Signal analysis method and system based on obtaining and identifying noise panoramic distribution model

Abstract

The present invention belongs to the technical field of signal analysis, and particularly relates to a signal analysis method and system based on obtaining and identifying a noise panoramic distribution model, including the following steps: Under a rich condition measurement environment, repeatedly measure a reference sample and a test sample to respectively obtain a plurality of measurement results; Process the measurement results of the reference sample and the test sample to respectively form training data of the reference sample and the test sample; Based on the training data of the reference sample and the test sample, enable the model to identify signals and noise from the measurement results and distinguish the reference sample and the test sample; Input the measurement results of the sample to be identified into the trained artificial intelligence model, and the output result is the specific type of the sample to be identified. Different from the noise processing solutions in the prior art, the present invention provides a signal analysis method based on obtaining and identifying a noise panoramic distribution model from a completely different technical perspective to solve the noise reduction problem difficult to handle in the prior art.

Description

Technical Field

[0001] The present invention belongs to the technical field of signal analysis, and particularly relates to a signal analysis method and system based on obtaining and identifying a panoramic distribution model of noise. Background Art

[0002] In response to the applications and requirements in many practical fields, there are already a wide variety of sample measurement methods in the prior art, such as electrocardiogram signals and electroencephalogram signals in the field of physiological detection, spectral signals in the field of substance detection, etc. However, limited by multiple factors such as the measurement environment, equipment accuracy, and the properties of the samples themselves, no matter which sample measurement method is selected, the measurement results are inevitably mixed data of signals and noise. For measurement results with a low signal-to-noise ratio, it is difficult to extract the signals therefrom, resulting in difficulties in comprehensively and effectively analyzing the signals, which will directly affect the accurate understanding of the samples.

[0003] To solve the above problems, the following two technical directions for dealing with noise are given in the prior art: 1. In the sample measurement stage, by measures such as improving equipment accuracy and improving the measurement environment, the noise is controlled or suppressed so that the signal intensity far exceeds the noise intensity, thereby obtaining a measurement result with a high signal-to-noise ratio; 2. In the result analysis stage, a mathematical method is adopted to construct a mathematical model based on a pre-assumed noise statistical distribution, and the noise is removed by using the mathematical model to further improve the overall signal-to-noise ratio of the measurement result.

[0004] The above two methods can solve the noise problem in some cases to a certain extent. However, there are always two insurmountable defects:

[0005] First of all, the noise mixed in the signal during the sample measurement process may be non-linear and may also have quite complex forms and contents. For example, for the measurement results of image samples, the noise at different positions may be different; for the measurement results of audio samples, the noise on different tracks or even at different times on the same track may be different. Facing this complex noise, the conventional mathematical noise reduction methods in the existing engineering technology are difficult to directly carry out, that is, it is difficult to design a noise reduction scheme through one or several common mathematical models, resulting in the signal-to-noise ratio of the measurement result not being able to be improved to a level suitable for analysis;

[0006] Secondly, during the actual sample measurement process, in order to obtain measurement results with a high signal-to-noise ratio, the sample measurement environment, equipment accuracy, and other conditions may be comprehensively optimized. However, even so, limited by the sample's own properties and other objective factors, the measurement results may have the following properties: 1. The signal in the measurement results can be detected, that is, the signal intensity is above the measurable lower limit of the measurement equipment; 2. The signal is extremely weak, and its intensity is of the same order of magnitude as the noise intensity or even lower; 3. The characteristics of the signal itself are very complex. In measurement results with the above properties, the signal is very likely to be submerged by noise. Conventional mathematical noise reduction methods have difficulties in processing such measurement results because it is difficult to establish a reasonable mathematical model to simulate and remove the noise mixed in the measurement results, making it difficult to separate the signal from the noise, and even causing high-value signals to be eliminated along with the noise during the denoising process. Summary of the Invention

[0007] The main object of the present invention is to provide a signal analysis method based on obtaining and identifying the panoramic distribution model of noise, aiming to solve the technical problem that it is difficult to analyze complex signals under ultra-low signal-to-noise ratio conditions in the prior art.

[0008] To achieve the above object, the technical solution of the present application is as follows:

[0009] A signal analysis method based on obtaining and identifying the panoramic distribution model of noise, comprising the following steps:

[0010] S1: Under rich-condition measurement environments, repeatedly measure the reference sample and the test sample to obtain multiple measurement results respectively; wherein, each measurement result includes a signal and different noise profiles; wherein, rich-condition refers to measurement conditions that are not aimed at maintaining external condition consistency, do not involve noise suppression, are natural, and include real complex noise factors.

