204 results about "Noisy data" patented technology

The invention discloses an underwater acoustic signal denoising method based on a time-frequency adaptive dual-path Conformer network, and the method comprises the steps: carrying out the preprocessing of real marine environment noise and underwater acoustic target signals, and generating a multi-signal-to-noise-ratio noisy data set; performing short-time Fourier transform on the noisy data, and extracting real part and imaginary part features to form a feature tensor; extracting features by using a feature encoder, and generating intermediate feature representation; respectively extracting a time path feature and a frequency path feature through a time-frequency adaptive dual-path Conformer network; performing multi-scale convolution processing and weighted fusion on the extracted time-frequency features by adopting a multi-scale fusion dynamic gating network; performing nonlinear mapping on the fused features by using a feature decoder to generate a mask matrix; and restoring the complex frequency spectrum based on the mask matrix, and restoring the denoised underwater acoustic time domain signal. According to the method, time-frequency information is fully mined in combination with underwater sound noise characteristics, and the underwater sound signal denoising effect is improved.

The invention discloses a data security sharing method and system, and relates to the technical field of data sharing. The method comprises the steps that a data provider uploads original data, a privacy budget value is calculated through a differential privacy algorithm, noise is added to obtain noisy data, and the noisy data is processed through an anonymous privacy protection algorithm based on maximum dissimilarity degree clustering to obtain desensitized data; encrypting the desensitized data by adopting a block chain decentralization-based ciphertext policy attribute-based encryption method, obtaining an encrypted data ciphertext, uploading the encrypted data ciphertext to a cloud server, and obtaining a content addressing hash value; encrypting the hash value through an elliptic curve encryption algorithm, and storing the encrypted hash value to a block chain account book; and when an access request of a data requester is received, the cloud server verifies the authority by using a non-interactive zero-knowledge proof protocol and returns an encrypted hash value after passing the verification, and the requester decrypts to obtain the hash value and decrypts the encrypted ciphertext to obtain desensitized data, thereby realizing data sharing.

The invention provides an industrial time sequence generation method based on a time sequence decomposition conditional diffusion model. The method comprises the steps that industrial time sequence data are collected and preprocessed to obtain a data set; introducing conditional variables to construct a conditional diffusion model, and injecting noise through forward diffusion; designing a time sequence decomposition and reconstruction UNet module, performing feature extraction on noisy data and conditional variables to obtain an intermediate state, performing time sequence decomposition on the intermediate state, and performing feature reconstruction by using a decoder; the Sinkhorn distance is used as a regularization term to be fused into conditional noise prediction loss to construct a loss function for training; and randomly generating pure Gaussian noise, inputting the pure Gaussian noise into the trained TDA-CDM model, obtaining predicted noise of the current time step, calculating noisy data of the next time step, performing cyclic operation until a time sequence without noise is obtained, and accelerating sampling by using a denoising diffusion implicit model in circulation. According to the method, coexisting multi-scale dynamic features in the complex industrial MTS can be carefully and effectively captured and restored, and the comprehensive quality of generated data is remarkably improved.

The invention discloses a method for suppressing magnetotelluric mixed noise in a shallow water area. The method comprises the following steps: reading a noisy ocean MT time sequence; setting phase space reconstruction parameters; constructing a phase-space vector to obtain a noisy data matrix; noise priori information is obtained through noise pre-estimation; constructing an original noisy data matrix after noise whitening; performing singular value decomposition to obtain a feature vector and a feature value matrix of a data covariance matrix after noise whitening; selecting a low-order principal component after noise adjustment principal component transformation to reconstruct a denoised data matrix; recovering the denoised data matrix into a time sequence; carrying out Fourier transform to obtain an ocean MT magnetic field component amplitude spectrum, and combining to obtain an ocean MT four-component amplitude spectrum; detecting the frequency point of the pulse in the four-component amplitude spectrum; and setting all amplitudes corresponding to pulse frequency points in the four-component amplitude spectrum to be 0, and performing inverse Fourier transform to obtain a processed four-component time sequence. According to the invention, the ocean magnetotelluric signals can be better processed, and the data quality is improved.

