Data noise reduction method and device, computer equipment and storage medium

Abstract

The invention discloses a data noise reduction method and device, computer equipment and a storage medium, and the method comprises the steps: carrying out the baseline filtering of motion state sample data, and obtaining the motion state sample data with the baseline drift removed, and the motion state sample data comprises static data and motion data; the static data and the motion data are input into a preset domain adaptation model, domain alignment processing and micro motion estimation processing are carried out respectively to obtain a domain alignment result and a micro motion estimation result, and the domain alignment result comprises static global features and motion global features; determining alignment loss based on the static global feature and the motion global feature; determining an estimation loss based on a micro motion estimation result; and based on the alignment loss and the estimation loss, carrying out iterative updating on the preset field adaptation model. According to the method, the data noise reduction is realized by adopting a multi-stage decoupling strategy, and the noise reduction effect and the model performance are remarkably improved by processing different types of noise in a staged and targeted manner.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular, to a data denoising method, device, computer device, and storage medium. Background Art

[0002] When an inertial sensor (IMU) measures the motion state of an object, it is often affected by external interference and internal errors, resulting in noise. Denoising is required to improve the measurement accuracy. Traditional denoising methods, such as complementary filter algorithms, Kalman filter algorithms, blind source separation, etc., highly depend on task characteristics and specific parameter selection, with poor generalization ability and difficulty in adapting to complex and changeable application scenarios.

[0003] Moreover, the above traditional denoising methods mostly focus on dealing with the inherent mechanical noise of the IMU and are ineffective in dealing with other complex noise sources. In micro-motion perception tasks based on the IMU, such as using an IMU placed on the human body to capture micro-motions, many challenges are faced. On the one hand, the motion signal itself will become a noise source; on the other hand, as Figure 1 shown, the spectra of macroscopic body motion and micro-facial motion in the IMU signal significantly overlap, and the intensity of the human body motion interference signal far exceeds that of the micro-motion feature signal, resulting in the micro-motion feature signal being easily masked, greatly hindering the accurate perception and analysis of micro-motions.

[0004] Although, with the development of deep learning, denoising autoencoders, generative adversarial networks, domain adaptation methods, etc. have been applied to tasks such as IMU denoising. However, most of the existing temporal domain adaptation methods are only applicable to classification tasks. In regression tasks, the aligned sample features lack time dynamic change information and are difficult to effectively handle regression tasks such as micro-motion estimation that have strict requirements for the dynamic characteristics of time series, and cannot meet the needs of practical applications such as micro-motion perception based on the IMU. Summary of the Invention

[0005] Based on this, it is necessary to provide a data denoising method, device, computer device, and storage medium for the above technical problems to solve at least one of the problems existing in the above prior art.

[0006] In a first aspect, a data denoising method is provided, including: Performing baseline filtering on the motion state sample data to obtain motion state sample data with baseline drift removed, where the motion state sample data with baseline drift removed includes stationary data and motion data; Inputting the stationary data and the motion data into a preset domain adaptation model to perform domain alignment processing and micro-motion estimation processing respectively, so as to obtain a domain alignment result and a micro-motion estimation result, where the domain alignment result includes stationary global features and motion global features; Determine an alignment loss based on the static global feature and the motion global feature; Determine an estimation loss based on the micro-motion estimation result; Iteratively update the preset domain adaptation model based on the alignment loss and the estimation loss.

[0007] In a possible implementation, inputting the static data and the motion data into a preset domain adaptation model to perform domain alignment processing and micro-motion estimation processing respectively includes: Extract features from the static data and the motion data respectively to obtain static features and motion features; Perform normalization processing on the static features and the motion features in the time dimension respectively to obtain standard static features and standard motion features; Perform domain alignment on the standard static features and the standard motion features based on class tokens to obtain the static global feature and the motion global feature; Obtain a micro-motion estimation result based on the standard static features and the standard motion features.

[0008] In a possible implementation, extracting features from the static data and the motion data respectively to obtain static features and motion features includes: Slice the static data and the motion data into multiple static time series segments and multiple motion time series segments respectively; Extract at least one target time series segment from multiple static time series segments and multiple motion time series segments respectively according to a preset time stamp label; Extract features from the target time series segments respectively through a preset feature extractor according to a preset extraction rule to obtain the static features and the motion features.

[0009] In a possible implementation, performing normalization processing on the static features and the motion features in the time dimension respectively to obtain standard static features and standard motion features includes: Determine the mean and standard deviation of the static features and the motion features; Perform normalization processing on the static features and the motion features based on the mean and the standard deviation to obtain standard static features and standard motion features.

[0010] In a possible implementation, performing domain alignment on the standard static features and the standard motion features based on class tokens to obtain the static global feature and the motion global feature includes: Add class tokens to the standard static features and standard motion features respectively to obtain static class token features and motion class token features, where the class tokens are used to identify domain features with time-varying dynamics; Input the static class token features and motion class token features into a preset encoder for processing to obtain static global features and motion global features.

[0011] In a possible implementation, obtaining the micro-motion estimation result based on the standard static features and standard motion features includes: Expand the prefix labels in the feature time series corresponding to the standard static features and standard motion features to obtain an expanded label sequence, where the prefix labels refer to the labels corresponding to the prefix motion state data, and the prefix motion state data is the motion state sample data corresponding to the first preset number of features in the feature time series; Input the expanded label sequence into the encoder layer for self-attention calculation; Input the label sequence after self-attention calculation into the decoder layer for feature fusion and decoding processing to obtain the micro-motion estimation result.

[0012] In a possible implementation, after iteratively updating the preset domain adaptation model, it further includes: Obtain the motion state data to be estimated; Perform baseline filtering on the motion state data to be estimated to obtain the motion state data to be estimated with baseline drift removed; Extract features from the motion state data to be estimated with baseline drift removed to obtain the features to be estimated; Perform normalization processing on the features to be estimated in a preset time dimension to obtain standard features to be estimated; Based on the standard features to be estimated, obtain the micro-motion estimation result corresponding to the motion state data to be estimated.