[0011] S2: Process the measurement results of the reference sample and the test sample to respectively form the training data of the reference sample and the test sample; wherein, the training data includes a noise panorama or at least part of the noise panorama composed of multiple noise profiles;

[0012] S3: Based on the training data of the reference sample and the test sample, with the observability presentation of noise as the convergence target, conduct artificial intelligence model training to enable the model to identify the signal and noise from the measurement results and distinguish the reference sample and the test sample;

[0013] S4: Input the measurement result of the sample to be identified into the trained artificial intelligence model, and the output result of the artificial intelligence model is the specific type of the sample to be identified.

[0014] Optionally, in step S1, before each measurement of the reference sample and the test sample, a rich-condition measurement environment is created by introducing slight perturbations, thereby increasing the noise observation dimension and making the measurement results of each measurement contain different noise profiles.

[0015] Furthermore, the slight perturbations can be selected but are not limited to spatial perturbations, temporal perturbations, physical perturbations, and environmental perturbations; spatial perturbations include but are not limited to: causing slight displacement of the measurement site, causing slight rotation of the measurement site; temporal perturbations include but are not limited to: increasing the measurement duration, shortening the measurement duration, and changing the time interval between multiple measurements; physical perturbations include but are not limited to: vibrating the measurement device or the sample during measurement, agitating the fluid sample; environmental perturbations include but are not limited to: changing the environmental temperature during measurement, changing the environmental humidity during measurement, changing the electromagnetic field during measurement, and changing the air pressure during measurement.

[0016] Furthermore, in step S2, the steps of processing the measurement results of the reference sample and the test sample to respectively form the training data of the reference sample and the test sample include:

[0017] S21. Normalize the measurement results of the reference sample and the test sample;

[0018] S22. Based on the normalization result of step S21, establish a posterior probability model framework;

[0019] After the measurement results of the reference sample and the test sample are processed through steps S21 - 22, the qualified training data will be respectively formed for subsequent artificial intelligence model training.

[0020] During the process of forming the training data from the measurement results of the reference sample and the test sample, different noise profiles constitute a noise panorama or at least part of the noise panorama. At the same time, the overall measurement results of the two types of samples, as well as the signals in the measurement results, will respectively exhibit stable statistical characteristics; the statistical distribution pattern presented by the noise will also tend to be stable with the construction of the noise panorama.

[0021] In step S3, the artificial intelligence model can be selected but is not limited to: artificial neural network, perceptron, support vector machine, Bayesian classifier, Bayesian network, random forest model, or clustering model.

[0022] In step S3, during the training process of the artificial intelligence model, the model will iteratively conduct a large amount of empirical learning, induction, and convergence on the features in the training data that can achieve signal - noise recognition and the features that can achieve the distinction between the reference sample and the test sample, and learn the relationship between the features and the preset labels.

[0023] Specifically, the features capable of signal recognition include the statistical distribution patterns presented after processing multiple measurement results, which conform to the true mathematical statistical laws of the signal; the features capable of noise recognition include the statistical distribution patterns presented by the noise panorama or at least part of the noise panorama constructed from diverse noise profiles, which approach the true mathematical statistical laws of the noise; the features capable of distinguishing reference samples and test samples include the statistical distribution patterns presented respectively after processing the multiple measurement results of the reference samples and test samples.

[0024] Furthermore, the preset tags include output tags and input tags. Among them, the output tags include two tags representing the reference sample and the test sample respectively; the input tags are two sets of coupled tags for the training data related to the reference sample and the test sample respectively, and each coupled tag is associated with the rich-condition measurement environment when the sample is measured.

[0025] Specifically, each coupled tag in different groups respectively represents: the coupling of the measurement result of the reference sample or the test sample with the noise panorama under each independent measurement environment in the rich-condition measurement environment; among them, the noise profile included in the measurement result is the noise profile obtained under this independent measurement environment.

[0026] The present invention also provides a signal analysis system based on obtaining and identifying the noise panorama distribution model, including a measurement module, a processing module, a training module, and an analysis module;

[0027] Under the rich-condition measurement environment, the measurement module repeatedly measures the reference sample and the test sample, and respectively obtains multiple measurement results; among them, each measurement result includes a signal and different noise profiles; where the rich condition refers to the measurement condition that is not aimed at maintaining the consistency of external conditions, does not involve suppressing noise, is natural, and includes real complex noise factors.