The invention discloses a rainfall prediction method based on LightGBM and a variable attention mechanism, and belongs to the technical field of weather prediction. The objective of the invention is to solve the problem of poor prediction effect of an existing model when noise or missing exists in data. According to the method, the LightGBM and the variable attention mechanism are combined, the variable attention mechanism is used for accurately capturing cross-variable dynamic association and multi-scale time dependency in meteorological data, the model can flexibly adapt to the relative dependency between different time steps through relative position coding, and the modeling capacity for long-time-sequence data is improved; by adopting a split weighting mechanism based on LightGBM, different types of exogenous variable data can be weighted, large-scale data can be effectively processed, and feature selection can be carried out, so that the model is helped to pay more attention to variables having great influence on meteorological prediction in the training process, and the robustness of the model to noise data and missing data is enhanced. The method can be applied to rainfall prediction.

Embodiments described herein provide systems and methods for noise-robust contrastive learning. In view of the need for a noise-robust learning system, embodiments described herein provides a contrastive learning mechanism that combats noise by learning robust representations of the noisy data samples. Specifically, the training images are projected into a low-dimensional subspace, and the geometric structure of the subspace is regularized with: (1) a consistency contrastive loss that enforces images with perturbations to have similar embeddings; and (2) a prototypical contrastive loss augmented with a predetermined learning principle, which encourages the embedding for a linearly-interpolated input to have the same linear relationship with respect to the class prototypes. The low-dimensional embeddings are also trained to reconstruct the high-dimensional features, which preserves the learned information and regularizes the classifier.

The invention relates to the technical field of numerical calculation and data analysis, in particular to a data interpolation and fitting analysis system which comprises a data processing module, a depth generation interpolation module, an uncertainty quantification module, an interpretability optimization module, a multi-source heterogeneous data processing module and a hybrid calculation acceleration module. In the prior art, a traditional interpolation method is easy to generate over-fitting or under-fitting in a complex data distribution and high noise scene, and a single deep learning model is insufficient in generalization ability in a small sample or data sparse region; through the dynamic fusion architecture of the depth generation interpolation module, the advantages of a traditional numerical method and deep learning are combined, the weight is automatically adjusted based on data characteristics, the prediction error under noise data is remarkably reduced, the adaptability to complex distribution is improved, and the stability of a data sparse region is enhanced; the problems of insufficient precision and weak generalization ability of a single model of a traditional method are effectively solved.

The invention discloses a seismic strong background noise removal method based on a two-channel network, and relates to the field of geophysical exploration data processing, and the method comprises the steps: selecting a noiseless signal and background noise when the noiseless signal and the background noise are not excited from a single-channel seismic record after a seismic source is excited, carrying out the superposition, generating noisy data, and carrying out the normalization processing; constructing a dual-channel constraint denoising network composed of a global constraint sub-network and a denoising sub-network; optimizing network parameters through the training sample set; and performing global constraint denoising and slice local denoising on the noisy seismic data by using the trained model, and finally splicing to obtain a complete denoising result. According to the method, the splicing effect problem caused by traditional blocking processing is effectively solved, the strong background noise removal effect is remarkably improved, and seamless high-quality denoising with the complete edge is achieved.

The invention discloses a face recognition method based on adaptive sine angle robust principal component analysis, and belongs to the technical field of face recognition, and the method comprises the steps: obtaining a face image, and carrying out the preprocessing of the face image; constructing a robust principal component analysis model based on an adaptive sine angle, and training the robust principal component analysis model by using the training set; solving an optimal projection matrix by using a non-greedy iterative algorithm; mapping the original face features to a discrimination space by using the optimal projection matrix; and performing face recognition by using a nearest neighbor classifier, and performing robustness evaluation on a recognition result. According to the method, the adaptive sine angle is added, abnormal values can be effectively suppressed in the face of noise data, the influence of the abnormal values on the projection direction is reduced, and the method plays a key role in improving robustness.

The invention relates to the field of field groundwater magnetic resonance detection methods, in particular to an augmentation system and method for field groundwater magnetic resonance detection small sample data. Comprising a generation network configured to receive random noise and generate augmented data based on the random noise; the discrimination network comprises a time domain discrimination network and a frequency domain discrimination network; the generative network and the time domain discrimination network form a first discrimination channel, the generative network and the frequency domain discrimination network form a second discrimination channel, and in the second discrimination channel, augmented data generated by the generative network and the noisy data are jointly input into the frequency domain discrimination network; the frequency domain discrimination network is used for discriminating whether the source of input data is a generation network or noisy data in a frequency domain. According to the method, a magnetic resonance detection small sample noisy data set is established, and multi-scale features in an amplitude spectrum and a phase spectrum are captured through the frequency domain discrimination network; the collaborative driving generation network generates a high-quality sample which is close to small sample noisy data in waveform form and spectral characteristics.