[0013] In a second aspect, a data denoising device is provided, including: A first denoising unit for performing baseline filtering on the motion state sample data to obtain motion state sample data with baseline drift removed, where the motion state sample data with baseline drift removed includes static data and motion data; A second denoising unit for inputting the static data and motion data into a preset domain adaptation model to perform domain alignment processing and micro-motion estimation processing respectively to obtain a domain alignment result and a micro-motion estimation result, where the domain alignment result includes static global features and motion global features; An alignment loss determination unit, configured to determine an alignment loss based on the static global feature and the motion global feature; An estimation loss determination unit, configured to determine an estimation loss based on the micro-motion estimation result; A model iterative update unit, configured to iteratively update the preset domain adaptation model based on the alignment loss and the estimation loss.

[0014] In a third aspect, a computer device is provided, including a memory, a processor, and computer-readable instructions stored in the memory and executable on the processor. When the processor executes the computer-readable instructions, the steps of the data noise reduction method as described above are implemented.

[0015] In a fourth aspect, a readable storage medium is provided. The readable storage medium stores computer-readable instructions, and when the computer-readable instructions are executed by a processor, the steps of the data noise reduction method as described above are implemented.

[0016] The above-mentioned data denoising method, device, computer equipment and storage medium, the method of which is implemented, includes: baseline filtering the motion state sample data to obtain the motion state sample data with baseline drift removed, the motion state sample data with baseline drift removed includes static data and motion data; inputting the static data and motion data into the preset domain adaptation model, performing domain alignment processing and small motion estimation processing respectively, to obtain domain alignment results and small motion estimation results, the domain alignment results include static global features and motion global features; based on the static global features and the motion global features, determining the alignment loss; based on the small motion estimation results, determining the estimated loss; based on the alignment loss and the estimated loss, iteratively updating the preset domain adaptation model. In the embodiment of the present application, a multi-stage decoupling strategy is innovatively adopted to realize data denoising, and the denoising effect and model performance are significantly improved by processing different types of noise in stages and in a targeted manner. First, in the first stage, a baseline filtering mechanism is designed to cope with the low-frequency interference noise caused by external interference. This mechanism can dynamically adjust the baseline according to signal changes, effectively suppress the drift characteristics of low-frequency noise, accurately filter out such noise interference, and lay a pure data foundation for subsequent processing. Secondly, the second stage focuses on the interference noise introduced by human movement. This stage introduces time domain information to deeply explore the changing characteristics of data in the time dimension. By capturing the dynamic change rules of data over a long time span, it can not only comprehensively handle complex and changeable interference noise, but also greatly enhance the model's adaptability to time series data. In complex motion scenes, whether it is fast and violent limb movements or subtle and continuous posture changes, this method can stably and efficiently handle noise, showing excellent robustness, and providing reliable guarantees for tasks such as micro-motion perception based on inertial sensors (IMUs). BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the description of the embodiments of the present application will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0018] Figure 1 is a curve diagram showing a frequency domain aliasing phenomenon between a human motion signal and a facial signal in an embodiment of the prior art; Figure 2 This is a schematic diagram of an overall framework of a data denoising method in an embodiment of the present application; Figure 3 This is a schematic diagram of a network architecture of a domain adaptation model TADA combined with time domain information in one embodiment of the present application; Figure 4 It is a schematic flowchart of a data denoising method in an embodiment of the present application; Figure 5 It is a schematic diagram of a network architecture of a DACT encoder in an embodiment of the present application; Figure 6 It is a schematic diagram of a network architecture of a micro motion estimation module in an embodiment of the present application; Figure 7 It is a schematic structural diagram of a data denoising device in an embodiment of the present application; Figure 8 It is a schematic diagram of a computer device in an embodiment of the present application. Detailed implementation manners

[0019] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0020] The data denoising method provided in this embodiment can be applied to an application environment such as Figure 2 . First, an adaptive baseline correction algorithm is used to process interference signals with drift characteristics. It can dynamically adjust the signal baseline by using baseline filtering, thereby effectively removing low-frequency interference signals and being able to flexibly adjust the baseline according to the actual situation, significantly improving the accuracy. In the second stage, a domain adaptation model TADA that combines time-domain information is constructed. This model can process interference signals within a longer time span and shows stronger robustness in complex motion scenarios. Finally, micro motion estimation is performed on the denoised signal.

[0021] Such as Figure 3As shown, the domain adaptation model TADA for the time-domain information may include a Feature Extractor, a TPN module (temporal segment normalization module), a DACT encoder, and a micro-motion estimation module. Among them, the Feature Extractor is used to extract features from static data and moving data respectively, to obtain the basic feature representation of the motion state sample data. The TPN module is used to receive the features output by the feature extractor, and further process them to capture the position information and variation law of the data in the time dimension, providing time perception characteristics for subsequent processing. The DACT encoder can use the self-attention mechanism to deeply encode the features processed by the TPN module, mine the correlations between different features inside the data, and through the class token, achieve domain alignment between the static domain and the motion domain, enhancing the feature representation ability. At the same time, the micro-motion estimation module can estimate the features processed by the TPN module and output the micro-motion estimation results related to the task, such as micro-motion estimation values, etc.

[0022] Among them, the DACT encoder may include multiple stacked Encoder Blocks, such as 3, a Feed-Forward (feed-forward neural network), and an Attention Block. The Encoder Block is the basic unit of the entire encoder structure, and by stacking multiple such modules, deep encoding processing of the input data is achieved. Inside the Encoder Block, the feed-forward neural network performs non-linear transformation on the input data, in the form of a multi-layer perceptron, enhancing the feature representation ability of the data, enabling the model to learn more complex feature patterns. The Attention Block is the key part of the Encoder Block. It determines the importance of each position in feature calculation by calculating the attention weights between different positions in the input sequence, thereby effectively capturing the long-range dependencies in the sequence, enabling the model to focus on the information more important for the current task.