[0028] The processing module processes the measurement results of the reference sample and the test sample, and respectively forms the training data of the reference sample and the test sample; among them, the training data includes the noise panorama or at least part of the noise panorama composed of multiple noise profiles;

[0029] Based on the training data of the reference sample and the test sample, the training module takes the observability presentation of the noise as the convergence target, and conducts artificial intelligence model training, so that the model can identify the signal and the noise from the measurement results, and distinguish the reference sample and the test sample;

[0030] For the measurement result of the sample to be identified, the analysis module inputs it into the trained artificial intelligence model, and the output result of the artificial intelligence model is the specific type of the sample to be identified.

[0031] Optionally, the measurement module includes a perturbation mechanism. Before each measurement of the reference sample and the test sample by the measurement module, the perturbation mechanism introduces a slight perturbation to create a rich-condition measurement environment, thereby increasing the noise observation dimension of the sample measurement and making the measurement results of each sample measurement contain different noise profiles.

[0032] Furthermore, before each measurement of the reference sample and the test sample, the slight perturbation introduced by the perturbation mechanism can be selected but is not limited to spatial perturbation, temporal perturbation, physical perturbation, and environmental perturbation.

[0033] Spatial perturbation includes but is not limited to: slightly displacing the measurement site, slightly rotating the measurement site; Temporal perturbation includes but is not limited to: increasing the measurement duration, shortening the measurement duration, and changing the time interval between multiple measurements; Physical perturbation includes but is not limited to: vibrating the measurement device or the sample during measurement, agitating the fluid sample; Environmental perturbation includes but is not limited to: changing the environmental temperature during measurement, changing the environmental humidity during measurement, changing the electromagnetic field during measurement, and changing the air pressure during measurement.

[0034] Furthermore, the processing module includes a normalization module and a posterior probability module;

[0035] Among them, the normalization module normalizes the measurement results of the reference sample and the test sample and outputs the normalization results respectively; The posterior probability module, based on the normalization results, establishes a posterior probability model framework and forms the training data of the reference sample and the test sample that meet the requirements respectively for subsequent artificial intelligence model training.

[0036] In the process of the processing module processing the measurement results of the reference sample and the test sample to form training data, different noise profiles constitute a noise panorama or at least part of a noise panorama. At the same time, the overall measurement results of the two types of samples and the signals in the measurement results will respectively exhibit stable statistical characteristics; The statistical distribution pattern presented by the noise will also tend to be stable as the noise panorama is constructed.

[0037] Furthermore, the artificial intelligence model can be selected but is not limited to: artificial neural network, perceptron, support vector machine, Bayesian classifier, Bayesian network, random forest model, or clustering model.

[0038] During the training process of the artificial intelligence model, the model will iteratively conduct a large amount of empirical learning, induction, and convergence on the features in the training data that can achieve signal and noise recognition and the features that can achieve the distinction between the reference sample and the test sample, and learn the relationship between the features and the preset labels.

[0039] Specifically, features that enable signal identification include statistical distribution patterns that are presented after multiple measurement results are processed and that conform to the true mathematical statistical laws of the signal; features that enable noise identification include statistical distribution patterns that are presented by a noise panorama or at least part of the noise panorama constructed by diverse noise profiles and that are close to the true mathematical statistical laws of noise; features that enable distinction between reference samples and test samples include statistical distribution patterns that are presented after multiple measurement results of the reference sample and the test sample are processed.

[0040] Furthermore, the preset labels include output labels and input labels. The output labels include two labels representing the reference sample and the test sample respectively; the input labels are two sets of coupling labels involving the training data of the reference sample and the test sample respectively, and each coupling label is associated with the rich condition measurement environment in which the sample is measured.

[0041] Specifically, each coupling label of different groups represents: the coupling between the measurement results of the reference sample or the test sample and the noise panorama in each independent measurement environment in the rich condition measurement environment; wherein the noise profile contained in the measurement result is the noise profile obtained in this independent measurement environment.

[0042] The beneficial effects of this application are:

[0043] 1. The present invention is different from the noise processing solutions in the prior art. From a completely different technical perspective, it provides a signal analysis method based on acquiring and identifying a noise panoramic distribution model to solve the noise reduction problem that is difficult to handle in the prior art.

