Micro-angle measurement time data denoising reconstruction method and system based on BiLSTM-CNN

Abstract

The application provides a kind of based on BiLSTM-CNN's micro-angle measurement time data denoising reconstruction method and system, the method includes: obtaining the micro-angle measurement time data of the micro-angle measurement system of visual axis pointing change;Analysis micro-angle measurement time data data variation characteristics and noise distribution law;According to data variation characteristics and noise distribution law generates micro-angle measurement data set;Based on micro-angle measurement data set trains for realizing the BiLSTM-CNN model of micro-angle measurement time sequence denoising, finds the optimal hyperparameter combination of BiLSTM-CNN model using Bayesian search algorithm, obtains micro-angle measurement time sequence denoising model;Micro-angle measurement time sequence denoising model is used to the micro-angle measurement time data of processing to denoising reconstruction, and then reduce the noise influence of micro-angle measurement time data, improve the precision of micro-angle measurement system.

Description

Technical Field

[0001] This invention relates to the field of data processing technology for micro-angle measurement systems for changes in line of sight, and particularly to a method and system for denoising and reconstructing micro-angle measurement time data based on BiLSTM-CNN. Background Technology

[0002] High-precision measurement and feedback of line-of-sight (LOS) pointing changes can be achieved using an angle meter based on the principle of optical autocollimation. Optical autocollimation angle measurement systems typically measure LOS pointing angle changes accurately by calculating the positional change of the reflected light spot on the image sensor. In practical engineering applications, micro-angle measurement systems may encounter errors and noise in optical and electronic components due to factors such as satellite on-orbit environments or harsh external conditions. Specifically, thermal noise and shot noise generated by the random operation of the image sensor's acquisition circuit resistance can cause deviations in the calculation of the light spot centroid. These centroid calculation deviations lead to inaccuracies in the LOS pointing change calculation results, reducing the reliability and accuracy of the micro-angle measurement system. When deploying high-precision centroid calculation algorithms, it is necessary to simultaneously consider calculation accuracy, resource consumption, and algorithm real-time performance. Therefore, developing a method to reduce the impact of noise from the final received LOS pointing change micro-angle measurement time series is a feasible and effective strategy. Summary of the Invention

[0003] To solve the above-mentioned technical problems, or at least partially solve them, the present invention provides a method and system for denoising and reconstructing micro-angle measurement time data based on BiLSTM-CNN.

[0004] One aspect of the present invention provides a noise reduction and reconstruction method for micro-angle measurement time data based on BiLSTM-CNN, the method comprising:

[0005] Acquire micro-angle measurement time data from the micro-angle measurement system for changes in line-of-sight pointing;

[0006] Analyze the data variation characteristics and noise distribution patterns of micro-angle measurement time data;

[0007] A standard micro-angle measurement dataset is constructed based on the data variation characteristics of the micro-angle measurement time data. Noise data conforming to the noise distribution law is added to each standard micro-angle measurement time data in the standard micro-angle measurement dataset according to the noise distribution law to form a micro-angle measurement dataset.

[0008] A BiLSTM-CNN model for denoising micro-angle measurement time series was trained based on the micro-angle measurement dataset, and the optimal hyperparameter combination of the BiLSTM-CNN model was found using the Bayesian search algorithm to obtain the micro-angle measurement time series denoising model.

[0009] The aforementioned micro-angle measurement time series denoising model is used to denoise and reconstruct the micro-angle measurement time data to be processed, so as to output the denoised micro-angle measurement time data.

[0010] The micro-angle measurement time data acquired by the micro-angle measurement system for changes in line-of-sight pointing includes:

[0011] Receive micro-angle measurement time data in binary format from the micro-angle measurement system for line-of-sight changes deployed on the satellite and transmitted to the ground station;

[0012] The binary format micro-angle measurement time data is decoded and converted into decimal data including positive and negative signs.

[0013] The data variation characteristics and noise distribution patterns of the micro-angle measurement time data include:

[0014] We conducted annual variation analysis, daily variation analysis, and hourly variation analysis on the micro-angle measurement time data to obtain the variation pattern of the micro-angle measurement time series over time.

[0015] Noise distribution analysis is performed on the micro-angle measurement time data to obtain the noise distribution pattern of the micro-angle measurement time series.

[0016] The noise distribution analysis of the micro-angle measurement time data includes:

[0017] The Monte Carlo simulation method was used to analyze the distribution law of centroid calculation deviation caused by image sensor noise, and the obtained centroid calculation deviation distribution law was used as the noise distribution law of micro-angle measurement time series.

[0018] Among them, the micro-angle measurement time data includes the first micro-angle measurement time series and the second micro-angle measurement time series output by different measurement lines of the line-of-sight pointing change micro-angle measurement system.

[0019] The variation pattern of each micro-angle measurement time series over time is as follows: it exhibits periodic variation in terms of year, day, and hour.

[0020] The noise distribution pattern of each micro-angle measurement time series conforms to a normal distribution.

[0021] The micro-angle measurement time series denoising model includes a first BiLSTM layer, a second BiLSTM layer set in parallel with the first BiLSTM layer, and a CNN layer connected to the outputs of the first BiLSTM layer and the second BiLSTM layer.