The invention discloses a satellite remote sensing image random stripe noise suppression method and system based on low-rank tensor approximation. The method specifically comprises the following steps: establishing a random stripe mixed noise image model; then converting a core task of satellite image denoising into an unconstrained optimization problem, and determining a target function; lRA operation is carried out on a noisy data matrix / tensor for random noise in a satellite image to realize random noise suppression, and stripe noise separation is realized through direction selective regularization by using a stripe noise removal model based on one-way total variation UTV for stripe noise in the satellite image; a UTV-LRTA mixed denoising model is established, and the mixed stripe noise in the satellite image is effectively suppressed through combination of low-tube-rank tensor constraint and one-way total variation regularization. According to the method, the random stripe mixed noise is effectively suppressed, the image quality is improved, and the visual effect is improved, so that the accuracy of information identification and analysis is improved.

The invention discloses a driving decision model optimization method and electronic equipment, and relates to the technical field of automatic driving, and the method comprises the steps: carrying out the denoising processing through a training diffusion model according to first driving perception data and first noisy data, so as to generate a first driving path point sequence of the first driving perception data at a future moment; therefore, abundant and diversified driving tracks are generated by utilizing the strong complex multi-modal distribution modeling capability of the diffusion model; using a first diffusion model to generate a plurality of second driving path point sequences at future moments according to the second driving perception data so as to construct an evaluator training sample, and training an evaluator model to select an optimal driving path point sequence from the plurality of second driving path point sequences, according to the method, the diffusion model and the evaluator model obtained through training are utilized to construct the driving decision-making model, through joint training of the evaluator model and the diffusion model, the decision-making ability of the driving decision-making model for generating candidate driving tracks and selecting the optimal driving track is enhanced, and the safety of automatic driving is improved.

The invention discloses a multi-mode chemical reaction yield prediction method based on adaptive data screening, and belongs to the technical field of chemical synthesis and machine learning crossing. According to the method, a multi-modal input system containing one-dimensional chemical attribute data, a two-dimensional molecular structure map and a three-dimensional spatial configuration is constructed, so that multi-dimensional analysis of molecular interaction is realized; and meanwhile, a multi-stage training process is adopted to simulate a human cognitive rule, so that the model progressively learns layer by layer from a basic reaction feature to a complex reaction mechanism. According to the method, the screening threshold is dynamically adjusted, noise interference is effectively suppressed, and the adaptability to long-tail distribution data is improved. Compared with a traditional single-mode prediction model, the method has the advantages that the prediction accuracy on a noisy data set is improved, the trial and error cost of chemical synthesis experiments is remarkably reduced, and reliable technical support is provided for efficiently screening reaction conditions and accelerating research and development of new compounds.

The invention provides an aeromagnetic data noise reduction method of a noise reduction auto-encoder based on adversarial regularization, and the method comprises the steps: constructing a simulation target signal library, and designing an adversarial regularization noise reduction auto-encoder model structure composed of a noise simulator-encoder-decoder-discriminator structure; inputting the clean signal into a noise simulator to generate noisy data, inputting the noisy data into a decoder-encoder, and outputting a de-noised signal; secondly, the de-noised signal is input into a discriminator, and the discriminator generates a score for the de-noised signal; the reconstruction loss and the adversarial loss form an adversarial loss function; then, the model is trained; and finally, inputting an actually measured signal into the trained model, and carrying out aeromagnetic signal denoising. According to the technical scheme, the technical problems that in the prior art, when an aeromagnetic interference compensation method is used for processing non-linear, time-varying and atypical interference, the compensation precision is low, manual modeling is depended on, and the generalization ability is weak are solved.

The invention discloses a robust target recognition method and device based on noise analysis and a storage medium thereof. The method comprises the following steps: acquiring original noisy data to establish an original data set, screening the original data set based on a K-nearest neighbor method in cooperation with Jensen-Shannon divergence, and respectively establishing a net data subset and a noise data subset; corresponding weight parameters and gradient parameters are obtained in the training process of the network, responses of the net data subset and the noise data subset are calculated, and corresponding sensitivity and contribution degree are obtained; the sensitivity and the contribution degree are fused, and the overall importance score of the parameters is obtained through integration; and improving a loss function in a pruning criterion of Taylor expansion through the overall importance score, and building a pruning model. According to the method, information in the feature space and information in the prediction space are fused, clean samples and noise samples in an original data set are accurately divided, and the reliability of weight parameter evaluation samples in the pruning process of the pruning model is improved.