[0044] The field of signal detection often involves signals that have analytical value but are submerged in noise due to their extremely weak intensity and / or extremely complex characteristics. In this case, the distribution model of the noise cannot be reasonably assumed, making the existing noise reduction methods that mathematically model the noise signal difficult to effectively implement.

[0045] The signal analysis method provided by the present invention is based on mathematical statistical principles and does not directly separate signals from noise, but it can still effectively distinguish noise from signals and realize the successful identification of multiple independent signals based on different measurement samples, thereby carrying out practical applications such as sample detection and material classification. In addition, the present invention uses artificial intelligence technology to perform mixed modeling of noise and the signal submerged therein. Even if there is no mathematical hypothesis for noise, the trained artificial intelligence model can deeply explore the mathematical statistical laws hidden in the measurement results and accurately obtain the mathematical distribution model of signals and noise.

[0046] 2. In the sample measurement stage of the present invention, instead of setting the measurement conditions to be consistent, a variety of measurement conditions are created to form multiple noise profiles that vary due to changes in the measurement conditions. Multiple noise profiles are combined to form a noise panorama or at least part of a noise panorama, and then the mathematical distribution model of the noise is identified. This operation method will not cause signal loss or misdeletion, and avoids the influence of the signal itself caused by the commonly used noise reduction steps in the prior art.

[0047] In the technical solution disclosed by the present invention, through a large number of repeated measurements under diverse measurement conditions, the data distribution form of the relatively stable signal can be presented more clearly, thereby improving the visibility of the signal in the measurement results and being beneficial to subsequent signal extraction and analysis. On the other hand, the perturbation environment provided by the diverse measurement conditions provides different noise observation dimensions for each sample measurement, ensuring the sample randomness of the noise. On this basis, a noise panorama can be obtained through a large number of repeated measurements, that is, a large number of noise "samples" can almost comprehensively cover all possibilities of the noise itself. At the same time, the distribution model of the noise will also tend to its true distribution form.

[0048] The present invention can discover the mathematical statistical laws of the noise from the mixed data distribution form of the sample measurement results, and distinguish the noise and the signal and identify different types of signals from the perspective of the data distribution model. Based on this technical idea, in the sample measurement results, the noise and the signal will respectively present their true mathematical statistical laws. Compared with the direct noise removal and signal extraction in the existing engineering technology, the present invention deeply explores the mathematical statistical laws of the noise and the signal, can avoid the miselimination of the signal caused by the noise reduction operation, and ensure the validity of the data. Therefore, the noise will not interfere with the signal analysis and will not affect the identification and classification between independent signals. It can be seen that for the problems that the prior art cannot solve, such as removing the noise itself or extracting the signal itself from the mixed sample measurement results, the present invention provides an effective solution.

[0049] 3. Since noise is an inevitable influencing factor in the actual sample measurement process, even under the currently most excellent sample measurement conditions, the measurement results obtained from each sample measurement may infinitely approach the true signal, but always only change "statistically stably" near the signal, and the "statistically stable" change of this measurement result and the noise therein is unpredictable, that is, the exact value of the next sample measurement result cannot be assumed.

[0050] However, after multiple repeated acquisitions, a large number of measurement results will overall present a data distribution model that tends to be stable. This stable data distribution model represents the macroscopic aggregation of the mutual influence of all components in the measurement results. That is, in addition to the true signal, interference factors that may cause noise, such as environmental complexity, equipment accuracy, and the inherent influence of sampling methods, are all incorporated into the overall distribution model of the above-mentioned measurement results. Therefore, the overall distribution model of the measurement results can fully reflect its own characteristics. While the distribution model of the measurement results tends to be stable, a noise panorama or at least part of the noise panorama composed of a large number of noise profiles will also present unique mathematical statistical laws and approach the true distribution model of the noise. The present invention realizes the distinction between noise and signal by identifying the complete mathematical model of noise, and this identification scheme will obtain more comprehensive and accurate identification results.

[0051] 4. By deeply mining the data distribution model of noise using an artificial intelligence model, highly empirical and more accurate analysis results can be obtained.

[0052] For a large number of sample repeated measurements carried out under diverse measurement conditions, the noise profiles obtained can at least construct part of the noise panorama, and the data distribution model of the noise has tended to be able to reflect the theoretically true distribution model of the noise. In this case, the present invention uses artificial intelligence technology to explore the distribution model of the noise.