[0022] Both the first BiLSTM layer and the second BiLSTM layer employ the Tanh activation function. The first BiLSTM layer and the second BiLSTM layer are used to extract features from the two micro-angle measurement time series samples included in each micro-angle measurement time data sample in the micro-angle measurement dataset, in order to obtain the long-term dependency features in each micro-angle measurement time series sample. Each micro-angle measurement time data sample includes a first micro-angle measurement time series sample and a second micro-angle measurement time series sample representing the output of different measurement lines of the micro-angle measurement system that represent the change in line of sight pointing.

[0023] The CNN layer consists of a first convolutional layer, a first activation function layer, a first dropout layer, and a second convolutional layer. The CNN network uses the ReLU activation function. The CNN layer is used to extract and fuse the outputs of the first BiLSTM layer and the second BiLSTM layer, and reconstruct the micro-angle measurement time series based on the fused features to obtain the denoised micro-angle measurement time data.

[0024] The method further includes:

[0025] The loss function of the micro-angle measurement time series denoising model is constructed. The loss function includes a first loss function and a second loss function. The first loss function is the deviation loss loss1 between the micro-angle measurement time series after model denoising and the original micro-angle time series corresponding to the first micro-angle measurement time series sample. The second loss function is the deviation loss loss2 between the micro-angle measurement time series after model denoising and the original micro-angle time series corresponding to the second micro-angle measurement time series sample.

[0026] A dynamic gradient balancing algorithm is used to dynamically balance the gradients of the first and second loss functions according to a preset error standard level. The specific formula is shown below:

[0027]

[0028] loss=weight·loss1+(1-weight)·loss2 (2)

[0029] In the formula: grad1 and grad2 are the average values ​​of the gradient norms generated by the first loss function and the second loss function, respectively; loss is the total loss of the micro-angle measurement time series denoising model; and weight is the weight coefficient.

[0030] The hyperparameters of the BiLSTM-CNN model include: the number of BiLSTM layers and the number of hidden layer nodes, the coefficients of the Dropout layer, the batch size, the learning rate, the convolutional kernel size, and the weight decay parameter.

[0031] Another aspect of the present invention provides a noise reduction and reconstruction system for micro-angle measurement time data based on BiLSTM-CNN, the system comprising:

[0032] The acquisition module is used to acquire the micro-angle measurement time data of the micro-angle measurement system for changes in line of sight pointing;

[0033] The analysis module is used to analyze the data variation characteristics and noise distribution patterns of micro-angle measurement time data;

[0034] The data generation module is used to construct a standard micro-angle measurement dataset based on the data change characteristics of the micro-angle measurement time data, and to add noise data that conforms to the noise distribution law to each standard micro-angle measurement time data in the standard micro-angle measurement dataset according to the noise distribution law to form a micro-angle measurement dataset.

[0035] The model building module is used to train a BiLSTM-CNN model for denoising micro-angle measurement time series based on the micro-angle measurement dataset, and to use the Bayesian search algorithm to find the optimal hyperparameter combination of the BiLSTM-CNN model to obtain the micro-angle measurement time series denoising model.

[0036] The reconstruction module is used to denoise and reconstruct the micro-angle measurement time data to be processed using the micro-angle measurement time series denoising model to output the denoised micro-angle measurement time data.

[0037] The analysis module includes:

[0038] The first analysis unit is used to perform annual variation analysis, daily variation analysis, and hourly variation analysis on the micro-angle measurement time data to obtain the variation pattern of the micro-angle measurement time series over time.

[0039] The second analysis unit is used to perform noise distribution analysis on the micro-angle measurement time data to obtain the noise distribution pattern of the micro-angle measurement time series.

[0040] The present invention provides a noise reduction and reconstruction method and system for micro-angle measurement time data based on BiLSTM-CNN, which integrates BiLSTM and CNN. First, BiLSTM is used to capture the long-term dependencies between micro-angle measurement time series, and then CNN is used to further extract features. In this way, the contextual information of the micro-angle measurement time series is encoded to ensure that each data point can utilize the information before and after it. This provides CNN with more comprehensive time series features. The present invention can reconstruct micro-angle measurement time series more accurately, effectively reduce the impact of noise, and improve the accuracy of micro-angle measurement system. Attached Figure Description

[0041] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0042] Figure 1 A flowchart illustrating the noise reduction and reconstruction method for micro-angle measurement time data based on BiLSTM-CNN provided in this embodiment of the invention;

[0043] Figure 2 This is a schematic diagram of the included angle measurement system according to an embodiment of the present invention;

[0044] Figure 3 This is a schematic diagram of the deviation distribution results obtained by solving the centroid of the light spot during the Monte Carlo simulation process proposed in this embodiment of the invention;

[0045] Figure 4 A schematic diagram illustrating the specific structure and principle of the BiLSTM-CNN model provided in this embodiment of the invention;

[0046] Figure 5 This is a flowchart of the Bayesian search optimization process proposed in an embodiment of the present invention;

[0047] Figure 6 A comparative diagram showing the noise reduction and reconstruction results of micro-angle measurement time data using different models;

[0048] Figure 7 A diagram comparing the denoising performance evaluation indicators of different models for denoising reconstruction of micro-angle measurement time data.