[0023] Among them, the micro motion estimation module may include Input Embedding (input embedding layer), Transformer Encoder (transformer encoder), Transformer Decoder (transformer decoder), and Linear Projection (linear projection). Input Embedding is used to convert the input features into a low-dimensional dense vector representation suitable for model processing, giving the data an initial feature expression form. The Transformer Encoder processes the features processed by the input embedding layer. It contains a self-attention mechanism inside, which can capture the associations between data features from different subspaces; the feed-forward neural network (Feed Forward) further performs non-linear transformations on the features; residual connection & layer normalization (Add&Norm) is used to stabilize network training, avoiding gradient disappearance or explosion by normalizing the inputs of each layer, thereby extracting deep features of the data. The Transformer Decoder receives the output of the Transformer encoder and processes it in combination with the observed IMU features. Similarly, using the cross-attention mechanism and the feed-forward neural network, a connection is established between the encoder output and the observed features to decode relevant information. Linear Projection maps the feature vector output by the Transformer decoder to the required output dimension through linear transformation to obtain the final micro motion estimation result or other task-related outputs.

[0024] In one embodiment, as Figure 4 shown, a data denoising method is provided, including the following steps: In step S110, baseline filtering is performed on the motion state sample data to obtain motion state sample data with baseline drift removed. The motion state sample data with baseline drift removed includes stationary data and motion data; It should be noted that the motion state sample data refers to IMU data, which is time series data collected by an inertial measurement unit. These data describe the motion state of an object in three-dimensional space, including angular velocity and acceleration. An inertial measurement unit is an inertial sensor that can measure the three-axis acceleration and three-axis angular velocity of an object. The inertial measurement sensor placed on the human body can not only capture micro motions, but also these motion signals themselves constitute noise sources to be eliminated.

[0025] It is understandable that the motion state sample data includes static data and motion data, which are presented in the form of a time series. The static data refers to the motion state data collected by the inertial measurement unit when the measured target such as an object or a human body is in a relatively static state. The motion data refers to data related to motion. It is usually generated by a moving object or a human body, such as the acceleration, velocity, displacement, etc. of an object during its motion recorded by a sensor, or various dynamic information collected by a mobile device during its movement. It should be noted that since the inertial measurement unit usually only collects sensitive data of minute motions, such as hand tremors (which can monitor diseases), such as facial muscle twitches, etc., but in a motion scenario, the signal of a person walking will also be collected by the IMU, and these data will interfere with the above-mentioned minute motion data. Therefore, it is necessary to perform noise reduction processing on the collected data.

[0026] The motion state sample data is the IMU signal originally collected by the IMU sensor, and it often has noise interference (such as noise generated by the sensor's own error and external environmental interference). For example, low-frequency interference noise with a drift characteristic and interference noise introduced for human motion cannot directly meet the requirements of high-precision applications. Since the motion frequency of this type of noise with a drift characteristic is relatively low, much lower than the frequency of minute motions, such as the detachment of the device where the IMU is located. Therefore, for such noise, this application constructs an adaptive baseline correction algorithm to extract the low-frequency baseline component in the IMU signal.

[0027] Specifically, a preset frequency threshold can be set, for example, 0.1 Hz. The signal components below 0.1 Hz are regarded as baseline signals. These baseline signals are mainly generated by low-frequency interference. Therefore, by extracting and removing these low-frequency baseline signals, the originally collected IMU data can be effectively corrected.

[0028] Among them, the specific processing process of the adaptive baseline correction algorithm is as follows: Perform high-pass filtering on the motion state sample data to remove the low-frequency components below the preset frequency threshold. Assume that the motion state sample data is x(t), and the signal after high-pass filtering is . Then its expression is as follows: ; Among them, represents the baseline signal extracted by the low-pass filter. The preset frequency threshold of the low-pass filter can be set to 0.1 Hz to ensure that only the low-frequency components of the motion are extracted. Then the high-pass filtering can be achieved through the following formula: ; Among them, H(s) is the transfer function of the filter, s is the complex frequency variable, c is the cut-off frequency (such as the preset frequency threshold of 0.1 Hz). Thus, the IMU signal can be adaptively corrected to remove low-frequency interference, thereby improving the accuracy of micro-motion perception. This adaptive baseline correction algorithm is not only simple and efficient but also can significantly improve the robustness of micro-motion perception.

[0029] In step S120, the stationary data and the motion data are input into a preset domain adaptation model for domain alignment processing and micro-motion estimation processing respectively to obtain a domain alignment result and a micro-motion estimation result. The domain alignment result includes a stationary global feature and a motion global feature. Specifically, as Figure 2 shown, the signal denoised by the first-stage adaptive baseline correction algorithm can enter the second stage and be denoised again through the domain adaptation model TADA that combines time-domain information. TADA may include a FeatureExtractor (feature extractor), a TPN Module (temporal segment normalization module), a DACT encoder, and a micro-motion estimation module. It should be noted that the preset domain adaptation model may include two branches. One branch is provided with a DACT encoder for aligning the stationary and motion domains, and the other branch is provided with a micro-motion estimation module for micro-motion estimation.

[0030] It should be noted that the specific processing process of this TADA is as follows: The stationary data and the motion data are respectively input into a shared feature extractor. The feature extractor processes the stationary data and the motion data respectively to extract preliminary stationary feature representations and motion feature representations. The stationary feature representation and the motion feature representation are input into the TPN module. Then, the TPN module normalizes the data information in the time dimension to ensure the stability of the subsequent network input, which helps the subsequent model better focus on the dependencies in the time-series data. Class tokens are introduced into the standard stationary features and the standard motion features obtained after the TPN module processes them and are input into the DACT encoder to obtain the corresponding stationary global feature [CLS]s and motion global feature [CLS]m to achieve domain alignment between the stationary domain and the motion domain and improve the generalization ability of the model.

[0031] Meanwhile, the standard stationary features and the standard motion features can be input into the micro-motion estimation module, and the micro-motion estimation module can perform supervised learning based on these features and the true labels.

[0032] In step S130, based on the stationary global feature and the motion global feature, an alignment loss is determined. Specifically, the DACT encoder can obtain the static global feature [CLS]s and the motion global feature [CLS]m based on the standardized static features and standardized motion features after the TPN module's standardization process. Then, based on the static global feature [CLS]s and the motion global feature [CLS]m, it calculates the Lalign (alignment loss), and uses this alignment loss to measure and reduce the difference between the static domain and the motion domain, prompting the model to learn more general and aligned features.

