practical applications, rapid processing is often required. Traditional deep network structures (such as the standard U-Net) have a large number of parameters, slow inference, and high GPU memory consumption, making them unsuitable for direct industrial deployment. 2. Unlabeled and unclean data: Most airborne magnetic surveys only have noisy raw data, lacking "clean" labels. Traditional supervised learning methods cannot be directly trained, and self-supervised mechanisms such as Noise2Void need to adapt to specific data distributions.

[0004] Therefore, there is an urgent need to invent a denoising method for airborne magnetic measurement data to solve the problems of high training cost, low processing efficiency and poor denoising effect of existing denoising models. Summary of the Invention

[0005] In view of this, embodiments of the present invention provide a method, device, medium, and program product for denoising and generating models of airborne magnetic data, which at least partially solves the problems existing in the prior art.

[0006] Other features and advantages of the invention will become apparent from the following detailed description, or may be learned in part by practice of the invention.

[0007] To achieve the above objectives, the embodiments of the present invention provide the following technical solutions:

[0008] According to a first aspect of the present invention, a method for generating a denoising model for airborne magnetic data is provided, the method comprising:

[0009] The raw two-dimensional magnetic data collected by the airborne magnetometer are acquired, and the raw two-dimensional magnetic data are normalized to obtain normalized magnetic data.

[0010] The normalized magnetic data is cropped into image blocks of a preset size, and a preset number of points to be covered are randomly generated for each image block.

[0011] For each point to be covered, the original pixels of the point to be covered are replaced with pixels in the neighborhood of the point to be covered to obtain a covered image block, and a cover mark image is generated based on the point to be covered to mark the position of the point to be covered.

[0012] Using the masked image patch as input and defining the loss calculation location based on the masked marker image, the lightweight U-net denoising model is trained to obtain the trained lightweight U-net denoising model.

[0013] Further, the original two-dimensional magnetic data is normalized to obtain normalized magnetic data, including:

[0014] The original two-dimensional magnetic data is subjected to extreme value clipping based on the preset quantile points to obtain clipped two-dimensional magnetic data.

[0015] The mean and standard deviation of the cropped two-dimensional magnetic data are normalized to obtain normalized magnetic data.

[0016] Furthermore, the normalized magnetic data is cropped into image blocks of a preset size, and for each image block, a preset number of points to be occluded are randomly generated, including:

[0017] From the normalized magnetic data, a starting point is randomly selected, and a preset number of image blocks of a preset size are cropped out.

[0018] For each of the image blocks, a preset number of points to be covered are randomly generated within the image block.

[0019] Furthermore, for each point to be masked, the original pixels of the point to be masked are replaced with pixels in the neighborhood of the point to be masked to obtain a masked image block, and a masking mark image for marking the position of the point to be masked is generated based on the point to be masked, including:

[0020] For each point to be covered, based on the coordinates of the point to be covered, pixels in the neighborhood of the point to be covered are randomly selected to replace the original pixels of the point to be covered, thus obtaining the covered image block.

[0021] A masking marker image is generated based on the coordinates of the point to be masked. The marker value of the masking marker image is 1 at the point to be masked and 0 at the other locations.

[0022] Furthermore, the lightweight U-net denoising model has a total of 1 input channel and 1 output channel.

[0023] The lightweight U-net denoising model includes an encoder and a decoder;

[0024] The encoder includes three sequentially connected downsampling stages, each consisting of a dual convolutional block and a max pooling layer connected in series. The number of output channels of the dual convolutional blocks in the three sequentially connected downsampling stages are 32, 64, and 128, respectively.

[0025] The decoder consists of a transposed convolutional layer, a spatial auto-alignment module, and a dual convolutional block connected in series. The spatial auto-alignment module is used to perform size alignment on the upsampled features and the encoded features before skip connection fusion. The size alignment includes centering cropping and / or automatic padding alignment.

[0026] Furthermore, the lightweight U-net denoising model calculates the loss only at the point to be occluded, and the loss function is: Where i is the pixel position index. Let i be the predicted pixel value corresponding to pixel position index i. Let i be the pixel value before occlusion at pixel position index i. The masking marker value is the value corresponding to pixel position index i, where pixel position index i is the point to be masked. ,otherwise .