[0053] In the technical solution disclosed by the present invention, the sample measurement stage involves multiple sample measurements carried out under diverse perturbation conditions, thereby obtaining a large number of measurement results mixed with signals and noise. The above operations are conducive to obtaining the noise panorama, and at the same time, the huge measurement values also provide a sufficient data basis for the training of the artificial intelligence model. The trained artificial intelligence model can discover true features that meet the analysis requirements or purposes of the experimenter in high background noise data, can provide more efficient mathematical operations, and can output highly empirical and more accurate analysis results in real time.

[0054] 5. The present invention creates a rich-condition measurement environment through different means of introducing perturbations to increase the noise observation dimension in sample measurements, so that the measurement results under a large number of repeated measurement conditions can show a complete noise panorama, or can provide at least part of the noise panorama with sufficient accuracy for subsequent signal analysis. There are differences in the practical difficulties of different means of introducing perturbations, and different impacts may also be caused in terms of increasing the noise observation dimension.

[0055] In the practical application of the technical solution involved in the present invention, considering various factors such as the characteristics of the sample itself, the sample measurement method, and the measurement accuracy requirements, the experimenter can fully select from the perturbation introduction means provided by the present invention according to the actual needs. The diverse perturbation introduction means disclosed in the present invention provide the experimenter with a broad choice, and to a certain extent, reduce the application difficulty of the present invention, making this technical solution more valuable for popularization and application. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 FIG. is a schematic flowchart of a signal analysis method based on obtaining and identifying a noise panoramic distribution model provided by the present invention.

[0057] Figure 2 For the description in the appended Figure 1 FIG. is a schematic flowchart of step 2 in the signal analysis method shown.

[0058] Figure 3 FIG. is a schematic structural diagram of a signal analysis system based on obtaining and identifying a noise panoramic distribution model provided by the present invention.

[0059] Figure 4 FIG. is a schematic diagram of the principle of multiple noise profiles constituting a noise panorama or at least a part of the noise panorama. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0060] The purpose of signal processing is to extract useful information from the sample measurement results. For example, the content with research value or the differential features different from other signals. Limited by many uncertain factors in the sample measurement process, these "contents with research value" and "differential features" are often not represented by independent numerical values, but are reflected by the data statistical distribution of the overall signal.

[0061] In the measurement results obtained in the actual sample measurement process, in addition to the signals reflecting the true characteristics, there must be noise mixed. In the signal processing solutions disclosed in the prior art, either the influence of noise is eliminated from the measurement results, or the signal is extracted from the measurement results. However, when the noise mixed in the measurement results cannot be simulated by a "known" mathematical model, it is extremely difficult to eliminate the noise or extract the signal. Therefore, the technical objective of the present invention is to find the "unknown" distribution models of the signal and noise from the measurement results based on the data statistical principle, so as to effectively distinguish the signal from the noise. In addition, when the measurement results are from different measurement samples, the accurate identification of the sample type is achieved by finding the differential data distribution models.

[0062] For each sample measurement, the signals and noises in the obtained measurement results will be slightly different from those obtained in previous measurements. From the perspective of samples and sampling, each sample measurement is equivalent to a random sampling from the sample population, and the measurement results corresponding to random sampling cannot reflect the true characteristics. However, on the premise that the signals and noises respectively have specific data statistical laws and conform to specific data distribution models, when the number of measurements increases and the sampling range expands to approach the sample population, the overall data statistical laws reflected by a large number of test results can tend to reflect the true situation.

[0063] More specifically, 1. For the signals in the measurement results, since the signals reflect the inherent characteristics of the test samples themselves, they must have a clear statistical distribution model. This clear statistical distribution model can be clearly presented through a large number of data samplings; 2. For the noises in the measurement results, the existing technology generally believes that the "ideal" mathematical statistical law of noises conforms to the Gaussian distribution. However, the actual sample measurement process usually cannot create an "ideal" noise situation. In addition, even by means of improving equipment accuracy, improving material purity, etc., and optimizing the sample measurement conditions as much as possible, it may still be impossible to obtain measurement results with ideal analysis conditions. That is, the target signal to be analyzed is submerged in the noise due to its weak intensity, or the characteristics of the signal are extremely complex and difficult to analyze.

[0064] It is often very difficult to mine the data distribution model of such measurement results, and it is even completely impossible to assume the distribution model of the noises therein. In such a situation, the present invention realizes sample measurement under a perturbed environment by creating diverse measurement conditions and generates a noise profile from multiple observation dimensions. When the number of sample measurements is large enough, a full-range observation of the noises can be realized, that is, a noise panorama that can show the complete data statistical distribution model is constructed, and this data statistical distribution model will infinitely approach the true distribution of the noises. Thus, it can be seen that from a statistical perspective, it is completely theoretically feasible to show the statistical distribution model of the noises through the construction of a noise panorama under a perturbed environment.