[0049] Figure 8 The diagram shows the structure of a noise reduction and reconstruction system for micro-angle measurement time data based on BiLSTM-CNN, as provided in an embodiment of the present invention. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0051] Figure 1 A flowchart illustrating the implementation of a noise reduction and reconstruction method for micro-angle measurement time data based on BiLSTM-CNN according to an embodiment of the present invention is shown. Figure 1 As shown, the noise reduction and reconstruction method for micro-angle measurement time data based on BiLSTM-CNN proposed in this invention specifically includes the following steps:

[0052] S11. Acquire the micro-angle measurement time data from the line-of-sight pointing change micro-angle measurement system. Specifically, this includes: receiving the micro-angle measurement time data in binary format transmitted from the line-of-sight pointing change micro-angle measurement system deployed on the satellite to the ground station; decoding the binary micro-angle measurement time data and converting it into decimal data including positive and negative signs. In a specific example, the line-of-sight pointing change micro-angle measurement system has two or more measurement lines. In one specific embodiment, the micro-angle measurement system structure is as follows: Figure 2 As shown. Among them, the micro-angle measurement time data includes the first micro-angle measurement time series and the second micro-angle measurement time series output by different measurement lines of the line-of-sight pointing change micro-angle measurement system.

[0053] In this embodiment, the line-of-sight pointing change micro-angle measurement system deployed on the satellite transmits data in binary to the host computer at the ground station. In order to facilitate analysis and processing, it is necessary to perform a number system conversion. These binary floating-point numbers need to be converted into decimal numbers. The binary data is decoded and converted into the corresponding decimal values. Not only the magnitude of the value needs to be considered, but also the positive or negative sign of the value needs to be processed.

[0054] The following explanation uses a micro-angle measurement system for line-of-sight pointing changes, comprising two measurement lines, as an example. The measurement method for the micro-angle measurement time series of each measurement line includes the following steps:

[0055] Step 1, as follows Figure 2 As shown, four angle measuring instruments and their reflectors are arranged at corresponding positions on the satellite frame.

[0056] Step 2: When the on-orbit space environment changes, the satellite's line of sight will experience a slight spatial three-axis angular shift, which will cause the position of the image point on the image plane of each angle measuring instrument sensor to change.

[0057] Step 3: Obtain the image point coordinates on the sensor image plane of the three angle measuring instruments, and calculate the pitch and yaw angle deflection of the corresponding reflecting mirror. J1, J2, and J3 form one group, and J1, J4, and J3 form another group. The positions of J2 and J4 are fixed in parallel on the same reference value, that is, the angle change of the same reference is being measured.

[0058] Step four: Calculate the spatial three-axis angular offset of the satellite's on-orbit line of sight using an algorithm based on the angular deflection measured by the three angle measuring instruments J1, J2, and J3. Assume the spatial three-axis angular changes of rigid bodies A and B relative to the world coordinate system are represented as (a1, b1, c1) and (a2, b2, c2), respectively. Then the corresponding image point coordinates on the sensor image plane of the three angle measuring instruments J1, J2, and J3 are (d... x1 d y1 ), (d x2 d y2 ) and (d x3 d y3 Then the corresponding coordinates of the image points on the sensor image planes of the three angle measuring instruments J1, J4, and J3 are (d...). x1 d y1 ), (d x4 d y4 ) and (d x3 d y4 ).

[0059] For combinations J1, J2, and J3, the relative pointing change (a1, b1, c1) at point A represents the satellite's orbital line-of-sight pointing change. This can be obtained by solving the following system of equations: MS = D, where M is a parameter matrix determined by the positions and relative angles of each angle measuring instrument and the reflector at the "zero position". S is [a1, b1, c1, a2, b2, c2]. T , representing the relative spatial three-axis angular changes of rigid bodies A and B, where D is [d x1 ,d y1 ,d x2 ,d y2 ,d x3 ,d y3 ] T , representing the change in image point coordinates on the sensor image plane measured by the three angle measuring instruments. Solving for S, we get S =

[0060] M -1 D, where S is [a1,b1,c1,a2,b2,c2] T The first three terms (a1, b1, c1) represent the micro-angle of the change in the orbital line of sight of the satellite.

[0061] For combinations J1, J4, and J3, the relative pointing change (a1, b1, c1) at point A represents the satellite's orbital line-of-sight pointing change. This can be obtained by solving the following system of equations: MS = D, where M is a parameter matrix determined by the positions and relative angles of each angle measuring instrument and the reflector at the "zero position". S represents [a1, b1, c1, a2, b2, c2]. T , representing the relative spatial three-axis angular changes of rigid bodies A and B, where D is [d x1 ,d y1 ,d x4 ,d y4 ,d x3 ,d y3 ] T , representing the change in image point coordinates on the sensor image plane measured by the three angle measuring instruments. Solving for S, we get S =

[0062] M -1 D, where S is [a1,b1,c1,a2,b2,c2] T The first three terms (a1, b1, c1) represent the micro-angle of the change in the orbital line of sight of the satellite.

[0063] S12. Analyze the data variation characteristics and noise distribution patterns of the micro-angle measurement time data.

[0064] The main causes of changes in line-of-sight (LOS) pointing are external vibrations, thermal deformation, and other factors, exhibiting strong temporal correlation and continuity. Therefore, the time series of LOS pointing changes measured by the micro-angle measurement system shows a strong time dependence. This invention analyzes the factors affecting LOS pointing changes, and combines the working environment of the micro-angle measurement system with measured data to analyze the data characteristics and noise sources of the micro-angle measurement time data, thereby obtaining the data variation patterns and noise distribution patterns of the micro-angle measurement time data.

