Inertial sensor over-range signal repairing and signal quality enhancing system

Abstract

The invention discloses an inertial sensor outrange signal restoration and signal quality enhancement system, and relates to the technical field of sensors, in the system, a signal acquisition module is used for acquiring a high-cost sensor signal and a low-cost sensor signal; the generator GANH-L and the generator GANL-H are used for conversion between a low-cost sensor signal and a high-cost sensor signal; the MLE module is used for injecting Laplace energy into the signal when the generator GANL-> H converts the low-cost sensor signal into the high-cost sensor signal, and injecting Laplace energy into the signal when the generator GANH-> L converts the high-cost sensor signal into the low-cost sensor signal; and the OTS module is used for constructing the optimal mapping between the low-cost and high-cost sensor signal features based on the optimal transportation theory. According to the method, through multi-level detail enhancement, the reconstruction precision and quality of the low-cost accelerometer signal are remarkably improved.

Description

Technical Field

[0001] The present application relates to the field of sensor technology, and in particular to an inertial sensor over-range signal repair and signal quality enhancement system. Background Art

[0002] Low-cost accelerometers have become key components in modern industry, medical care, and smart devices due to their affordability and wide application. Nevertheless, these sensors are often limited by accuracy and measurement range, and have difficulty providing accurate signals in high-dynamic environments, which may lead to data loss or measurement errors in some critical applications. Therefore, improving the signal quality and extending the measurement range of low-cost accelerometers has significant practical value. Although traditional hardware upgrade methods can improve sensor performance to a certain extent, their high cost limits their promotion in large-scale applications. On the other hand, as an advanced generative model, generative adversarial network (GAN) can convert low-quality input into high-quality output by learning the distribution characteristics of data, providing an economical and efficient way to enhance the signal of low-cost accelerometers.

[0003] However, due to the lack of frame-by-frame paired data of low-cost and high-cost sensor signals, traditional GAN ​​architectures and fully supervised training methods are not applicable in this case. Summary of the invention

[0004] The purpose of this application is to provide an inertial sensor over-range signal repair and signal quality enhancement system, which can significantly improve the reconstruction accuracy and quality of low-cost accelerometer signals through multi-level detail enhancement.

[0005] To achieve the above objectives, this application provides the following solutions:

[0006] In a first aspect, the present application provides an inertial sensor over-range signal repair and signal quality enhancement system, comprising: a generator GAN L→H , Generator GAN H→L , OTS module, MLE module and signal acquisition module;

[0007] The signal acquisition module is used to acquire a high-cost sensor signal and a low-cost sensor signal; the correlation between the low-cost sensor signal and the high-cost sensor signal is shown as unpaired or weakly paired;

[0008] The generator GAN H→L , used to convert high-cost sensor signals into low-cost sensor signals;

[0009] The generator GAN L→H , used to convert low-cost sensor signals into high-cost sensor signals;

[0010] The MLE module is used to: L→H When converting low-cost sensor signals into high-cost sensor signals, Laplace energy is injected into the low-cost sensor signals, and the generator GAN H→L When converting a high-cost sensor signal into a low-cost sensor signal, injecting Laplace energy into the high-cost sensor signal; the Laplace energy is a Laplace energy of a neural network; the Laplace energy is used to adjust the Laplace energy of a model in a generative deep learning architecture;

[0011] The OTS module is used to mine the potential correlation between unpaired and weakly paired data based on the optimal transport theory, and to construct an optimal mapping between the features of the low-cost sensor signal and the features of the high-cost sensor signal according to the potential correlation.