[0027] According to a second aspect of the present invention, a method for denoising aeromagnetic data is provided, the method comprising:

[0028] The raw two-dimensional magnetic data collected by the airborne magnetometer are acquired, and the raw two-dimensional magnetic data are normalized to obtain normalized magnetic data.

[0029] The normalized magnetic data is input into the trained lightweight U-net denoising model in the airborne magnetic data denoising model generation method described in any of the above claims, and the denoising result is output.

[0030] The denoising result is subjected to inverse normalization processing to obtain the original physical range of airborne magnetic denoising data, wherein the inverse normalization processing is based on the normalization parameters corresponding to the normalization processing.

[0031] According to a third aspect of the present invention, an aeromagnetic data denoising device is provided, the device comprising: a processor and a memory;

[0032] The memory is used to store one or more program instructions;

[0033] The processor is configured to run one or more program instructions to perform the steps of an airborne magnetic data denoising model generation method as described above or an airborne magnetic data denoising method as described above.

[0034] According to a fourth aspect of the present invention, a computer-readable storage medium is provided, wherein a computer program is stored on the computer-readable storage medium, and the computer program, when executed by a processor, implements the steps of an airborne magnetic data denoising model generation method as described in any of the preceding claims or an airborne magnetic data denoising method as described above.

[0035] According to a fifth aspect of the present invention, a computer program product is provided, the computer program product comprising a computing program stored on a non-transitory computer-readable storage medium, the computer program comprising program instructions that, when executed by a computer, cause the computer to implement the steps of an airborne magnetic data denoising model generation method as described in any of the preceding claims or an airborne magnetic data denoising method as described above.

[0036] This invention provides a method, device, medium, and program product for denoising and model generation of airborne magnetic data. By generating masked image blocks and corresponding masked marker images and training a lightweight U-net denoising model, airborne magnetic data denoising is achieved without the need for clean label data, thereby improving the denoising quality and efficiency of airborne magnetic measurement data. Attached Figure Description

[0037] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary, and those skilled in the art can derive other embodiments based on the provided drawings without creative effort.

[0038] Figure 1 A flowchart illustrating a method for generating a noise reduction model for airborne magnetic data, provided in an embodiment of the present invention;

[0039] Figure 2 This is a flowchart illustrating an airborne magnetic data denoising method provided in an embodiment of the present invention. Detailed Implementation

[0040] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0041] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0042] First, some nouns or terms that appear in the description of the embodiments of this application shall be interpreted as follows:

[0043] U-Net: A convolutional neural network with an encoder-decoder structure, widely used for image segmentation and denoising. Its core principle is to fuse low-level details with high-level semantics through multi-layer feature extraction and upsampling, combined with skip connections.

[0044] Noise2Void (N2V): An unsupervised denoising method that randomly covers some pixels of the input image and replaces them with neighboring pixels. The training model only predicts the original value at the points to be covered, avoiding the model "peeking" at the original pixels and achieving unlabeled training.

[0045] Vectorized sampling: refers to using libraries such as NumPy to generate random mask points and neighborhood replacements in batches, avoiding point-by-point looping and improving efficiency.

[0046] Normalization / denormalization: Scaling the original data to a standard distribution to fit deep learning training, and then restoring it to the original range after inference.

[0047] Automatic Mixed Precision (AMP): A deep learning training acceleration technique that automatically selects the appropriate precision for computation, improving speed while reducing memory usage.

[0048] Figure 1 A flowchart of a method for generating a noise reduction model for airborne magnetic data according to an embodiment of the present invention is shown.

[0049] like Figure 1 As shown, the method for generating a noise reduction model for airborne magnetic data according to an embodiment of the present invention may include steps S100, S200, S300 and S400.

[0050] In step S100, the raw two-dimensional magnetic data collected by the airborne magnetometer is acquired, and the raw two-dimensional magnetic data is normalized to obtain normalized magnetic data.

[0051] Specifically, the above steps include:

[0052] Raw magnetic measurement data is collected using an airborne magnetometer and saved as a two-dimensional floating-point array (e.g., .npy format), with the data shape as (H, W). The input file used in this invention is MagR2_noisy.npy.