[0065] S13. Construct a standard micro-angle measurement dataset based on the data change characteristics of the micro-angle measurement time data, and add noise data that conforms to the noise distribution law to each standard micro-angle measurement time data in the standard micro-angle measurement dataset according to the noise distribution law to form a micro-angle measurement dataset.

[0066] This invention obtains the data variation pattern and noise distribution pattern of micro-angle measurement time data based on theoretical analysis and received micro-angle measurement data. Then, a micro-angle measurement dataset is generated based on the data variation pattern and noise distribution pattern. Each micro-angle measurement data in the micro-angle measurement dataset includes initial data and noisy data. The micro-angle measurement dataset is divided into a training set, a validation set, and a test set, which are used for model training, validation, and testing, respectively.

[0067] S14. Based on the micro-angle measurement dataset, train a BiLSTM-CNN model for denoising micro-angle measurement time series, and use the Bayesian search algorithm to find the optimal hyperparameter combination of the BiLSTM-CNN model to obtain the micro-angle measurement time series denoising model.

[0068] This invention constructs a BiLSTM-CNN model for denoising micro-angle measurement time series, trains the BiLSTM-CNN network, and uses Bayesian search to find the optimal hyperparameter combination to obtain and save the micro-angle measurement time series denoising model. BiLSTM stands for Bi-directional Long Short-Term Memory, composed of a forward LSTM and a backward LSTM. It is suitable for handling sequence labeling tasks with contextual relationships, can capture bidirectional semantic dependencies, better model contextual information, is suitable for handling sequence problems, and can capture long-distance dependencies.

[0069] S15. The micro-angle measurement time series denoising model is used to denoise and reconstruct the micro-angle measurement time data to be processed, so as to output the denoised micro-angle measurement time data.

[0070] The present invention provides a noise reduction and reconstruction method for micro-angle measurement time data based on BiLSTM-CNN, which integrates BiLSTM and CNN. First, BiLSTM is used to capture the long-term dependencies between micro-angle measurement time series, and then CNN is used to further extract features. By encoding the contextual information of the micro-angle measurement time series, it is ensured that each data point can utilize the information before and after it. This provides CNN with more comprehensive time series features. The present invention can reconstruct micro-angle measurement time series more accurately, effectively reduce the impact of noise, and improve the accuracy of micro-angle measurement system.

[0071] In this embodiment of the invention, the analysis of the data variation characteristics and noise distribution patterns of the micro-angle measurement time data in step S12 specifically includes the following steps, not shown in the accompanying drawings:

[0072] S121. Analyze the annual variation pattern, daily variation pattern, and hourly variation pattern of the micro-angle measurement time data to obtain the variation pattern of the micro-angle measurement time series over time. Specifically, the variation pattern of each micro-angle measurement time series over time is as follows: it exhibits periodic variation on an annual basis, periodic variation on a daily basis, and periodic variation on an hourly basis.

[0073] Specifically, based on the working principle and working environment of the included angle measurement system, data analysis was conducted. According to the analysis, the experimental data should have the following characteristics:

[0074] Overall, the changes should exhibit a certain regularity on an annual basis. This is because one cycle of the Earth's revolution around the sun plus its rotation is about one year. Therefore, the intensity and area of ​​sunlight received by a geostationary satellite should change periodically on an annual basis. These on-orbit changes will lead to periodic changes in external heat flow.

[0075] In a localized context, the data should exhibit a certain regularity and periodicity on a daily basis. This is because the Earth rotates on its axis while also revolving around the Sun, causing the relative position of the satellite to be different every day and every moment. However, since geostationary orbit satellites rotate simultaneously with the Earth and their relative positions do not change, although the values ​​may differ between different dates within a year, the daily variation follows a fixed trend.

[0076] Furthermore, the analysis of the received micro-angle measurement data shows that the data range is within a certain positive and negative value range, and changes slowly and smoothly over a 24-hour period.

[0077] S122. Perform noise distribution analysis on the micro-angle measurement time data to obtain the noise distribution pattern of the micro-angle measurement time series. Further, the noise distribution analysis of the micro-angle measurement time data includes: using Monte Carlo simulation to analyze the centroid calculation deviation distribution pattern caused by image sensor noise, and using the obtained centroid calculation deviation distribution pattern as the noise distribution pattern of the micro-angle measurement time series. Specifically, the noise distribution pattern of each micro-angle measurement time series conforms to a normal distribution.

[0078] In this embodiment, Monte Carlo simulation and D'Agostino-Pearson normality test are used to analyze and verify the noise distribution characteristics. The Monte Carlo simulation method is implemented as follows: the Monte Carlo simulation method is used to generate an initial standard simulation image, and the centroid of the light spot is calculated by locating the centroid using the OTSU centroid method. Since the noise mainly comes from shot noise and thermal noise generated by resistive motion, appropriate Poisson-distributed shot noise and Gaussian-distributed thermal noise are added to the initial ideal image. The D'Agostino-Pearson normality test is then used, and the P-value standard is set to 0.05.