[0012] According to the specific embodiments provided in this application, this application discloses the following technical effects:

[0013] This application provides an inertial sensor over-range signal repair and signal quality enhancement system, in which the generator GAN L→H , Generator GAN H→L , OTS module, MLE module and signal acquisition module constitute this system; the signal acquisition module is responsible for collecting high-cost sensor signals and low-cost sensor signals; the generator GAN H→L The function of GAN is to convert high-cost sensor signals into low-cost sensor signals; L→H It is responsible for converting low-cost sensor signals into high-cost sensor signals; the function of the MLE module is in the generator GAN L→H In the process of converting the signal, the modulated Laplace energy is injected into the low-cost sensor signal, and the generator GAN H→L In the process of signal conversion, the modulated Laplace energy is injected into the high-cost sensor signal; the OTS module establishes the optimal mapping relationship between the low-cost sensor signal characteristics and the high-cost sensor signal characteristics based on the optimal transport theory. This application can effectively enhance the signal quality of low-cost inertial sensors without relying on frame-by-frame pairing data. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments 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 creative work.

[0015] Figure 1 A schematic diagram of the HEROS-GAN framework provided in one embodiment of the present application;

[0016] Figure 2 A schematic diagram of optimal transmission supervision provided in an embodiment of the present application, wherein (a) is element-by-element supervision and (b) is optimal transmission supervision;

[0017] Figure 3 A schematic diagram of Modulated Laplace Energy (MLE) provided in one embodiment of the present application. DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0019] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0020] Embodiment 1

[0021] This embodiment provides an inertial sensor over-range signal repair and signal quality enhancement system, including: a generator GAN L→H , Generator GAN H→L , OTS module, MLE module and signal acquisition module.

[0022] The signal acquisition module is used to acquire a high-cost sensor signal and a low-cost sensor signal; the correlation between the low-cost sensor signal and the high-cost sensor signal is expressed as unpaired or weakly paired.

[0023] The generator GAN H→L , used to convert high-cost sensor signals into low-cost sensor signals.

[0024] The generator GAN L→H , used to convert low-cost sensor signals into high-cost sensor signals.

[0025] The MLE module is used to: L→H When converting low-cost sensor signals into high-cost sensor signals, Laplace energy is injected into the low-cost sensor signals, and the generator GAN H→LWhen converting a high-cost sensor signal into a low-cost sensor signal, Laplace energy is injected into the high-cost sensor signal; the Laplace energy is a Laplace energy of a neural network; the Laplace energy is used to adjust the Laplace energy of a model in a generative deep learning architecture.

[0026] The OTS module is used to mine the potential correlation between unpaired and weakly paired data based on the optimal transport theory, and to construct an optimal mapping between the features of the low-cost sensor signal and the features of the high-cost sensor signal according to the potential correlation.

[0027] In some embodiments, the overall architecture of HEROS-GAN is as follows: Figure 1 As shown in Figure 1, it mainly includes two generators. Generator GAN L→H Responsible for converting low-cost sensor signals into high-cost sensor signals, while the generator GAN H→L The HEROS-GAN is responsible for mapping the high-cost sensor signal back to the low-cost sensor signal. The design of this bidirectional generation architecture is inspired by CycleGAN, but unlike the traditional CycleGAN, HEROS-GAN introduces OTS and MLE modules in the generation and supervision process to enhance the learning ability and generation effect of the model.

[0028] The high-cost sensor signal acquired by the signal acquisition module is an unpaired data signal in the inertial sensor. The low-cost sensor signal acquired by the signal acquisition module is an unpaired data signal in the inertial sensor.

[0029] In some embodiments, the OTS module includes a transportation cost submodule, a feature alignment submodule, and an optimal mapping calculation submodule.

[0030] The transportation cost submodule is used to calculate the transportation cost between the characteristics of the low-cost sensor signal and the characteristics of the high-cost sensor signal; the transportation cost is determined according to the similarity between the characteristics of the low-cost sensor signal and the characteristics of the high-cost sensor signal.

[0031] The feature alignment submodule is used to align the features of the low-cost sensor signal with the features of the high-cost sensor signal.

[0032] The optimal mapping calculation submodule is used to determine the optimal mapping of the features of the low-cost sensor signal to the features of the high-cost sensor signal according to the minimum transportation cost when the features of the low-cost sensor signal are aligned with the features of the high-cost sensor signal.