The invention provides a soil moisture sensor data quality inspection and interpolation method, which comprises the following steps: screening multivariable soil moisture time sequence data based on a preset core physical feature list, carrying out abnormal value detection through physical rule constraint and an isolation forest algorithm, and carrying out data labeling by creating a complete time axis; generating a training sample from the preprocessed data through sliding window sampling, and performing deep feature learning by using a denoising network based on a space-time diffusion probability model; performing interpolation on missing values in the original data by using the trained space-time diffusion probability model, generating a noisy data sample through a forward noise adding process, and performing conditional data interpolation based on a known observation value and a mask matrix in a reverse denoising process to generate a preliminary interpolation result; and performing post-processing correction on the preliminary interpolation result, wherein the post-processing correction comprises clamping correction based on a monthly historical range and correction based on interlayer physical logic.

The invention provides a U-shaped expansion convolutional neural network microseismic signal noise reduction method and system fused with a residual space attention mechanism, and belongs to the field of microseismic signal noise reduction. The problem that an existing noise reduction network model is poor in noise reduction effect and causes effective signal loss is solved. The method comprises the following steps: synthesizing noise-containing microseismic data and clean data, and constructing a training data set; inputting noisy data into the network model, and performing feature extraction through convolution operation in down-sampling; effective signal features are highlighted through a space attention mechanism, and interference noise is suppressed; in up-sampling, restoring the original size of the feature map through deconvolution operation, and fusing deep information and shallow information through a jump connection structure; residual learning is utilized to obtain estimation of a clean micro-seismic signal; and the trained network model is used for noise reduction of actual microseismic signals. According to the method, more effective information of the microseismic data can be reserved, and the noise reduction effect is effectively improved.

The invention provides artificial intelligence-based characteristic guidance method and system for enhancing quality of a diffusion model in generating a data from a noisy data based on a condition information. The method comprises: generating a nonlinear precorrection vector; performing, by a regularization module, a context regularization iteration to obtain an updated nonlinear correction vector and a nonlinear correction gradient; checking if a convergence criterion is met; and denoising the noisy data based on the updated nonlinear correction vector to generate the data. By using the regularization module, the provided characteristic guidance method not only greatly improves the stability of data generation by the diffusion model, but also provides enhanced control over context through two context modes: the detail enhancement mode and the context enhancement mode. The present invention can enhance the semantic characteristics of prompts and mitigate irregularities in image generation.

Noisy data parsed from raw data can be received. The noisy data indicates first transaction events determined to be noise in the raw data. Using the noisy data, a first detection model can be trained to assign anomaly event indicators to second transaction events. The first detection model can receive an anomaly record. The anomaly record can indicate at least a portion of anomalous transaction events identified in runtime data. The first detection mode can assign the anomaly event indicators to the anomalous transaction events. The anomaly event indicators can indicate levels of severity of the anomalous transaction events identified in the runtime data.

The present invention discloses a device (100), a system (200), and a method (300) for analysing a document received from one or more users. The invention includes a system (200) for document analysis. The system (200) comprises a user interface (101) to interact with users. The user interface (101) generates queries and receives documents from users. The documents are then transmitted to a device (100) equipped with processors (102) for analysis. The processors (102) extract key sections from the document, prioritize them, remove noisy data, and generate a summary. An interactive tool (105) within the user interface (101) facilitates user feedback on the analysis. The user feedback is used to update a keyword dictionary (104) at predetermined intervals, allowing the system (200) to continuously improve its document analysis capabilities.

The invention relates to a spatio-temporal data rule extraction method based on spatio-temporal Fourier expert mixing, belongs to the technical field of urban spatio-temporal data processing, solves the problem that global stability characteristics and local disturbance characteristics driven by real physical rules are difficult to effectively describe in the prior art, and comprises the steps that S1, a spatio-temporal data acquisition module acquires historical data information; s2, establishing a space-time Fourier expert hybrid network, and carrying out expert mixing to obtain mixed features; s3, using adaptive group normalization as a conditional fusion module to obtain fusion features; s4, establishing a space-time Fourier attention mechanism module to obtain output features; s5, establishing a diffusion model for performing back diffusion on the noisy data in combination with the noise estimation network to obtain predicted denoised data, and performing training to obtain a trained diffusion model; and S6, performing sampling to obtain spatio-temporal data, inputting the spatio-temporal data into the trained diffusion model, and obtaining an extracted causal law for urban traffic flow prediction.