[0079] In a specific example, for noise source analysis, there is thermal noise error caused by the random motion of resistors in the image sensor electronic device, which mainly manifests as Gaussian white noise; and shot noise composed of input photon shot noise, photocurrent shot noise, and dark current shot noise, which is caused by the statistical uncertainty of a large number of individual events and follows a Poisson distribution in time. We choose to add shot noise with a Poisson distribution and a gain of 5 to the simulated image; and we choose to add Gaussian noise with a mean of 2000 and a standard deviation of 0.1 to the simulated image.

[0080] Specifically, an initial simulation image of size 1100x1100 is created, with the target spot covering an area of ​​11x11 pixels. The gray values ​​of the spot are concentrated within a central 5x5 pixel area and follow a two-dimensional Gaussian function distribution. The gray value at the center of the spot is set to 20000, and the centroid of the spot is calculated using the OTSU centroid method. By adding the aforementioned noise to the ideal image, the distribution characteristics of the image noise mapped to the spot centroid algorithm are verified using Monte Carlo simulation and the D'Agostino-Pearson normality test. The p-value standard is set to 0.05, and the offset distribution is observed. If the final p-value is greater than 0.05, it can be considered that the distribution law of the deviation distribution of the image sensor's imaging noise mapped to the spot centroid calculation conforms to a normal distribution. The results are as follows. Figure 3 As shown, the P-value for the offset distribution in the X direction is 0.812276, and the P-value for the offset distribution in the Y direction is 0.685084. This indicates that the deviation distribution of the imaging noise of the image sensor mapped onto the centroid solution of the light spot conforms to a normal distribution, that is, the deviation distribution of the micro-angle measurement data conforms to a normal distribution.

[0081] Furthermore, this invention can generate a dataset based on theoretical analysis and received micro-angle measurement data, and divide it into a training set, a validation set, and a test set. In a specific example, a total of eight years of micro-angle measurement data can be generated. Based on the noise distribution characteristics in the noise analysis, corresponding Gaussian white noise is added to form the micro-angle measurement dataset. The standard deviation of the added Gaussian noise ranges from 0 to 0.8, and the step size is 0.05. The first five years of the micro-angle measurement dataset are used as the training set, the sixth and seventh years as the validation set, and the eighth year as the test set. The test set is used to test the generalization level of the model's denoising ability.

[0082] like Figure 4 As shown, the micro-angle measurement time series denoising model constructed in this embodiment of the invention includes a first BiLSTM layer, a second BiLSTM layer set in parallel with the first BiLSTM layer, and a CNN layer connected to the outputs of the first BiLSTM layer and the second BiLSTM layer.

[0083] The first and second BiLSTM layers both employ the Tanh activation function. These layers are used to extract features from the two micro-angle measurement time series samples included in each micro-angle measurement time series sample in the micro-angle measurement dataset, respectively, to obtain long-term dependency features in each sample. Each micro-angle measurement time series sample includes a first micro-angle measurement time series sample and a second micro-angle measurement time series sample representing the outputs of different measurement lines of the micro-angle measurement system indicating changes in line-of-sight pointing. The CNN layer consists of a first convolutional layer, a first activation function layer, a first dropout layer, and a second convolutional layer. The CNN network uses the ReLU activation function and is used to extract and fuse features from the outputs of the first and second BiLSTM layers. Based on the fused features, the micro-angle measurement time series is reconstructed to obtain denoised micro-angle measurement time data. In this embodiment, the BiLSTM layer extracts features from the input micro-angle measurement time series data to capture long-term dependencies in the sequence. The CNN part further extracts and fuses features, and reconstructs the micro-angle measurement time series data through a noise reduction process. The model's optimizer is the Adam optimizer.

[0084] Furthermore, the denoising and reconstruction method for micro-angle measurement time data based on BiLSTM-CNN proposed in this invention also includes the step of constructing a loss function for the micro-angle measurement time series denoising model. The loss function of the model includes a first loss function and a second loss function. The first loss function is the deviation loss loss1 between the model-denoised micro-angle measurement time series corresponding to the first micro-angle measurement time series sample and the original micro-angle time series. The second loss function is the deviation loss loss2 between the model-denoised micro-angle measurement time series corresponding to the second micro-angle measurement time series sample and the original micro-angle time series.

[0085] This invention employs a dynamic gradient balancing algorithm, which dynamically balances the gradients of the first and second loss functions according to a preset error standard level. The specific formula is shown below:

[0086]

[0087] loss=weight·loss1+(1-weight)·loss2 (2)

[0088] In the formula: grad1 and grad2 are the average values ​​of the gradient norms generated by the first loss function and the second loss function, respectively; loss is the total loss of the micro-angle measurement time series denoising model; and weight is the weight coefficient.

[0089] This invention incorporates a gradient balancing algorithm, which dynamically balances the gradients of two different loss functions based on the different error standard levels of the input. This allows for the dynamic adjustment of the weights of the two micro-angle change measurement time series in the model's loss, ensuring that each loss contributes as equally as possible to the final gradient and preventing the model from overlearning one of the micro-angle time series.