[0053] Based on preset quantiles of 2% and 98%, extreme value clipping is performed on the original two-dimensional magnetic data to avoid the influence of outliers on normalization, resulting in clipped two-dimensional magnetic data.

[0054] Then, the mean and standard deviation of the clipped two-dimensional magnetic data are normalized to obtain normalized magnetic data. The normalization formula is as follows: Save the normalization parameters (lo, hi, mean, std) for subsequent inverse normalization.

[0055] In this embodiment of the invention, extreme values ​​and abnormal distributions of magnetic measurement data are standardized after percentile pruning, making the data distribution more suitable for deep learning model training and ensuring the consistency of the physical range of the model output.

[0056] Next, in step S200, the normalized magnetic data is cropped into image blocks of a preset size, and a preset number of points to be covered are randomly generated for each image block.

[0057] Specifically, the above steps include:

[0058] Starting from a random point in the normalized magnetic data, a preset number of image patches of a preset size are cropped out. The size of the image patch is 64×64 and the preset number of samples is 20000.

[0059] For each image block, a preset number (64) of points to be occluded are randomly generated within the image block.

[0060] In step S300, for each point to be covered, the original pixels of the point to be covered are replaced with pixels in the neighborhood of the point to be covered to obtain a covered image block, and a cover mark image for marking the position of the point to be covered is generated based on the point to be covered.

[0061] Specifically, the above steps include:

[0062] For each point to be masked, a pixel is randomly selected as a replacement value within a preset neighborhood of the point to be masked, based on the coordinates of the point. The original pixel value of the point to be masked is then replaced with the replacement value to obtain the masked image block. Preferably, the neighborhood is a four-neighborhood (top, bottom, left, and right), and in some embodiments, an eight-neighborhood can be used. When the point to be masked is located at the boundary of the image block, causing the neighborhood to exceed the boundary, a replacement value is randomly selected only from the set of neighboring pixels that have not exceeded the boundary, or the image block is filled with boundary information before the replacement is performed.

[0063] For each point to be masked, a masking map is generated based on the coordinates of the point to be masked. The masking map has a marker value of 1 at the point to be masked and a marker value of 0 at other locations. The masked image patch and the masking map are then converted into PyTorch tensor format (e.g., (C,H,W)) to facilitate subsequent neural network training.

[0064] This invention utilizes libraries such as NumPy to generate the coordinates of the points to be covered and their neighborhood replacements in a vectorized manner in batches, thereby completing the covering and replacement of multiple pixels at once, improving data construction efficiency and supporting batch training.

[0065] Traditional Noise2Void often uses a point-by-point for loop sampling method to generate the points to be covered and replacement values ​​in some image domains, which has a large computational cost when dealing with large-scale data from aeromagnetic methods.

[0066] To address the aforementioned issues, this invention utilizes libraries such as NumPy to generate the coordinates of points to be masked and their neighboring replacements in a vectorized manner in batches. This allows for the simultaneous masking and replacement of multiple pixels, improving data construction efficiency and supporting large-scale training. This innovation overcomes the bottleneck of large data volumes and high real-time requirements in aeromagnetic data. Furthermore, considering the spatial correlation of aeromagnetic data, this invention innovatively designs a neighborhood random replacement algorithm, making sampling more closely match the distribution of geophysical noise, thereby effectively improving denoising performance.

[0067] Finally, in step S400, the masked image patch is used as input, and the loss calculation position is defined based on the masked marker image to train the lightweight U-net denoising model, thereby obtaining the trained lightweight U-net denoising model.

[0068] Specifically, the above steps include:

[0069] The lightweight U-net denoising model described above has 1 input channel and 1 output channel, with a base number of 32 channels.

[0070] The lightweight U-net denoising model includes an encoder and a decoder:

[0071] The encoder described above contains three sequentially connected downsampling stages. Each downsampling stage consists of a dual convolutional block (two convolutional layers + ReLU) and a max pooling layer connected in series. The number of output channels of the dual convolutional blocks in the three sequentially connected downsampling stages are 32, 64 and 128, respectively.