The present invention discloses a self-label modification method for processing noisy labels. The method randomly selects small batches of data samples, performs data augmentation processing on the data samples to obtain different views, uses them as inputs to a pseudo-twin neural network, and outputs the predicted probability of the data sample category. The prediction calculations of different networks for different views and the JS divergence of the data sample label distribution are used to determine the possibility of them being clean data samples. According to a given judgment threshold, the batch of data samples are divided into clean data samples and noisy data samples. The labels of the clean data samples are only smoothed, and the noisy data samples are dynamically weighted according to the model's predictions and the sample's own labels to give them reliable labels. Finally, the classification loss function and the consistency loss function are used to update the model. The method of the present invention is used to solve image classification tasks under label noise and achieve good performance results.

The application discloses a method for discovering control equations from scarce and noisy data based on physics. The application combines a physical information neural network and a sparse regression method to discover the partial differential control equations of a dynamic system from scarce and noisy data. Firstly, the size of the candidate function library is effectively reduced through a dimension verification method. Then, the powerful nonlinear fitting capability and automatic differentiation characteristics of a deep neural network are used to model a physical system and calculate candidate functions. Finally, the form of the control equation and the coefficients of the equation are obtained through sparse regression, and the coefficients are fine-tuned through a DNN. The method can not only discover control equations from data, but also obtain a network model to realize response prediction of a dynamic system. The method is simple, efficient, high-precision and highly universal, and can be widely applied to physical knowledge mining, modeling and reasoning of complex dynamic systems.

A body fluid movement apparatus includes a body fluid movement apparatus tube with a lumen, a proximal end, a distal end and a balloon coupled to the proximal end. The balloon is configured to be positioned in an interior of a bladder. The proximal end is configured to provide flow of body fluid from the bladder through the lumen, with a draining bag collecting body fluid from the bladder through the lumen. The drainage bag has an inlet port for receiving body fluid and an outlet port for draining body fluid from the drainage bag. The urinary catheter tube includes the proximal end and the proximal end, with a plurality of body fluid draining holes that receive body fluid from the bladder and allow it to be transported to and though the body fluid movement apparatus tube. One or more sensors are positioned in an interior of the catheter tube and are in contact with the patient's urine. The one or more sensors provide sensor data, at least a portion of sensor data being noisy data that contains one or more of errors, outliers, and inconsistencies. Logic resources provide preprocessing of the noisy data to create cleaned sensor data used for one or more of: identification, cleaning, and transforming of noisy data for the machine learning algorithms to produce the cleaned sensor data. An artificial intelligence system coupled to or including an AI database. The AI engine. with a plurality of machine learning algorithms, provide analysis of the cleaned sensor data used for medical monitoring of one or more medical conditions of the patient by the machine learning algorithms, the analysis of the cleaned sensor data being used for the medical monitoring of the patient.

This disclosure describes one or more implementations of systems, non-transitory computer-readable media, and methods that train a named entity recognition (NER) model with noisy training data through a self-cleaning discriminator model. For example, the disclosed systems utilize a self-cleaning guided denoising framework to improve NER learning on noisy training data via a guidance training set. In one or more implementations, the disclosed systems utilize, within the denoising framework, an auxiliary discriminator model to correct noise in the noisy training data while training an NER model through the noisy training data. For example, while training the NER model to predict labels from the noisy training data, the disclosed systems utilize a discriminator model to detect noisy NER labels and reweight the noisy NER labels provided for training in the NER model.