[0090] This invention utilizes a Bayesian search algorithm to fine-tune the parameters of a BiLSTM-CNN model. The parameters to be adjusted include: the number of BiLSTM layers (num_layers), the number of hidden nodes in the BiLSTM layer (hidden_size), the dropout rate, the batch size (batch_size), the learning rate (lr), the kernel size (kernel_size), and the weight decay parameter (weight_decay). The Bayesian search flowchart is shown below. Figure 5 As shown, the training and validation sets are fed into the model. Using the Bayesian optimization algorithm, the model parameters are set as follows: the number of BiLSTM layers is set to 4, the number of hidden nodes in the BiLSTM layer is set to 44, the coefficient of the Dropout layer is set to 0.15, the batch size is set to 6, the learning rate is set to 0.0002, the convolutional kernel size is 13, the weight decay parameter is set to 0.0001, the optimizer is Adam optimizer, the loss function is set to MSE, and the number of iterations is 200.

[0091] This invention also provides a model evaluation step, which specifically includes inputting test data into a trained micro-angle measurement time series denoising model and evaluating the results.

[0092] Specifically, the test set is fed into a pre-trained micro-angle measurement time series denoising model, such as... Figure 4 As shown, the data is arranged according to... Figure 4 The output of the model after the process is the denoised and reconstructed micro-angle measurement time series. For ease of demonstration, a portion of the noise identification and removal results for the model with input standard deviations of 0.2 and 0.7 are shown in the image. Figure 6 As shown. MSE (Mean Squared Error), RMSE (Root Mean Square Error), MAE (Mean Absolute Error), and R... 2 A comparative analysis was conducted using four evaluation indicators: (R-squared, coefficient of determination). The results are as follows: Figure 7As shown, the denoising effect of BiLSTM-CNN is significantly improved compared to traditional RNNs (Recurrent Neural Networks), LSTMs (Long Short-Term Memory Networks), GRUs (Gated Recurrent Units), and BiLSTMs (Bi-directional Long Short-Term Memory Networks). Compared to BiLSTM, which has the best performance among the four traditional models, BiLSTM-CNN achieves significant reductions of 58.01%, 35.20%, and 35.19% in MSE, RMSE, and MAE, respectively. 2 It also showed an improvement of 0.062%, indicating that the model was able to identify and remove noise well, while preserving the original features of the micro-angle data to the greatest extent.

[0093] Compared with the prior art, the present invention has the following advantages:

[0094] In the BiLSTM-CNN architecture proposed in this invention, the bidirectional dependency of micro-angle measurement time series is first captured by BiLSTM, fully integrating the bidirectional temporal information of the sequence. This provides the CNN with richer temporal features, enabling a more comprehensive understanding of the changing trends of micro-angle measurement data when processing time series data. Consequently, the CNN can more effectively extract these temporally enhanced features. This structural optimization makes the BiLSTM-CNN model exhibit better performance and stronger generalization ability, especially in the task of denoising and reconstructing micro-angle measurement time series, which has high requirements for long-term dependencies. Compared with traditional models, this model outperforms in multiple performance indicators, proving its effectiveness in denoising and data reconstruction. It provides a novel and reliable approach for time series denoising in line-of-sight pointing micro-angle measurement systems, which is of great significance for improving the accuracy of line-of-sight pointing micro-angle measurement systems.

[0095] Figure 8 This diagram illustrates a structural block diagram of a BiLSTM-CNN-based denoising and reconstruction system for micro-angle measurement time data, according to an embodiment of the present invention. This system is used to implement the BiLSTM-CNN-based denoising and reconstruction method for micro-angle measurement time data as described above. Figure 8 As shown, the noise reduction and reconstruction system for micro-angle measurement time data based on BiLSTM-CNN provided in this embodiment of the invention includes:

[0096] The acquisition module 701 is used to acquire the micro-angle measurement time data of the micro-angle measurement system for line-of-sight pointing change.

[0097] Analysis module 702 is used to analyze the data variation characteristics and noise distribution patterns of micro-angle measurement time data;

[0098] The data generation module 703 is used to construct a standard micro-angle measurement dataset based on the data change characteristics of the micro-angle measurement time data, and to add noise data that conforms to the noise distribution law to each standard micro-angle measurement time data in the standard micro-angle measurement dataset according to the noise distribution law to form a micro-angle measurement dataset.

[0099] The model building module 704 is used to train a BiLSTM-CNN model for denoising micro-angle measurement time series based on the micro-angle measurement dataset, and to use the Bayesian search algorithm to find the optimal hyperparameter combination of the BiLSTM-CNN model to obtain the micro-angle measurement time series denoising model.

[0100] The reconstruction module 705 is used to perform noise reduction and reconstruction on the micro-angle measurement time data to be processed using the micro-angle measurement time series denoising model to output the denoised micro-angle measurement time data.

[0101] In this embodiment of the invention, the analysis module 702 includes a first analysis unit and a second analysis unit, wherein:

[0102] The first analysis unit is used to perform annual variation analysis, daily variation analysis, and hourly variation analysis on the micro-angle measurement time data to obtain the variation pattern of the micro-angle measurement time series over time.

[0103] The second analysis unit is used to perform noise distribution analysis on the micro-angle measurement time data to obtain the noise distribution pattern of the micro-angle measurement time series.

[0104] In this embodiment, the micro-angle measurement time series denoising model includes a first BiLSTM layer, a second BiLSTM layer set in parallel with the first BiLSTM layer, and a CNN layer connected to the outputs of the first BiLSTM layer and the second BiLSTM layer.