[0072] The decoder described above consists of a transposed convolutional layer, a spatial auto-alignment module, and a series of dual convolutional blocks, and fuses encoded features through skip connections. The spatial auto-alignment module is used to align the size of the upsampled features and the encoded features before each skip connection fusion. The size alignment includes centering and / or padding.

[0073] In some embodiments, the lightweight U-net denoising model is trained using the Adam optimizer with a learning rate of 1e-3, a batch size of 64, and 5000 training steps. GPU devices and automatic mixed precision (AMP) are enabled to improve training speed and GPU memory utilization.

[0074] The lightweight U-net denoising model described above uses masked mean squared error (masked MSE) as its loss function, calculating the loss only at the points to be masked to avoid overfitting the model to the uncovered regions. The loss function is... Where i is the pixel position index. Let i be the predicted pixel value corresponding to pixel position index i. Let i be the pixel value before occlusion at pixel position index i. The masking marker value is the value corresponding to pixel position index i, where pixel position index i is the point to be masked. ,otherwise .

[0075] During model training, training speed (steps / second) and memory usage are monitored in real time to ensure project availability.

[0076] After training is complete, save the model weight file n2v_unet_small_gpu_fast.pth.

[0077] The embodiments of the present invention employ GPU acceleration and automatic mixed precision training (AMP), which supports GPU training and automatic mixed precision, significantly improving training speed and reducing memory consumption to meet the real-time or near-real-time processing requirements of engineering projects.

[0078] To address the issues of large parameter count, spatial sensitivity of skip connections, and feature loss or memory overflow when directly applied to existing standard U-Nets, this invention employs a lightweight U-Net design: reducing the number of channels and depth in convolutional layers, specifically tailored for two-dimensional aerospace magnetics data, significantly reducing parameter count and memory consumption, and ensuring the processing capability of large-scale data. Furthermore, this invention addresses the inconsistency in patch size after cropping and upsampling in aerospace magnetics data by innovatively designing an automatic centering cropping and padding algorithm. This ensures lossless feature fusion during skip connections, avoiding information loss. This mechanism solves the spatial alignment challenges caused by the variable resolution and patch size of geophysical data.

[0079] In addition, embodiments of the present invention also provide a method for denoising aeromagnetic data. Figure 2 A flowchart of an airborne magnetic data denoising method according to an embodiment of the present invention is shown.

[0080] like Figure 2 As shown, the airborne magnetic data denoising method according to an embodiment of the present invention includes steps S500, S600 and S700.

[0081] In step S500, the raw two-dimensional magnetic data collected by the airborne magnetometer is acquired, and the raw two-dimensional magnetic data is normalized to obtain normalized magnetic data.

[0082] In step S600, the normalized magnetic data is input into the trained lightweight U-net denoising model in the airborne magnetic data denoising model generation method described above for inference, and the denoising result is output.

[0083] In step S700, the denoising result is subjected to inverse normalization processing to obtain the original physical range of airborne magnetic denoised data. The inverse normalization processing is based on the normalization parameters corresponding to the normalization processing. The denoised result file MagR2_denoised_n2v_gpu_fast.npy is then saved.

[0084] Finally, the denoising results are presented in three images: the original input image, the denoised image, and a residual image (input-denoised result) used to visually evaluate the denoising effect. The color range is based on the 2% to 98% quantile of the input data, and the residual is based on the 99th quantile absolute value to ensure a prominent visualization effect.

[0085] In summary, this invention, through a deep understanding of the characteristics, spatial distribution patterns, noise types, and practical engineering requirements of airborne magnetics data, as well as the principles and performance bottlenecks of deep learning algorithms, and after repeated experiments and optimizations, innovatively proposes a denoising method for airborne magnetics measurement data based on a lightweight U-Net neural network and a vectorized Noise2Void self-supervised mechanism. By generating masked image patches and their corresponding masking marker images, and calculating the loss at the masking point locations, model training and denoising inference can be achieved without clean labeled data.

[0086] In some embodiments, lightweight network structures, vectorized masking and replacement, and GPU and automatic mixed-precision training can reduce computational overhead and improve processing efficiency, thus making them suitable for engineering processing scenarios of large-scale aeromagnetic data.