[0033] Among them, the alignment loss can be calculated through the following formula: ; In step S140, based on the micro-motion estimation result, an estimation loss is determined; Specifically, the standardized static features and standardized motion features after the TPN module's standardization process are input into the micro-motion estimation module. The micro-motion estimation module first performs augmentation and embedding processing on the standardized static features and standardized motion features. Augmentation is to make the features contain richer information, and embedding is to convert the features into a vector representation suitable for model processing. Then, the processed features are input into the encoder layer, and the encoder layer uses the self-attention mechanism for attention calculation to capture the correlation information between the features. Finally, through the decoder layer, decoding processing is performed based on a specific decoding algorithm to output the micro-motion estimation result.

[0034] Among them, the micro-motion estimation result may include displacement information, velocity information, motion direction, finger joint angle, or facial muscle movement intensity, etc.

[0035] Then, based on the preset loss function and the true label, the estimation loss can be calculated.

[0036] Among them, the estimation loss can be calculated through the following formula: ; Among them, represents estimation, L is the total length of the time series, h represents the number of prefix sequences, represents the sequence output by the decoder layer, denoted as , since the prefix h is a known condition, the estimation loss starts from m and goes backward.

[0037] In step S150, based on the alignment loss and the estimation loss, the preset domain adaptation model is iteratively updated.

[0038] Specifically, the estimation loss is used to measure the difference between the micro-motion metrics (such as vibration amplitude, motion intensity, etc.) and the actual values, while the alignment loss is used to ensure that the feature distributions in the static domain and the motion domain are as similar as possible, thereby enhancing the generalization ability of the model in different domains. To comprehensively consider the optimization of these two tasks, based on this alignment loss and the estimation loss , the overall loss is obtained, and based on this overall loss, the preset domain adaptation model is iterated to adjust the model weight coefficients until the preset convergence conditions are met. For example, when the overall loss is less than the preset loss threshold or the number of iterations reaches the preset number, the training is completed, and the trained domain adaptation model is obtained. By optimizing the model through the alignment loss and the estimation loss, the importance of estimation and domain feature alignment can be balanced to optimize the model performance.

[0039] Among them, the overall loss can be calculated by the following formula:

[0040] where λ is a hyperparameter used to control the relative importance of the alignment loss of

[0041] In an embodiment of the present application, a data denoising method is provided, comprising: performing baseline filtering on motion state sample data to obtain motion state sample data with baseline drift removed, wherein the motion state sample data with baseline drift removed includes static data and motion data; inputting the static data and motion data into a preset domain adaptation model, performing domain alignment processing and micro-motion estimation processing respectively, so as to obtain domain alignment results and micro-motion estimation results, wherein the domain alignment results include static global features and motion global features; determining alignment loss based on the static global features and the motion global features; determining estimated loss based on the micro-motion estimation results; iteratively updating the preset domain adaptation model based on the alignment loss and the estimated loss. In an embodiment of the present application, a multi-stage decoupling strategy is innovatively adopted to achieve data denoising, and the denoising effect and model performance are significantly improved by processing different types of noise in stages and in a targeted manner. First, in the first stage, a baseline filtering mechanism is designed to cope with low-frequency interference noise caused by external interference. This mechanism can dynamically adjust the baseline according to signal changes, effectively suppress the drift characteristics of low-frequency noise, accurately filter out such noise interference, and lay a pure data foundation for subsequent processing. Secondly, the second stage focuses on the interference noise introduced by human motion. This stage introduces time domain information to deeply explore the changing characteristics of data in the time dimension. By capturing the dynamic change rules of data over a long time span, it can not only comprehensively handle complex and changeable interference noise, but also greatly enhance the model's adaptability to time series data. In complex motion scenes, whether it is fast and violent body movements or subtle and continuous posture changes, this method can stably and efficiently handle noise, showing excellent robustness, and providing reliable guarantees for tasks such as micro-motion perception based on inertial sensors (IMUs).

[0042] In one embodiment of the present application, the static data and the motion data are input into a preset domain adaptation model, and domain alignment processing and micro-motion estimation processing are performed respectively, including: Extracting features from the static data and the motion data respectively to obtain static features and motion features; Standardizing the static features and the motion features in the time dimension respectively to obtain standard static features and standard motion features; Based on the category token, domain aligning the standard static features and the standard motion features to obtain the static global features and the motion global features; Based on the standard static features and the standard motion features, a micro-motion estimation result is obtained.

[0043] Specifically, the signals denoised by the first-stage adaptive baseline correction algorithm remove the low-frequency interference noise with drift characteristics. Then, in the second stage, the domain adaptation model TADA that combines time-domain information removes the interference noise introduced by the motion of objects or human bodies to be measured. By processing different types of noise in stages and in a targeted manner, the denoising effect and model performance are significantly improved.

[0044] It should be noted that the domain adaptation model TADA for time-domain information in the training stage may include a FeatureExtractor (feature extractor), a TPN module (temporal segment normalization module), a DACT encoder, and a micro-motion estimation module.

[0045] Among them, the specific processing process of TADA is as follows: The static data and the motion data are respectively input into the shared feature extractor, and the feature extractor processes the original data to extract the preliminary static feature representation and the motion feature representation. The static feature representation and the motion feature representation are input into the TPN module. The TPN module normalizes the data information from the time dimension to ensure the stability of the subsequent network input, which helps the subsequent model to better focus on the dependencies in the time series data. The features with time-aware characteristics enter the DACT encoder. The DACT encoder uses the self-attention mechanism to establish associations between different feature dimensions and different time steps, deeply encodes the features, and obtains a more representative and discriminative feature representation. At the same time, after the static data and the motion data are respectively processed by their own DACT encoders, the corresponding and , that is, the global feature representation. Based on the obtained and , the alignment loss is calculated. Through this alignment loss, the differences between the static domain and the motion domain are measured and reduced, prompting the model to learn more general and aligned features and improving the model generalization ability. Normalizing the time series from the time dimension before domain alignment helps to improve the stationarity of the domain alignment input and helps the subsequent model to better capture the dependencies in the time series data.