[0105] Both the first BiLSTM layer and the second BiLSTM layer employ the Tanh activation function. The first BiLSTM layer and the second BiLSTM layer are used to extract features from the two micro-angle measurement time series samples included in each micro-angle measurement time data sample in the micro-angle measurement dataset, in order to obtain the long-term dependency features in each micro-angle measurement time series sample. Each micro-angle measurement time data sample includes a first micro-angle measurement time series sample and a second micro-angle measurement time series sample representing the output of different measurement lines of the micro-angle measurement system that represent the change in line of sight pointing.

[0106] The CNN layer consists of a first convolutional layer, a first activation function layer, a first dropout layer, and a second convolutional layer. The CNN network uses the ReLU activation function. The CNN layer is used to extract and fuse the outputs of the first BiLSTM layer and the second BiLSTM layer, and reconstruct the micro-angle measurement time series based on the fused features to obtain the denoised micro-angle measurement time data.

[0107] Furthermore, the model building module 704 is also used to construct the loss function of the micro-angle measurement time series denoising model. The loss function includes a first loss function and a second loss function. The first loss function is the deviation loss loss1 between the model-denoised micro-angle measurement time series corresponding to the first micro-angle measurement time series sample and the original micro-angle time series. The second loss function is the deviation loss loss2 between the model-denoised micro-angle measurement time series corresponding to the second micro-angle measurement time series sample and the original micro-angle time series. A dynamic gradient balancing algorithm is used to dynamically balance the gradients of the first loss function and the second loss function according to a preset error standard level. The specific formula is as follows:

[0108]

[0109] loss=weight·loss1+(1-weight)·loss2 (2)

[0110] In the formula: grad1 and grad2 are the average values ​​of the gradient norms generated by the first loss function and the second loss function, respectively; loss is the total loss of the micro-angle measurement time series denoising model; and weight is the weight coefficient.

[0111] The present invention provides a noise reduction and reconstruction method and system for micro-angle measurement time data based on BiLSTM-CNN, which integrates BiLSTM and CNN. First, BiLSTM is used to capture the long-term dependencies between micro-angle measurement time series, and then CNN is used to further extract features. In this way, the contextual information of the micro-angle measurement time series is encoded to ensure that each data point can utilize the information before and after it. This provides CNN with more comprehensive time series features. The present invention can reconstruct micro-angle measurement time series more accurately, effectively reduce the impact of noise, and improve the accuracy of micro-angle measurement system.

[0112] Furthermore, those skilled in the art will understand that although some embodiments herein include certain features included in other embodiments but not others, combinations of features from different embodiments are intended to be within the scope of the invention and form different embodiments. For example, any of the claimed embodiments can be used in any combination.

[0113] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for denoising and reconstructing micro-angle measurement time data based on BiLSTM-CNN, characterized in that, The method includes: Acquire micro-angle measurement time data from the micro-angle measurement system for changes in line-of-sight pointing; Analyze the data variation characteristics and noise distribution patterns of micro-angle measurement time data; A standard micro-angle measurement dataset is constructed based on the data variation characteristics of the micro-angle measurement time data. Noise data conforming to the noise distribution law is added to each standard micro-angle measurement time data in the standard micro-angle measurement dataset according to the noise distribution law to form a micro-angle measurement dataset. A BiLSTM-CNN model for denoising micro-angle measurement time series was trained based on the micro-angle measurement dataset, and the optimal hyperparameter combination of the BiLSTM-CNN model was found using the Bayesian search algorithm to obtain the micro-angle measurement time series denoising model. The micro-angle measurement time series denoising model is used to denoise and reconstruct the micro-angle measurement time data to be processed, so as to output the denoised micro-angle measurement time data. The micro-angle measurement time series denoising model includes a first BiLSTM layer, a second BiLSTM layer set in parallel with the first BiLSTM layer, and a CNN layer connected to the outputs of the first BiLSTM layer and the second BiLSTM layer. Both the first BiLSTM layer and the second BiLSTM layer employ the Tanh activation function. The first BiLSTM layer and the second BiLSTM layer are used to extract features from the two micro-angle measurement time series samples included in each micro-angle measurement time data sample in the micro-angle measurement dataset, in order to obtain the long-term dependency features in each micro-angle measurement time series sample. Each micro-angle measurement time data sample includes a first micro-angle measurement time series sample and a second micro-angle measurement time series sample representing the output of different measurement lines of the micro-angle measurement system that represent the change in line of sight pointing. The CNN layer consists of a first convolutional layer, a first activation function layer, a first dropout layer, and a second convolutional layer. The CNN network uses the ReLU activation function. The CNN layer is used to extract and fuse the outputs of the first BiLSTM layer and the second BiLSTM layer, and reconstruct the micro-angle measurement time series based on the fused features to obtain the denoised micro-angle measurement time data. The method further includes: The loss function of the micro-angle measurement time series denoising model is constructed. The loss function includes a first loss function and a second loss function. The first loss function is the deviation loss loss1 between the micro-angle measurement time series after model denoising and the original micro-angle time series corresponding to the first micro-angle measurement time series sample. The second loss function is the deviation loss loss2 between the micro-angle measurement time series after model denoising and the original micro-angle time series corresponding to the second micro-angle measurement time series sample. A dynamic gradient balancing algorithm is used to dynamically balance the gradients of the first and second loss functions according to a preset error standard level. The specific formula is shown below: ； ； In the formula: grad1 and grad2 are the average of the gradient norms generated by the first loss function and the second loss function, respectively, and loss is the total loss of the micro-angle measurement time series denoising model. These are the weighting coefficients.