This invention proposes a method for observing ocean waves based on a binocular camera, belonging to the field of image processing technology. The method includes: S1: The binocular camera captures ocean wave images, generating ocean wave point cloud data at the current time t; S2: The ocean wave point cloud data in S1 is converted to ocean wave point cloud data in a geodetic coordinate system; S3: The ocean wave data in S2 undergoes quality screening and optimization, including: selecting the region of interest, quality screening, converting noisy data points to null values, null value filling, and Gaussian filtering; the quality screening classifies the ocean wave data into three levels: excellent, good, and poor; S4: Based on the excellent and good quality ocean wave point cloud data, the wave height, wavelength, and period at that time are calculated; S5: S1-S4 are repeated to obtain multiple frames of ocean wave point cloud data at different times, and the effective wave height, wavelength, period, and wave spectrum are calculated. This method can fill in null values, making the overall data more reasonable and the obtained elevation data more accurate.

The invention belongs to the technical field of biomedical engineering, and discloses a gene regulatory network optimization method based on a diffusion model, which comprises the following steps: acquiring gene data of cells under a steady state condition, and constructing a gene expression matrix according to the gene data; injecting Gaussian noise into the gene expression matrix based on a diffusion model method to generate a series of noisy data sequences; performing noise estimation and structure estimation on the noisy data sequence by a noise estimator and a structure estimator based on a gene regulation and control network, and performing reverse denoising processing according to the noise estimation and the structure estimation to obtain gene structure estimation after reverse denoising; performing structure optimization on the gene structure estimation after reverse denoising by adopting an acyclic constraint function and a regularization substitution method; and outputting the optimized gene structure estimation. According to the method, the regulation and control relation between the genes is accurately recognized from high-dimensional gene expression data, and the modeling precision of the regulation and control relation between the genes is improved.

The invention discloses a novel distributed local fault detection method based on neighborhood preserving embedding-canonical variable analysis, and aims to construct an accurate fault detection model for high-dimensional dynamic data with noise. The core of the method is that data is divided into four subspaces according to Gaussian and non-Gaussian features and dynamic and non-dynamic features. For non-dynamic features, high-dimensional data are projected to a low-dimensional embedding space based on a neighborhood preserving embedding (NPE) dimension reduction technology to reserve a local structure relationship of the data, so that the noise influence is reduced, and the modeling accuracy is improved; for the dynamic features, the time sequence correlation of the data is modeled by using canonical variable analysis (CVA), and the dynamic relationship of the time sequence is captured by constructing a feature matrix, so that the accurate extraction of the features is realized. Secondly, calculating T2 statistic of the data after dimension reduction, and estimating a threshold value of the T2 statistic through kernel density estimation (KDE); in addition, mutual information is used for judging the correlation strength of the subspaces, based on a local outlier factor strategy, the statistics of the subspaces and the cross-correlation information of the subspaces are considered, comprehensive statistics are established, and a statistical threshold value of the statistics is solved. And finally, performing fault detection on the test data according to the threshold value, and visualizing the change of the statistical magnitude. Compared with a traditional method, the method can more effectively deal with high-dimensional data with noise, improves the accuracy and stability of fault detection, and is a better fault detection method.

In various examples, a multi-sensor fusion machine learning model—such as a deep neural network (DNN)—may be deployed to fuse data from a plurality of individual machine learning models. As such, the multi-sensor fusion network may use outputs from a plurality of machine learning models as input to generate a fused output that represents data from fields of view or sensory fields of each of the sensors supplying the machine learning models, while accounting for learned associations between boundary or overlap regions of the various fields of view of the source sensors. In this way, the fused output may be less likely to include duplicate, inaccurate, or noisy data with respect to objects or features in the environment, as the fusion network may be trained to account for multiple instances of a same object appearing in different input representations.

The invention discloses a low-cost imaging sensor noise modeling method and system. According to the method, calibration data are collected and processed, sensor noise is decomposed into time-invariant noise, stripe noise and pixel-level noise components, and modeling calibration is carried out on the time-invariant noise, the stripe noise and the pixel-level noise components respectively; the method comprises the following steps: constructing a neural proxy network taking random noise, sensitivity and exposure time as input conditions, and training the neural proxy network by using a high-precision pixel-level noise sample, so that the neural proxy network can generate synthetic noise adaptive to sensor characteristics and imaging parameters; a random mixing strategy is adopted to fuse synthetic noise generated by the network and real sampling noise, then the synthetic noise and the real sampling noise are superposed with other calibration noise components and signal correlation noise, a complete synthetic noise field is generated, and finally the complete synthetic noise field and a clean image are synthesized into vivid noisy data used for training a denoising network. According to the method, the complex noise characteristics of the low-cost sensor can be accurately described, and the authenticity and diversity of the noise are effectively considered.