[0065] When at least part of the noise panorama is obtained and the noises have shown a clear and stable distribution model, the present invention uses artificial intelligence technology to deeply explore the statistical laws of the noises. Artificial intelligence technology is an effective means applicable to various data analyses and solving empirical data processing. For example, the artificial intelligence deep learning model can simulate the learning process of humans, quickly summarize the empirical data processing methods of humans, and thus realize signal recognition and judgment behaviors. In the present invention, the accuracy of the empirical analysis results output by the artificial intelligence model trained by big data can be guaranteed, so as to effectively identify the mathematical distribution model of the noises and carry out specific analysis work such as subsequent noise separation and signal classification based on this.

[0066] In order to make the objectives, technical solutions and advantages of the present invention more clear and understandable, the present invention will be further described in detail below in conjunction with embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0067] Embodiment 1

[0068] A signal analysis method based on obtaining and identifying a noise panoramic distribution model, the process schematic diagram of which can be seen in the attached Figure 1 of the specification, and this method includes the following steps:

[0069] S1: Under a rich-condition measurement environment, repeatedly measure a reference sample and a test sample to respectively obtain a plurality of measurement results; wherein, each measurement result includes a signal and different noise profiles;

[0070] In the technical field of signal acquisition and analysis, maintaining the consistency of external conditions during the sample measurement process is a conventional means to reduce noise fluctuations and form a good signal-to-noise ratio, and repeatedly measuring the sample is also recognized as an effective way to reduce random errors. However, in the embodiments of the present invention, the setting of external condition consistency is not involved, but the repeated measurement of the reference sample and the test sample is carried out under a rich-condition measurement environment. Among them, rich-condition means: Rich-condition refers to a measurement condition that is not aimed at maintaining the consistency of external conditions, does not involve suppressing noise, is natural, and includes real complex noise factors. The purposes of both the "rich-condition measurement environment" and the "repeated measurement" are to obtain rich noise profiles sufficient to construct a noise panorama.

[0071] Specifically, under a rich-condition measurement environment, limited by the properties of the sample itself, the signal in the measurement results of the reference sample and the test sample will always remain statistically unchanged, but the noise will vary due to environmental changes, that is, the environmental changes will increase the observation dimension of the noise. Based on the noise observation dimensions with multi-aspect, multi-angle and multi-time-space characteristics, the repeated measurement of the reference sample and the test sample will form rich noise profiles. The rich noise profiles are the basis for constructing a noise panorama and identifying noise based on data statistical laws in the subsequent steps.

[0072] S2: Process the measurement results of the reference sample and the test sample to respectively form training data of the reference sample and the test sample; wherein, the training data includes a noise panorama or at least part of a noise panorama composed of a plurality of noise profiles;

[0073] As shown in the attached Figure 4As shown, in a single noise observation dimension, only the noise profile reflecting the local noise can be obtained, and a comprehensive noise observation result cannot be obtained. That is, the noise cannot present a complete data statistical law that conforms to its true distribution characteristics. However, in the embodiments of the present invention, the repeated measurements of the reference sample and the test sample are carried out in a rich-condition measurement environment. The rich noise profiles obtained under different noise observation dimensions will be sufficient to construct a noise panorama or at least a partial noise panorama. While constructing the noise panorama, the data statistical law of the noise will tend to its true mathematical statistical law.

[0074] In the embodiments of the present invention, the noise panorama means that the distribution model of the noise can already comprehensively reflect its theoretically true distribution model; the partial noise panorama means that the distribution model of the noise cannot completely reflect its theoretically true distribution model, but the distribution model already has the accuracy that can be used for subsequent signal analysis.

[0075] S3: Based on the training data of the reference sample and the test sample, with the observability presentation of the noise as the convergence target, perform artificial intelligence model training so that the model can identify signals and noise from the measurement results and distinguish the reference sample and the test sample.

[0076] In the embodiments of the present invention, the training data of the reference sample and the test sample will be randomly assigned as learning data and detection data at a preset ratio. Use the learning data to train the artificial intelligence model, input the detection data into the trained artificial intelligence model, calculate the signal recognition result. If the signal recognition accuracy is lower than the preset threshold, continue to train with the learning data. If the signal recognition accuracy is higher than the preset threshold, it is considered that the artificial intelligence model has completed training.