2. The method according to claim 1, characterized in that, The micro-angle measurement time data obtained by the micro-angle measurement system for line-of-sight pointing change includes: Receive micro-angle measurement time data in binary format from the micro-angle measurement system for line-of-sight changes deployed on the satellite and transmitted to the ground station; The binary format micro-angle measurement time data is decoded and converted into decimal data including positive and negative signs.

3. The method according to claim 1, characterized in that, The data variation characteristics and noise distribution patterns of the micro-angle measurement time data analyzed include: We conducted annual variation analysis, daily variation analysis, and hourly variation analysis on the micro-angle measurement time data to obtain the variation pattern of the micro-angle measurement time series over time. Noise distribution analysis is performed on the micro-angle measurement time data to obtain the noise distribution pattern of the micro-angle measurement time series.

4. The method according to claim 3, characterized in that, The noise distribution analysis of the micro-angle measurement time data includes: The Monte Carlo simulation method was used to analyze the distribution law of centroid calculation deviation caused by image sensor noise, and the obtained centroid calculation deviation distribution law was used as the noise distribution law of micro-angle measurement time series.

5. The method according to claim 3, characterized in that, The micro-angle measurement time data includes the first micro-angle measurement time series and the second micro-angle measurement time series output by different measurement lines of the line-of-sight pointing change micro-angle measurement system. The variation pattern of each micro-angle measurement time series over time is as follows: it exhibits periodic variation in terms of year, day, and hour. The noise distribution pattern of each micro-angle measurement time series conforms to a normal distribution.

6. The method according to claim 1, characterized in that, The hyperparameters of the BiLSTM-CNN model include: the number of BiLSTM layers and the number of hidden layer nodes, the coefficients of the Dropout layer, the batch size, the learning rate, the kernel size, and the weight decay parameter.

7. A noise reduction and reconstruction system for micro-angle measurement time data based on BiLSTM-CNN, characterized in that, The system includes: The acquisition module is used to acquire the micro-angle measurement time data of the micro-angle measurement system for changes in line of sight pointing; The analysis module is used to analyze the data variation characteristics and noise distribution patterns of micro-angle measurement time data; The data generation module is used to construct a standard micro-angle measurement dataset based on the data change characteristics of the micro-angle measurement time data, and to add noise data that conforms to the noise distribution law to each standard micro-angle measurement time data in the standard micro-angle measurement dataset according to the noise distribution law to form a micro-angle measurement dataset. The model building module is used to train a BiLSTM-CNN model for denoising micro-angle measurement time series based on the micro-angle measurement dataset, and to use the Bayesian search algorithm to find the optimal hyperparameter combination of the BiLSTM-CNN model to obtain the micro-angle measurement time series denoising model. The reconstruction module is used to denoise and reconstruct the micro-angle measurement time data to be processed using the micro-angle measurement time series denoising model to output the denoised micro-angle measurement time data. The micro-angle measurement time series denoising model includes a first BiLSTM layer, a second BiLSTM layer set in parallel with the first BiLSTM layer, and a CNN layer connected to the outputs of the first BiLSTM layer and the second BiLSTM layer. Both the first BiLSTM layer and the second BiLSTM layer employ the Tanh activation function. The first BiLSTM layer and the second BiLSTM layer are used to extract features from the two micro-angle measurement time series samples included in each micro-angle measurement time data sample in the micro-angle measurement dataset, in order to obtain the long-term dependency features in each micro-angle measurement time series sample. Each micro-angle measurement time data sample includes a first micro-angle measurement time series sample and a second micro-angle measurement time series sample representing the output of different measurement lines of the micro-angle measurement system that represent the change in line of sight pointing. The CNN layer consists of a first convolutional layer, a first activation function layer, a first dropout layer, and a second convolutional layer. The CNN network uses the ReLU activation function. The CNN layer is used to extract and fuse the outputs of the first BiLSTM layer and the second BiLSTM layer, and reconstruct the micro-angle measurement time series based on the fused features to obtain the denoised micro-angle measurement time data. The model building module is also used to construct the loss function of the micro-angle measurement time series denoising model. The loss function includes a first loss function and a second loss function. The first loss function is the deviation loss loss1 between the model-denoised micro-angle measurement time series corresponding to the first micro-angle measurement time series sample and the original micro-angle time series. The second loss function is the deviation loss loss2 between the model-denoised micro-angle measurement time series corresponding to the second micro-angle measurement time series sample and the original micro-angle time series. A dynamic gradient balancing algorithm is used to dynamically balance the gradients of the first loss function and the second loss function according to a preset error standard level. The specific formula is shown below: ； ； In the formula: grad1 and grad2 are the average of the gradient norms generated by the first loss function and the second loss function, respectively, and loss is the total loss of the micro-angle measurement time series denoising model. These are the weighting coefficients.

8. The system according to claim 7, characterized in that, The analysis module includes: The first analysis unit is used to perform annual variation analysis, daily variation analysis, and hourly variation analysis on the micro-angle measurement time data to obtain the variation pattern of the micro-angle measurement time series over time. The second analysis unit is used to perform noise distribution analysis on the micro-angle measurement time data to obtain the noise distribution pattern of the micro-angle measurement time series.

Images