[0033] Specifically, the role of the OTS module is to extract as much supervisory information as possible from the unpaired data and feed it back to the generator. Specifically, the OTS module finds the optimal mapping between the low-cost sensor signal features and the high-cost sensor signal features through the optimal transport theory to minimize the difference between the two. The introduction of this module effectively solves the problem of insufficient supervision in the case of unpaired data and provides the generator with richer training signals. The MLE module enhances the local details of the generated signal by calculating the Laplace energy of the features and performing appropriate energy modulation. This module is particularly important in high dynamic signal generation because it ensures that the generator does not introduce too much noise when capturing subtle changes in the signal, thereby generating a physically consistent and detailed signal.

[0034] Specifically, the core idea of ​​the OTS mechanism is to construct the optimal mapping between the low-cost sensor signal features and the high-cost sensor signal features through the optimal transportation theory, so as to maximize the supervision information in the unpaired data. The feature representation of the low-cost sensor signal is set as The characteristic representation of the high-cost sensor signal is At this point, the goal of the optimal transportation problem is to find an optimal mapping that aligns the feature distribution of the low-cost sensor signal with the feature distribution of the high-cost sensor signal to minimize the “transportation cost” between the two feature distributions, such as Figure 2 This problem can be formalized as the following optimization problem:

[0035]

[0036] In the formula, Γ(F L ,F H ) represents the characteristic F of all connected low-cost sensor signals L and the characteristic F of the high-cost sensor signal H The set of joint distributions of , and c(f Li ,f Hj ) is a function defined in the feature f Li and f Hj In order to ensure the rationality of the optimal mapping, the calculation formula of the transportation cost in the transportation cost submodule in this embodiment is:

[0037]

[0038] Among them, f Li is the feature of the low-cost sensor signal in the i-th dimension, f Hj is the feature of the high-cost sensor signal in the jth dimension, c(f Li ,f Hj ) is the transportation cost.

[0039] The design of the transportation cost function takes into account the influence of the inner product (i.e., similarity) between features: the more similar the features are, the lower the cost is, thereby guiding the optimal mapping to align similar features.

[0040] Due to the computational complexity of the optimal transportation problem, this embodiment uses the Sinkhorn algorithm to approximate the solution to the above optimization problem. The Sinkhorn algorithm can obtain a better approximate solution within a reasonable computing time by introducing entropy regularization. Through the Sinkhorn algorithm, this embodiment can obtain an optimal mapping T:X→Y from low-cost sensor signal features to high-cost sensor signal features. This mapping allows this embodiment to identify and align the most similar features in the feature space of the generated model, thereby providing effective supervision information for unpaired data. Based on this mapping, this embodiment defines the OTS loss function To exert appropriate supervision, the features of low-cost sensor signals are encouraged to align with the features of certain high-cost sensor signals.

[0041] Specifically, a specific supervision mechanism is applied in the feature alignment submodule; the OTS loss function after applying the specific supervision mechanism specifically includes:

[0042]

[0043] Among them, P L and P H denote the domain distribution of low-cost sensor signals and high-cost sensor signals, respectively, and F H (x H ) is the characteristic of the high-cost sensor signal, F L (x L ) is the characteristic of the low-cost sensor signal, for, is the low cost signal x L After the generator G L→H The virtual high-cost signal feature generated after is different from the real high-cost signal feature F H (x H ) are compared and the potential consistency constraints between them are established with the help of the optimal transport mapping T. is a high cost signal x H After the generator G H→L The virtual low-cost signal feature generated after is different from the real low-cost signal feature F L (x L ) and use the optimal transport inverse mapping T -1 Establish potential consistency constraints between the two.

[0044] By minimizing The generative model is guided to generate low-cost sensor signal features that are aligned as closely as possible with high-cost sensor signal features, thereby achieving signal enhancement.

[0045] To ensure the effectiveness of the OTS mechanism, this embodiment provides a rigorous mathematical proof of the existence of the optimal mapping and demonstrates its convergence during training. In addition, this embodiment also found that in the traditional GAN ​​architecture, the conflict between the goals of the generator and the discriminator often leads to oscillations and instability in the gradient update, while the OTS mechanism effectively reduces the oscillations during training by finely controlling the feature space of the generated signal, ensuring that the generator is more stably optimized to the optimal solution. These properties of the OTS mechanism enable the quality of the generated signal to be effectively improved in the case of unpaired data, and play a key role in low-cost accelerometer signal enhancement.