[0046] At the same time, the features processed by the TPN module can also enter the micro-motion estimation module. The micro-motion estimation module performs micro-motion estimation based on these features and outputs the micro-motion estimation result. Then, based on the micro-motion estimation result and the true label, the estimation loss is calculated. This estimation loss can measure the gap between the estimation result and the true value and is used to guide the parameter update of the model in the estimation task.

[0047] Finally, according to the calculated and , the parameters in the model are adjusted and optimized through the backpropagation algorithm, and continuous iterative training is carried out to gradually improve the performance of the TADA model in domain feature alignment and estimation tasks.

[0048] In an embodiment of the present application, the step of respectively extracting features from the static data and the motion data to obtain static features and motion features includes: Respectively segment the static data and the motion data into multiple static time series segments and motion time series segments; According to a preset timestamp label, extract at least one target time series segment from multiple static time series segments and multiple motion time series segments respectively; According to a preset extraction rule, respectively extract features from the target time series segments through a preset feature extractor to obtain the static features and the motion features.

[0049] Specifically, the static data and the motion data can be respectively input into a shared feature extractor, and after convolution, pooling, and activation processing, static features and motion features are respectively obtained. The specific process of feature extraction is as follows: since the motion state sample data is continuously collected by the sensor, the static data and the motion data are time series signals. The time series signal is segmented into multiple time series segments, and then the corresponding sequence segments are selected and input into the feature extractor for feature extraction. The feature extractor is composed of a multi-layer convolutional neural network (CNN), for example, consisting of three layers. The time series segment enters the convolutional layer, and the convolutional layer may include multiple convolutional kernels. The convolutional kernels move point by point in the time dimension and perform convolutional operations with the corresponding data. One convolutional operation can obtain a feature map, and multiple convolutional kernels can obtain multiple feature maps. Each feature map represents a local feature representation of the data. In order to reduce the amount of data and complexity, the data processed by the convolutional layer can be pooled through the pooling layer to reduce the data dimension. Finally, activation processing can be performed through an activation function (such as ReLU, Sigmoid, etc.) to extract the static features and the motion features.

[0050] Exemplarily, for a time series signal with a length of T, x = , , …, , extract the sequence segment centered on the label timestamp, and the window size is ω. Given the label timestamp T = { , , …, }, then the corresponding sequence segment is , where . Then input the into the feature extractor for feature extraction.

[0051] It should be noted that, in order to reduce the number of parameters and improve the generalization ability of the model, the same feature extractor can be applied to different sequence segments, that is, share parameters. The shared parameters are represented as: ; wherein, is the feature representation of the sequence segment obtained through the feature extractor, and θ represents the shared parameters of the feature extractor.

[0052] In an embodiment of the present application, the static features and the motion features are respectively normalized in the time dimension to obtain standard static features and standard motion features, including: Determine the mean and standard deviation of the static features and the motion features; Based on the mean and the standard deviation, normalize the static features and the motion features to obtain standard static features and standard motion features.

[0053] Specifically, although the shared parameter mechanism of the feature extractor can ensure that the meanings of each sequence segment channel are consistent, the feature values themselves are not restricted. Therefore, in order to ensure the stability of the subsequent network input, the static features and the motion features extracted by the feature extractor can be normalized through the TPN module (time slice normalization module).

[0054] The feature representation extracted by the feature extractor is , where B is the batch size, C is the number of channels, L is the length of the time series, and the TPN module can perform normalization in the L dimension. In order to perform TPN on H, the mean μ and standard deviation σ of the features in the time dimension can be calculated first. The mean μ and standard deviation σ can be calculated respectively through the following formulas: ; ; wherein, is a small constant used to prevent division by zero.

[0055] Then, normalize the feature H and introduce the learnable parameters and for scaling and translation, then the following calculation formula can be obtained: ; wherein, the normalized feature representation is , which will be used as the input of the subsequent network.

[0056] In an embodiment of the present application, the domain alignment of the standard static features and the standard motion features based on the class token to obtain the static global feature and the motion global feature includes: Add class tokens to the standard static features and the standard motion features respectively to obtain static class token features and motion class token features, where the class token is used to identify the domain features of time dynamic changes; Input the static class token features and the motion class token features into a preset encoder for processing to obtain the static global feature and the motion global feature.

[0057] It should be noted that time dynamic changes can effectively reflect the time dependence of data, while the samples in traditional domain adaptation lack this time dynamic pattern. Therefore, class tokens (CLS Tokens) are introduced into the standard static features and the standard motion features. Through the CLS Token, the domain features of time dynamic changes can be identified, so as to align the features in this dimension. By capturing the features of data changing over time, the robustness of the model in complex motion scenarios is improved.

[0058] Then, input the input time series data (normalized features) with the [CLS] tag into the DACT encoder, as Figure 5As shown, the DACT encoder may include multiple stacked Encoder Blocks (encoder modules), such as three, a Feed-Forward (feed-forward neural network), and an Attention Block (attention module). In the Attention Block part of each Encoder Block, attention calculations can be performed on the input sequence. Specifically, by calculating the attention scores between different positions in the sequence, an attention weight matrix is obtained. This weight matrix represents the importance of each position in the sequence to other positions. Then, based on this weight matrix, the input sequence is weighted and summed to highlight the information at important positions and suppress unimportant information, thereby capturing long-range dependencies and key features in the sequence. The feature representation after being processed by the attention module is input into the Feed-Forward (feed-forward neural network). The feed-forward neural network performs non-linear transformations on these features. Through the calculations of multiple layers of neurons, features are further extracted and transformed, enhancing the expressive power of the features and learning more complex feature patterns. The output after being processed by one Encoder Block is used as the input for the next Encoder Block, repeating the above steps of attention calculation and feed-forward neural network processing. Through the stacking of multiple Encoder Blocks, deep feature extraction and encoding can be performed on the input data, continuously enhancing the richness and accuracy of the feature representation. After being processed by all Encoder Blocks, a feature representation containing the global information of the entire input sequence is finally obtained. The feature representation at the [CLS] position aggregates the global semantics of the sequence and can be used for subsequent tasks, such as serving as the basis for classification in classification tasks or playing a role in other tasks that require global features.