[0077] S4: Input the measurement result of the sample to be identified into the trained artificial intelligence model, and the output result of the artificial intelligence model is the specific type of the sample to be identified.

[0078] In the embodiments of the present invention, the reference sample and the test sample are two known samples. The training data respectively formed after processing multiple measurement results of the two can train an artificial intelligence model that can effectively distinguish the two known samples. When the sample to be identified is one of the two known samples, the artificial intelligence model can accurately identify the specific type of the sample to be identified.

[0079] Optionally, in step S1, before each measurement of the reference sample and the test sample, create a rich-condition measurement environment by introducing slight perturbations, thereby increasing the noise observation dimension and making the measurement results of each measurement contain different noise profiles.

[0080] Furthermore, the slight perturbations introduced before each measurement of the reference sample and the test sample can be selected but are not limited to spatial perturbation, temporal perturbation, physical perturbation, and environmental perturbation.

[0081] Spatial perturbations include but are not limited to: slightly displacing the measurement site, slightly rotating the measurement site; Temporal perturbations include but are not limited to: increasing the measurement duration, shortening the measurement duration, and changing the time interval between multiple measurements; Physical perturbations include but are not limited to: vibrating the measurement device or the sample during measurement, agitating the convective mass sample; Environmental perturbations include but are not limited to: changing the environmental temperature during measurement, changing the environmental humidity during measurement, changing the electromagnetic field during measurement, and changing the air pressure during measurement.

[0082] See the attached Figure 2 In step S2, the steps of processing the measurement results of the reference sample and the test sample to respectively form the training data of the reference sample and the test sample include:

[0083] S21. Normalize the measurement results of the reference sample and the test sample;

[0084] S22. Based on the normalization result of step S21, establish a posterior probability model framework.

[0085] After the measurement results of the reference sample and the test sample are processed through steps S21 - 22, the qualified training data will be respectively formed for subsequent artificial intelligence model training.

[0086] Specifically, in the embodiments of the present invention, for the measurement results of the reference sample and the test sample, they are regarded as the measured values obtained by measuring the measurement target composed of a complex system.

[0087] Define the measurement density function as where S is the measurement space dimension; V is the measurement environment; then in the measurement target, the number of systems is N, and N is defined by formula (1):

[0088]

[0089] Define B(V) as the measurement function, then the measured value There is:

[0090]

[0091] where

[0092]

[0093] Equation (3) is the normalization condition. In the embodiments of the present invention, in order to make the measurement results of the reference sample and the test sample satisfy the normalization condition of Equation (3), step S21 is adopted to perform normalization processing on the measurement results of the reference sample and the test sample.

[0094] Since the measurements of the reference sample and the test sample are repeated, and the repetition process is represented in a discrete manner, Equation (2) is rewritten in the form of an ensemble:

[0095]

[0096] Define H as the ensemble density function, then there is:

[0097] <v> ＝H <n>(5)

[0098] The statistical fluctuations of complex systems are as follows:

[0099]

[0100] Among them, corresponding to repeated measurements, δS is the information entropy of the measurement, and δP is the environmental change amount of the measurement.

[0101] Taking δP as the statistical space of the noise panorama and δS as the statistical space of the signal. Therefore, according to Bayes' formula, we have:

[0102]

[0103] In Equation (7), it is defined as Equation (8), and Equation (8) is the posterior probability condition. In the embodiments of the present invention, in order to make the measurement results of the reference sample and the test sample satisfy the posterior probability condition of Equation (8), step S22 is adopted to establish a posterior probability model framework based on the normalization result obtained in step S21.

[0104] Then, the estimation δn of the statistical fluctuations of the complex system * is as follows:

[0105] δn * = argmax δn P(H <n>|δn)P(δn) (9)

[0106] In the embodiments of the present invention, the measurement results processed through steps S21 - S22 can meet the normalization condition of formula (3) and the posterior probability condition of formula (8). The measurement results that meet the above two conditions can be used to implement the estimation of the statistical fluctuations of complex systems in formula (9). The measurement results that meet the above two conditions will be used as training data for subsequent artificial intelligence model training steps.