[0046] In some embodiments, in the task of low-cost accelerometer signal enhancement, in addition to the limitation of signal range, the lack of signal details is also a problem that needs to be solved. Low-cost sensors usually have low sensitivity and are difficult to capture subtle signal changes, resulting in a smoother signal that lacks necessary details. This phenomenon is particularly obvious in high-dynamic environments, which directly affects the reliability of the signal and application performance.

[0047] Although traditional generative adversarial networks (GANs) perform well in generating signals, they often have difficulty generating signals with rich details. This is because the signals generated by GANs tend to be globally smooth and ignore local subtle changes. To solve this problem, this embodiment proposes a "Modulated Laplace Regularization" (MLE) mechanism, which aims to enhance the details of the signal by regulating the local energy of the generated signal features, thereby improving the overall quality of the signal. Figure 3 shown.

[0048] The core of the MLE mechanism is to use the Laplace operator to measure the local changes in the feature layer and enhance these changes through appropriate energy modulation. The Laplace operator is a second-order differential operator that can effectively capture subtle changes and local fluctuations in the signal. Specifically, given the feature h (n) (represented as the nth layer feature in the network, with dimension d), this embodiment designs a Laplace energy of a neural network feature The calculation formula of Laplace energy in the MLE module is:

[0049]

[0050] in, Features The second-order derivative in the i-th dimension; d is the dimension, n is the number of layers; h(n) Represented as the nth layer feature in the network; is the Laplace energy. Defined as:

[0051]

[0052] is the n-th layer feature in the i-th dimension; is the n-th layer feature in the i+1th dimension; is the n-th layer feature in the i-1th dimension.

[0053] The higher the Laplace energy, the stronger the volatility and the more significant the local variation of the feature, while the lower the Laplace energy, the smoother the feature and the less variation. Therefore, by regulating the Laplace energy, this embodiment can guide the generation network to enhance the local details of the signal when necessary, and maintain the smoothness of the signal when not necessary.

[0054] In some embodiments, the present application further includes an energy modulation module for modulating the Laplace energy based on an energy modulation regularization term.

[0055] Specifically, in order to reasonably regulate the Laplace energy, this embodiment designs an energy modulation regularization term R MLE , which is used to dynamically adjust the Laplace energy of the feature layer during the generation process, so as to avoid excessive signal noise caused by too high energy, or loss of signal details caused by too low energy.

[0056] The formula expression of the energy modulation regularization term in the energy modulation module is:

[0057] R MLE = -log(E Laplace )-κ·log(1-E Laplace ).

[0058] Among them, σ is the Sigmoid function, which is used to normalize the Laplace energy to the (0,1) interval, κ is a modulation parameter, and the modulation formula of the modulation parameter κ is specifically:

[0059]

[0060] In the formula, d is the dimension, n is the number of layers; h (n) Represented as the nth layer feature in the network, for The mean of

[0061] In this embodiment, κ is designed as the kurtosis of the feature to measure the volatility of the feature. The κ parameter determines the minimum position of the regularization term and the degree of Laplace energy injection. Specifically, when the volatility of the feature is strong (κ is large), less energy injection is required. At this time, R MLE The minimum value of will be located at a lower Laplace energy value; on the contrary, when the volatility of the feature is weak (κ is small), more energy injection is required, and R MLE The minimum of will be located at higher Laplace energy values. This adaptive behavior ensures that the MLE mechanism can dynamically adjust the energy injection according to the volatility of the feature, thereby enhancing signal details in a high dynamic environment and reducing noise in a stable state.