[0059] Specifically, a CLS Token can be added in front of the extracted sequence features, and then the sequence representation can be obtained. . By adding CLS Tokens to the standard static features and the standard motion features respectively, the marked static features and the marked motion features can be obtained. After introducing the class token CLS Token, in order to further process the data with the [CLS] mark, it is input into the DACT encoder. The DACT encoder performs attention calculations on the input features through the self-attention mechanism, captures the long-range dependencies and key information in the sequence, then performs non-linear transformations to enhance the feature expression ability, and finally can perform normalization and residual connection. After being processed by multiple layers of encoders, the added classification mark [CLS] integrates the information of the entire static feature sequence or motion feature sequence. At this time, the feature vector corresponding to [CLS] represents the global feature of the sequence, which are respectively represented as and , where and , D represents the feature dimension. For example, the vector of the static feature sequence contains global information of the static state, and the vector of the motion feature sequence contains global information of the motion state, which can be used for subsequent classification, estimation and other tasks.

[0060] In an embodiment of the present application, based on the standard static feature and the standard motion feature, a micro-motion estimation result is obtained, including: Expanding the prefix labels in the feature time series corresponding to the standard static feature and the standard motion feature to obtain an expanded label sequence, where the prefix label refers to the label corresponding to the prefix motion state data, and the prefix motion state data is the motion state sample data corresponding to the first preset number of features in the feature time series; Inputting the expanded label sequence into the encoder layer for self-attention calculation; Inputting the label sequence after self-attention calculation into the decoder layer for feature fusion and decoding processing to obtain the micro-motion estimation result; where the micro-motion estimation result may include displacement information, velocity information, motion direction, finger joint angle, or facial muscle movement intensity, etc.

[0061] Specifically, the standard static feature and the standard motion feature can be input into a micro-motion estimation module for estimation processing. The micro-motion estimation module may include Input Embedding, Transformer Encoder, and Transformer Decoder. The input embedding is used to convert the input features into a low-dimensional dense vector representation suitable for model processing through a specific mapping function, giving the data an initial feature expression form. The encoder processes the features after input embedding through a self-attention mechanism, a feed-forward neural network, and residual connection & layer normalization operations to extract deep features. The decoder also processes the deep features output by the encoder through a cross-attention mechanism and a feed-forward neural network to decode information related to specific tasks. The micro-motion estimation module regresses the IMU micro-motion features to the output metrics of specific tasks (such as finger joint angle, facial muscle movement intensity, etc.). Through the micro-motion estimation module, the model can learn the mapping change between the label and the IMU from the prefix information to better estimate the micro-motion information.

[0062] As Figure 6 shown, assuming that the feature representation output by the TPN module is , where the first m features come from the prefix motion state data, L is the total length of the time series, and Lm is the length of the observed feature. In order to avoid the serious exposure bias problem caused by the autoregressive training strategy, the use of the lower triangle mask for regression training is cancelled. It should be noted that the motion state sample data is the data continuously collected by the sensor, so the features of the IMU data observed at the current time can be used as the observed features, and the IMU data collected before the current time can be used as the prefix motion state data, and the corresponding features are the prefix IMU features.

[0063] The prefix label can be extended to the total length by zero padding. The extended label is represented as . Make the label length consistent with the total length of the feature time series.

[0064] These expanded labels can then be used as input to the encoder layer, so that the labels and features match in dimension and length. The encoder layer includes a self-attention mechanism that performs self-attention calculations on the input features, allowing the model to learn the association between the prefix IMU features and the prefix labels. The cross-attention mechanism allows the decoder to focus on the features output by the encoder. In this case, it allows the model to learn how the prefix IMU features correspond to the labels and explore the mapping relationship. Finally, the decoder layer can output a time series of estimated metrics ,in The part represents the estimation result. This means that the model outputs an estimated value of a time series through learning and processing. This estimated value corresponds to the output indicator of a specific task (such as the estimated value of micro-motion, etc.), which can be used for subsequent analysis and application.

[0065] It should be noted that the IMU length can be set to 200 sampling points, and the label time series length L of the output indicator can be set to the corresponding number of sampling points according to the actual task requirements, for example, it can be set to 60 sampling points, and the prefix time series length m can be set to 15 sampling points. It should be noted that although L is 60 (corresponding to 2 seconds), this method can still maintain a continuous high frame rate output by overlapping sliding windows. In model training, the Adam optimizer was used with a learning rate of 1e-4 and a batchsize of 32. The loss weight parameter λ was set to 10. In addition, the OneCycleLR scheduling strategy was applied to adjust the learning rate. In the micro-motion estimation module, the number of Transformer layers can be set to 6, the number of heads can be set to 8, and the hidden dimension size can be set to 256.

[0066] In an implementation of the present application, after the iterative update of the preset domain adaptation model, the method further includes: Acquiring motion state data to be estimated; Performing baseline filtering on the motion state data to be estimated to obtain the motion state data to be estimated with baseline drift removed; Performing feature extraction on the motion state data to be estimated with the baseline drift removed to obtain features to be estimated; Standardizing the features to be estimated in a preset time dimension to obtain standard features to be estimated; Based on the standard feature to be estimated, a micro-motion estimation result corresponding to the motion state data to be estimated is obtained.

[0067] It should be noted that when performing reasoning through the trained domain adaptation model TADA, there is no need for domain alignment. Therefore, the TADA model during reasoning may include Feature Extractor, TPN Module (time segment normalization module) and micro-motion estimation module.

[0068] After the motion state data to be estimated is subjected to baseline filtering, the motion state data to be estimated with baseline drift removed can be obtained, and low-frequency noise filtering can be achieved. Then, it is input into the feature extractor, and the feature extractor performs feature extraction on the motion state data to be estimated with baseline drift removed, and extracts preliminary features to be estimated. The features to be estimated are input into the TPN module. The TPN module standardizes data information from the time dimension to ensure the stability of subsequent network inputs, which helps subsequent models to better focus on dependencies in time series data. The features with time-aware characteristics processed by the TPN module enter the micro-motion estimation module, which performs micro-motion estimation based on these features and outputs micro-motion estimation results.