[0087] This invention is not a simple combination of U-Net and Noise2Void, but rather an unsupervised adaptive noise removal mechanism designed for the specific needs of airborne magnetic measurement data. The entire process requires no clean labels and relies solely on the original measurement data for automatic training, adapting to different noise types and effectively improving the convenience and generalization ability of practical applications.

[0088] In addition, embodiments of the present invention also provide an airborne magnetic data denoising device, the device comprising: a processor and a memory; the memory being used to store one or more program instructions; the processor being used to execute one or more program instructions to perform the steps of an airborne magnetic data denoising model generation method or an airborne magnetic data denoising method as described above.

[0089] In addition, embodiments of the present invention also provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the airborne magnetic data denoising model generation method or the airborne magnetic data denoising method as described above.

[0090] In addition, embodiments of the present invention also provide a computer program product, which includes computer program instructions. When the computer program instructions are executed by a processor, they implement the steps of the airborne magnetic data denoising model generation method or the airborne magnetic data denoising method as described above.

[0091] The airborne magnetics data denoising and model generation method, equipment, medium, and program products provided in this invention have the following advantages:

[0092] No labels required, automatic adaptation: Only raw measurement data is needed, no manual labeling or prior noise models are required, and it automatically adapts to various noise types.

[0093] High efficiency and speed: Lightweight network, vectorized sampling and GPU acceleration enable fast training and inference speeds with low memory usage, making it suitable for large-scale engineering applications.

[0094] Robust and reliable: Normalization processing enhances the model's adaptability to extreme data, and the automatic size alignment mechanism ensures lossless feature fusion.

[0095] Visualization and Convenience: The three-image comparison clearly demonstrates the noise reduction effect, facilitating engineer evaluation and application.

[0096] In summary, this invention, addressing the characteristics of noise in airborne magnetics measurement data, innovatively combines a lightweight U-Net network structure, vectorized Noise2Void self-supervised sampling, robust normalization, and GPU-accelerated automatic mixed precision. This proposes an efficient, robust, and easy-to-use unsupervised noise removal method, significantly improving the quality and efficiency of geophysical data processing and demonstrating broad engineering application prospects.

[0097] In this embodiment of the invention, the processor can be an integrated circuit chip with signal processing capabilities. The processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in this embodiment of the invention. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in this embodiment of the invention can be directly implemented by a hardware decoding processor, or implemented by a combination of hardware and software modules in the decoding processor. The software modules can be located in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The processor reads information from the storage medium and, in conjunction with its hardware, completes the steps of the above methods. The storage medium can be memory, for example, volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. Non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. Volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDRSDRAM), Enhanced Synchronous DRAM (ESDRAM), Synchlink DRAM (SLDRAM), and Direct Rambus RAM (DRRAM).The storage media described in the embodiments of this invention are intended to include, but are not limited to, these and any other suitable types of memory. Those skilled in the art will recognize that the functions described in the above examples can be implemented using a combination of hardware and software. When applied software, the corresponding functions can be stored in a computer-readable medium or transmitted as one or more instructions or code on a computer-readable medium. Computer-readable media include computer storage media and communication media, wherein communication media include any medium that facilitates the transmission of computer programs from one place to another. Storage media can be any available medium accessible to general-purpose or special-purpose computers. Although the invention has been described in detail above with general description and specific embodiments, modifications or improvements can be made to it, which will be apparent to those skilled in the art. Therefore, such modifications or improvements made without departing from the spirit of the invention are all within the scope of protection claimed by the invention.

[0098] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Any simple modifications, equivalent changes, or alterations made by those skilled in the art using the disclosed technical content shall fall within the protection scope of the present invention.

Claims

1. A method for generating a denoising model for airborne magnetic data, characterized in that, The method includes: The raw two-dimensional magnetic data collected by the airborne magnetometer are acquired, and the raw two-dimensional magnetic data are normalized to obtain normalized magnetic data. The normalized magnetic data is cropped into image blocks of a preset size, and a preset number of points to be covered are randomly generated for each image block. For each point to be covered, the original pixels of the point to be covered are replaced with pixels in the neighborhood of the point to be covered to obtain a covered image block, and a cover mark image is generated based on the point to be covered to mark the position of the point to be covered. Using the masked image patch as input and defining the loss calculation location based on the masked marker image, the lightweight U-net denoising model is trained to obtain the trained lightweight U-net denoising model.