[0062] When constructing the MLE mechanism, this embodiment pays special attention to the mathematical properties of its regularization term. First, R MLE In E Laplace When it approaches 0 or 1, it tends to infinity, thus imposing a strong penalty on extremely low or extremely high Laplace energies, ensuring that the Laplace energy remains in a moderate range. Secondly, the modulation parameter κ controls the amount of injected energy, so that less energy is injected when the characteristic volatility is strong, and more energy is injected when the volatility is weak. This adaptive energy modulation mechanism allows the generated signal to show sufficient local details without introducing additional noise due to excessive energy injection. By minimizing R MLE , the generative network can effectively improve the local details of the signal while maintaining the overall smoothness of the signal, thereby generating a more accurate and realistic signal. This mechanism plays a vital role in low-cost accelerometer signal enhancement and effectively solves the shortcomings of traditional generative adversarial networks in signal detail enhancement.

[0063] In actual experiments, this embodiment verifies the effectiveness of the MLE mechanism through multiple data sets and evaluation indicators. The experimental results show that the HEROS-GAN model with the MLE mechanism not only significantly outperforms other methods in signal details, but also shows extremely high reliability in the physical consistency of the signal. These experimental results prove the effectiveness of the MLE mechanism in low-cost accelerometer signal enhancement and further establish HEROS-GAN's leading position in this field.

[0064] In summary, this application has the following technical effects:

[0065] 1) This application proposes an innovative "Optimal Transport Supervision" (OTS) mechanism, which aims to solve the problem of signal enhancement of low-cost inertial sensor signals in the case of unpaired data. OTS maximizes the potential supervision information in unpaired data by utilizing the optimal transport theory to construct the optimal mapping between low-cost and high-cost sensor signal features. Specifically, the OTS module guides the generator to achieve optimal alignment of signal features in the feature space by calculating the transportation cost between features. This mechanism effectively overcomes the challenge of insufficient supervision of traditional GAN ​​architectures in the case of unpaired data, and significantly improves the learning ability and signal enhancement effect of the generation model.

[0066] 2) The "Energy Modulated Laplace Regularization" (MLE) mechanism in this application is a novel feature layer energy regulation method that aims to enhance the local detail performance of the generated signal. MLE captures subtle changes in the signal by calculating the Laplace energy of the neural network feature layer, and dynamically adjusts the energy input in combination with the volatility of the features (such as kurtosis). This adaptive adjustment mechanism can not only effectively enhance signal details in a highly dynamic environment, but also avoid the introduction of noise caused by over-adjustment. MLE ensures the rationality and stability of the Laplace energy through rigorous mathematical derivation, further improving the physical consistency and realism of the generated signal.

[0067] 3) This application proposes a bidirectional generation architecture (HEROS-GAN) that combines OTS and MLE mechanisms. L→H and G H→L ) work together to achieve bidirectional mapping between low-cost and high-cost sensor signals. Different from the traditional CycleGAN, HEROS-GAN introduces the OTS module in the generation process to enhance the supervision information of unpaired data, and uses the MLE module to optimize the local details of the generated signal. This bidirectional architecture design not only ensures the global consistency of the generated signal, but also significantly improves the reconstruction accuracy and quality of the low-cost accelerometer signal through multi-level detail enhancement.

[0068] 4) The HEROS-GAN architecture proposed in this application combines the optimal transport theory with the energy regulation mechanisms OTS and MLE for the first time, innovatively solving the problems of insufficient supervision information and missing details in traditional GAN ​​models in low-cost sensor signal enhancement. This solution obtains the maximum potential supervision information through the OTS mechanism, and achieves precise energy regulation in the feature layer through the MLE mechanism to ensure the consistency and accuracy of the generated signal in global and local details. The uniqueness of this solution is that it can effectively enhance the signal quality of low-cost inertial sensors without relying on frame-by-frame pairing data, providing a cost-effective solution for a wide range of industrial, medical and smart device applications.

[0069] 5) This application realizes the reconstruction of inertial sensor over-range signals for the first time.