[0069] In the embodiment of the present application, a multi-stage decoupling strategy is innovatively adopted to achieve data denoising, and different types of noise are processed in stages and in a targeted manner, which significantly improves the denoising effect and model performance. First, in the first stage, a baseline filtering mechanism is designed to deal with low-frequency interference noise caused by external interference. This mechanism can dynamically adjust the baseline according to signal changes, effectively suppress the drift characteristics of low-frequency noise, accurately filter out such noise interference, and lay a pure data foundation for subsequent processing. In the second stage, the interference noise introduced by human motion is focused. In this stage, time domain information is introduced to deeply explore the changing characteristics of data in the time dimension. By capturing the dynamic change laws of data over a long time span, it can not only comprehensively handle complex and changeable interference noise, but also greatly enhance the model's adaptability to time series data. In complex motion scenes, whether it is fast and violent body movements or subtle and continuous posture changes, this method can stably and efficiently handle noise, showing excellent robustness, and providing reliable guarantees for tasks such as micro-motion perception based on inertial sensors (IMUs).

[0070] It should be understood that the size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0071] In one embodiment, a data denoising device is provided, which corresponds one-to-one to the data denoising method in the above embodiment. Figure 7 As shown, the data denoising device includes a first denoising unit 10, a second denoising unit 20, an alignment loss determining unit 30, an estimation loss determining unit 40 and a model iteration updating unit 50. Each functional module is described in detail as follows: A first noise reduction unit 10 is used to perform baseline filtering on the motion state sample data to obtain the motion state sample data with baseline drift removed, wherein the motion state sample data with baseline drift removed includes static data and motion data; A second denoising unit 20 is used to input the static data and the motion data into a preset domain adaptation model, and perform domain alignment processing and small motion estimation processing respectively to obtain a domain alignment result and a small motion estimation result, wherein the domain alignment result includes a static global feature and a motion global feature; an alignment loss determining unit 30, configured to determine an alignment loss based on the static global feature and the motion global feature; an estimated loss determining unit 40, configured to determine an estimated loss based on the micro-motion estimation result; The model iterative updating unit 50 is used to iteratively update the preset domain adaptation model based on the alignment loss and the estimation loss.

[0072] In one embodiment of the present application, the second noise reduction unit 20 is further used for: Extracting features from the static data and the motion data respectively to obtain static features and motion features; Standardizing the static features and the motion features in the time dimension respectively to obtain standard static features and standard motion features; Based on the category token, domain aligning the standard static features and the standard motion features to obtain the static global features and the motion global features; Based on the standard static features and the standard motion features, a micro-motion estimation result is obtained.

[0073] In one embodiment of the present application, the second noise reduction unit 20 is further used for: The static data and the motion data are divided into a plurality of static time series segments and a plurality of motion time series segments respectively; According to a preset timestamp label, extract at least one target time series segment from the plurality of static time series segments and the plurality of moving time series segments respectively; According to the preset extraction rules, the target time series segments are respectively subjected to feature extraction by a preset feature extractor to obtain the static features and the motion features.

[0074] In one embodiment of the present application, the second noise reduction unit 20 is further used for: Determining the mean and standard deviation of the static features and the motion features; Based on the mean value and the standard deviation, the static feature and the motion feature are standardized to obtain a standard static feature and a standard motion feature.

[0075] In one embodiment of the present application, the second noise reduction unit 20 is further used for: Adding category tokens to the standard static feature and the standard motion feature respectively to obtain a static category token feature and a motion category token feature, wherein the category token is used to identify a domain feature that changes dynamically over time; The stationary category token features and the motion category token features are input into a preset encoder for processing to obtain a stationary global feature and a motion global feature.

[0076] In one embodiment of the present application, the second noise reduction unit 20 is further used for: Expanding the prefix labels in the feature time series corresponding to the standard static features and the standard motion features to obtain an expanded label sequence, wherein the prefix label refers to the label corresponding to the prefix motion state data, and the prefix motion state data is the motion state sample data corresponding to the first preset features in the feature time series; Inputting the expanded label sequence into the encoder layer for self-attention calculation; The label sequence calculated after self-attention is input into the decoder layer for feature fusion and decoding processing to obtain the micro-motion estimation result.

[0077] In one embodiment of the present application, the device further includes an actual prediction unit, which is used to: After the preset domain adaptation model is iteratively updated, the method further includes: Acquiring motion state data to be estimated; Performing baseline filtering on the motion state data to be estimated to obtain the motion state data to be estimated with baseline drift removed; Performing feature extraction on the motion state data to be estimated with the baseline drift removed to obtain features to be estimated; Standardizing the features to be estimated in a preset time dimension to obtain standard features to be estimated; Based on the standard feature to be estimated, a micro-motion estimation result corresponding to the motion state data to be estimated is obtained.

[0078] In the embodiment of the present application, a multi-stage decoupling strategy is innovatively adopted to achieve data denoising, and different types of noise are processed in stages and in a targeted manner, which significantly improves the denoising effect and model performance. In the first stage, a baseline filtering mechanism is designed to deal with low-frequency interference noise caused by external interference. This mechanism can dynamically adjust the baseline according to signal changes, effectively suppress the drift characteristics of low-frequency noise, accurately filter out such noise interference, and lay a pure data foundation for subsequent processing. In the second stage, the interference noise introduced by human body movement is focused. In this stage, time domain information is introduced to deeply explore the changing characteristics of data in the time dimension. By capturing the dynamic change rules of data over a long time span, it can not only comprehensively handle complex and changeable interference noise, but also greatly enhance the model's adaptability to time series data. In complex motion scenes, whether it is fast and violent body movements or subtle and continuous posture changes, this method can stably and efficiently handle noise, showing excellent robustness, and providing reliable guarantees for tasks such as micro-motion perception based on inertial sensors (IMU).

[0079] For the specific definition of the data denoising device, please refer to the definition of the data denoising method above, which will not be repeated here. Each module in the above-mentioned data denoising device can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0080] In one embodiment, a computer device is provided. The computer device may be a terminal device, and its internal structure diagram may be as follows: Figure 8 As shown. The computer device includes a processor, a memory, and a network interface connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a readable storage medium. The readable storage medium stores computer-readable instructions. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer-readable instructions are executed by the processor, a data noise reduction method is implemented. The readable storage medium provided in this embodiment includes a non-volatile readable storage medium and a volatile readable storage medium.