2. The method for generating a denoising model for airborne magnetic data according to claim 1, characterized in that, The original two-dimensional magnetic data is normalized to obtain normalized magnetic data, including: The original two-dimensional magnetic data is subjected to extreme value clipping based on the preset quantile points to obtain clipped two-dimensional magnetic data. The mean and standard deviation of the cropped two-dimensional magnetic data are normalized to obtain normalized magnetic data.

3. The method for generating a denoising model for airborne magnetic data according to claim 1, characterized in that, The normalized magnetic data is cropped into image blocks of a preset size. For each image block, a preset number of points to be occluded are randomly generated, including: From the normalized magnetic data, a starting point is randomly selected, and a preset number of image blocks of a preset size are cropped out. For each of the image blocks, a preset number of points to be covered are randomly generated within the image block.

4. The method for generating a denoising model for airborne magnetic data according to claim 1, characterized in that, For each point to be masked, the original pixels of the point to be masked are replaced with pixels in the neighborhood of the point to be masked to obtain a masked image block. A masking marker image is then generated based on the points to be masked to mark their locations, including: For each point to be covered, based on the coordinates of the point to be covered, pixels in the neighborhood of the point to be covered are randomly selected to replace the original pixels of the point to be covered, thus obtaining the covered image block. A masking marker image is generated based on the coordinates of the point to be masked. The marker value of the masking marker image is 1 at the point to be masked and 0 at the other locations.

5. The method for generating a denoising model for airborne magnetic data according to claim 1, characterized in that, The lightweight U-net denoising model has a total of 1 input channel and 1 output channel. The lightweight U-net denoising model includes an encoder and a decoder; The encoder includes three sequentially connected downsampling stages, each consisting of a dual convolutional block and a max pooling layer connected in series. The number of output channels of the dual convolutional blocks in the three sequentially connected downsampling stages are 32, 64, and 128, respectively. The decoder consists of a transposed convolutional layer, a spatial auto-alignment module, and a dual convolutional block connected in series. The spatial auto-alignment module is used to perform size alignment on the upsampled features and the encoded features before skip connection fusion. The size alignment includes centering cropping and / or automatic padding alignment.

6. The method for generating a denoising model for airborne magnetic data according to claim 1, characterized in that, The lightweight U-net denoising model calculates the loss only at the points to be occluded, and the loss function is: Where i is the pixel position index. Let i be the predicted pixel value corresponding to pixel position index i. Let i be the pixel value before occlusion at pixel position index i. The masking marker value is the value corresponding to pixel position index i, where pixel position index i is the point to be masked. ,otherwise .

7. A method for denoising airborne magnetic data, characterized in that, The method includes: The raw two-dimensional magnetic data collected by the airborne magnetometer are acquired, and the raw two-dimensional magnetic data are normalized to obtain normalized magnetic data. The normalized magnetic data is input into the trained lightweight U-net denoising model in the airborne magnetic data denoising model generation method as described in any one of claims 1 to 6, and the denoising result is output. The denoising result is subjected to inverse normalization processing to obtain the original physical range of airborne magnetic denoising data, wherein the inverse normalization processing is based on the normalization parameters corresponding to the normalization processing.

8. A noise reduction device for airborne magnetic data, characterized in that, The device includes: a processor and a memory; The memory is used to store one or more program instructions; The processor is configured to run one or more program instructions to perform the steps of the airborne magnetic data denoising model generation method as described in any one of claims 1 to 6 or the airborne magnetic data denoising method as described in claim 7.

9. 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 steps of the airborne magnetic data denoising model generation method as described in any one of claims 1 to 6 or the airborne magnetic data denoising method as described in claim 7.

10. A computer program product, characterized in that, The computer program product includes computer program instructions that, when executed by a processor, implement the steps of the airborne magnetic data denoising model generation method as described in any one of claims 1 to 6 or the airborne magnetic data denoising method as described in claim 7.

Images