[0070] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0071] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. An inertial sensor over-range signal repair and signal quality enhancement system, characterized in that: include: Generator GAN L→H , Generator GAN H→L , OTS module, MLE module and signal acquisition module; The signal acquisition module is used to acquire a high-cost sensor signal and a low-cost sensor signal; the correlation between the low-cost sensor signal and the high-cost sensor signal is shown as unpaired or weakly paired; The generator GAN H→L , used to convert high-cost sensor signals into low-cost sensor signals; The generator GAN L→H , used to convert low-cost sensor signals into high-cost sensor signals; The MLE module is used to: L→H When converting low-cost sensor signals into high-cost sensor signals, Laplace energy is injected into the low-cost sensor signals, and the generator GAN H→L When converting a high-cost sensor signal into a low-cost sensor signal, injecting Laplace energy into the high-cost sensor signal; the Laplace energy is a Laplace energy of a neural network; the Laplace energy is used to adjust the Laplace energy of a model in a generative deep learning architecture; The OTS module is used to mine the potential correlation between unpaired and weakly paired data based on the optimal transport theory, and to construct an optimal mapping between the features of the low-cost sensor signal and the features of the high-cost sensor signal according to the potential correlation.

2. The inertial sensor over-range signal repair and signal quality enhancement system according to claim 1, characterized in that: The calculation formula of Laplace energy in the MLE module is: in, Features The second-order derivative in the i-th dimension; d is the dimension, n is the number of layers; h (n) Represented as the nth layer feature in the network; is the Laplace energy.

3. The inertial sensor over-range signal repair and signal quality enhancement system according to claim 1, characterized in that: An energy modulation module is also included, which is used to modulate the Laplace energy based on an energy modulation regularization term.

4. The inertial sensor over-range signal repair and signal quality enhancement system according to claim 3, characterized in that: The formula expression of the energy modulation regularization term in the energy modulation module is: R MLE j-log(E Laplace )-κ·log(1-E Laplace )4 in, σ is the Sigmoid function, which is used to normalize the Laplace energy to the (0,1) interval, and κ is the modulation parameter.

5. The inertial sensor over-range signal repair and signal quality enhancement system according to claim 4, characterized in that: The modulation formula of the modulation parameter κ is specifically: In the formula, d is the dimension, n is the number of layers; h (n) Represented as the nth layer feature in the network, for The mean of .

6. The inertial sensor over-range signal repair and signal quality enhancement system according to claim 1, characterized in that: The OTS module includes a transportation cost submodule, a feature alignment submodule and an optimal mapping calculation submodule; The transportation cost submodule is used to calculate the transportation cost between the characteristics of the low-cost sensor signal and the characteristics of the high-cost sensor signal; the transportation cost is determined according to the similarity between the characteristics of the low-cost sensor signal and the characteristics of the high-cost sensor signal; The feature alignment submodule is used to align the features of the low-cost sensor signal with the features of the high-cost sensor signal; The optimal mapping calculation submodule is used to determine the optimal mapping of the features of the low-cost sensor signal to the features of the high-cost sensor signal according to the minimum transportation cost when the features of the low-cost sensor signal are aligned with the features of the high-cost sensor signal.

7. The inertial sensor over-range signal repair and signal quality enhancement system according to claim 6, characterized in that: The calculation formula of the transportation cost in the transportation cost submodule is: Among them, f Li is the feature of the low-cost sensor signal in the i-th dimension, f Hj is the feature of the high-cost sensor signal in the jth dimension, c(f Li ,f Hj ) is the transportation cost.

8. The inertial sensor over-range signal repair and signal quality enhancement system according to claim 7, characterized in that: A specific supervision mechanism is applied in the feature alignment submodule; the OTS loss function after applying the specific supervision mechanism specifically includes: Among them, P L and P H denote the domain distribution of low-cost sensor signals and high-cost sensor signals, respectively, and F H (x H ) is the characteristic of the high-cost sensor signal, F L (x L ) is the characteristic of the low-cost sensor signal, is the low cost signal x L After the generator G L→H The virtual high-cost signal characteristics generated later, is a high cost signal x H After the generator G H→L The virtual low-cost signal characteristics generated later, T is the optimal transportation mapping.

Images