[0081] In an embodiment of the present application, a computer device is provided, including a memory, a processor, and computer-readable instructions stored in the memory and executable on the processor. When the processor executes the computer-readable instructions, the steps of the data denoising method described above are implemented.

[0082] In an application embodiment, a readable storage medium is provided. The readable storage medium stores computer-readable instructions, and when the computer-readable instructions are executed by a processor, the steps of the data noise reduction method as described above are implemented.

[0083] Those of ordinary skill in the art can understand that all or part of the processes in the above-described embodiment methods can be completed by instructing relevant hardware through computer-readable instructions. The computer-readable instructions can be stored in a non-volatile readable storage medium or a volatile readable storage medium. When the computer-readable instructions are executed, they can include the processes of the embodiments of the above various methods. Among them, any reference to a memory, storage, database, or other medium used in the various embodiments provided in the present application can include non-volatile and / or volatile memories. Non-volatile memories can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memories can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and Rambus dynamic RAM (RDRAM), etc.

[0084] Those skilled in the art can clearly understand that for the convenience and brevity of description, only the above-mentioned division of each functional unit and module is used as an example. In actual applications, the above functions can be allocated to different functional units and modules as needed, that is, the internal structure of the device is divided into different functional units or modules to complete all or part of the functions described above.

[0085] The above embodiments are only used to illustrate the technical solutions of the present application, and are not intended to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the various embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A data denoising method, characterized in that: The method comprises: Performing baseline filtering on the motion state sample data to obtain motion state sample data with baseline drift removed, wherein the motion state sample data with baseline drift removed includes static data and motion data; Inputting the static data and the motion data into a preset domain adaptation model, performing domain alignment processing and small motion estimation processing respectively, so as to obtain a domain alignment result and a small motion estimation result, wherein the domain alignment result includes a static global feature and a motion global feature; Determining an alignment loss based on the static global feature and the motion global feature; Determining an estimated loss based on the micro-motion estimation result; The preset domain adaptation model is iteratively updated based on the alignment loss and the estimation loss.

2. The data denoising method according to claim 1, characterized in that: The step of inputting the static data and the motion data into a preset domain adaptation model and performing domain alignment processing and micro-motion estimation processing respectively includes: Extracting features from the static data and the motion data respectively to obtain static features and motion features; Standardizing the static features and the motion features in the time dimension respectively to obtain standard static features and standard motion features; Based on the category token, domain aligning the standard static features and the standard motion features to obtain the static global features and the motion global features; The micro-motion estimation result is predicted based on the standard static feature and the standard motion feature.

3. The data denoising method according to claim 2, characterized in that: The extracting features of the static data and the motion data respectively to obtain static features and motion features includes: The static data and the motion data are divided into a plurality of static time series segments and a plurality of motion time series segments respectively; According to a preset timestamp label, extract at least one target time series segment from the plurality of static time series segments and the plurality of moving time series segments respectively; According to the preset extraction rules, the target time series segments are respectively subjected to feature extraction by a preset feature extractor to obtain the static features and the motion features.

4. The data denoising method according to claim 3, wherein: The standardizing the static features and the motion features in the time dimension to obtain standard static features and standard motion features includes: Determining the mean and standard deviation of the static features and the motion features; Based on the mean value and the standard deviation, the static feature and the motion feature are standardized to obtain a standard static feature and a standard motion feature.

5. The data denoising method according to claim 2, wherein: The domain alignment of the standard static features and the standard motion features based on the category token to obtain the static global features and the motion global features includes: Adding category tokens to the standard static feature and the standard motion feature respectively to obtain a static category token feature and a motion category token feature, wherein the category token is used to identify a domain feature that changes dynamically over time; The stationary category token features and the motion category token features are input into a preset encoder for processing to obtain the stationary global features and the motion global features.

6. The data denoising method according to claim 2, wherein: The predicting and obtaining the micro-motion estimation result based on the standard static feature and the standard motion feature includes: Expanding the prefix labels in the feature time series corresponding to the standard static features and the standard motion features to obtain an expanded label sequence, wherein the prefix label refers to the label corresponding to the prefix motion state data, and the prefix motion state data is the motion state sample data corresponding to the first preset features in the feature time series; Inputting the expanded label sequence into the encoder layer for self-attention calculation; The label sequence calculated after self-attention is input into the decoder layer for feature fusion and decoding processing to obtain the micro-motion estimation result.

7. The data denoising method according to any one of claims 1 to 6, characterized in that: After the preset domain adaptation model is iteratively updated, the method further includes: Acquiring motion state data to be estimated; Performing baseline filtering on the motion state data to be estimated to obtain the motion state data to be estimated with baseline drift removed; Performing feature extraction on the motion state data to be estimated with the baseline drift removed to obtain features to be estimated; Standardizing the features to be estimated in a preset time dimension to obtain standard features to be estimated; Based on the standard feature to be estimated, a micro-motion estimation result corresponding to the motion state data to be estimated is obtained.

8. A data noise reduction device, characterized in that: The device comprises: A first noise reduction unit is used to perform baseline filtering on the motion state sample data to obtain the motion state sample data with baseline drift removed, wherein the motion state sample data with baseline drift removed includes static data and motion data; A second denoising unit is used to input the static data and the motion data into a preset domain adaptation model, and perform domain alignment processing and small motion estimation processing respectively to obtain a domain alignment result and a small motion estimation result, wherein the domain alignment result includes a static global feature and a motion global feature; an alignment loss determining unit, configured to determine an alignment loss based on the static global feature and the motion global feature; an estimated loss determining unit, configured to determine an estimated loss based on the micro-motion estimation result; A model iterative updating unit is used to iteratively update the preset domain adaptation model based on the alignment loss and the estimation loss.

9. A computer device comprising a memory, a processor, and computer-readable instructions stored in the memory and executable on the processor, characterized in that: When the processor executes the computer-readable instructions, the steps of the data noise reduction method according to any one of claims 1 to 7 are implemented.

10. A readable storage medium storing computer-readable instructions, characterized in that: When the computer-readable instructions are executed by a processor, the steps of the data noise reduction method according to any one of claims 1 to 7 are implemented.

Images