[0107] In the process of forming training data from the measurement results of reference samples and test samples, different noise profiles constitute a noise panorama or at least part of a noise panorama. At the same time, the overall measurement results of the two types of samples and the signals in the measurement results will respectively exhibit stable statistical characteristics; the statistical distribution patterns presented by the noise will also tend to be stable as the noise panorama is constructed.

[0108] In step S3, the artificial intelligence model can be selected but not limited to: artificial neural network, perceptron, support vector machine, Bayesian classifier, Bayesian network, random forest model or clustering model.

[0109] In the embodiments of the present invention, the estimation of the statistical fluctuations of complex systems in the above formula (9) will be implemented by an artificial intelligence model.

[0110] Embodiment 2

[0111] A signal analysis system based on obtaining and identifying a noise panorama distribution model, for its structural schematic diagram, see the attached Figure 3 description. The system includes a measurement module 1, a processing module 2, a training module 3 and an analysis module 4;

[0112] In a rich - condition measurement environment, the measurement module 1 repeatedly measures reference samples and test samples, respectively obtaining multiple measurement results; among them, each measurement result contains a signal and different noise profiles;

[0113] In the technical field of signal acquisition and analysis, maintaining the consistency of external conditions during the sample measurement process is a conventional means to reduce noise fluctuations and form a good signal - to - noise ratio, and repeatedly measuring samples is also recognized as an effective way to reduce random errors. However, in the embodiments of the present invention, the setting of external condition consistency is not involved. Instead, in a rich - condition measurement environment, the measurement module 1 conducts repeated measurements on reference samples and test samples. Among them, rich - condition means: a measurement condition that is not aimed at maintaining external condition consistency, does not involve noise suppression, is natural, and includes real complex noise factors. The purposes of "rich - condition measurement environment" and "repeated measurement" are both to obtain rich noise profiles sufficient to construct a noise panorama.

[0114] Specifically, in a rich-condition measurement environment, due to the limitations of the sample's own attributes, the signals in the measurement results of the reference sample and the test sample will always remain statistically invariant, but the noise will vary due to environmental changes, that is, the environmental changes will increase the observation dimension of the noise. Based on the noise observation dimensions with multi-aspect, multi-angle, and multi-space-time characteristics, the measurement module 1 repeatedly measures the reference sample and the test sample, and a rich noise profile will be formed. The rich noise profile is the basis for subsequently constructing a noise panorama and identifying noise based on data statistical laws.

[0115] The processing module 2 processes the measurement results of the reference sample and the test sample to respectively form the training data of the reference sample and the test sample; wherein, the training data includes a noise panorama composed of multiple noise profiles or at least part of the noise panorama;

[0116] As shown in the Figure 4 specification appendix, in a single noise observation dimension, only a noise profile reflecting the local part of the noise can be obtained, and a comprehensive noise observation result cannot be obtained, that is, the noise cannot present a complete data statistical law that conforms to its true distribution characteristics. However, in the embodiments of the present invention, the repeated measurement of the reference sample and the test sample is carried out in a rich-condition measurement environment. The rich noise profiles obtained under different noise observation dimensions will be sufficient for the processing module to construct a noise panorama or at least part of the noise panorama. While the processing module 2 constructs the noise panorama, the data statistical law of the noise will tend to its true mathematical statistical law.

[0117] In the embodiments of the present invention, the noise panorama means that the distribution model of the noise can already comprehensively reflect its theoretically true distribution model; the partial noise panorama means that the distribution model of the noise cannot completely reflect its theoretically true distribution model, but the distribution model already has the accuracy that can be used for subsequent signal analysis.

[0118] Based on the training data of the reference sample and the test sample, the training module 3 takes the observability presentation of the noise as the convergence target to perform artificial intelligence model training, so that the model can identify signals and noise from the measurement results and distinguish the reference sample and the test sample;

[0119] In the embodiments of the present invention, the training data of the reference sample and the test sample will be randomly assigned as learning data and detection data at a preset ratio. The training module 3 uses the learning data to train the artificial intelligence model, and inputs the detection data into the trained artificial intelligence model to calculate the signal recognition result. If the signal recognition accuracy rate is lower than the preset threshold, continue to train with the learning data. If the signal recognition accuracy rate is higher than the preset threshold, it is considered that the artificial intelligence model has completed training.

[0120] For the measurement results of the samples to be recognized, the analysis module 4 inputs them into the trained artificial intelligence model, and the output result of the artificial intelligence model is the specific type of the samples to be recognized.

[0121] In the embodiments of the present invention, the reference sample and the test sample are two known samples. The training data respectively formed after processing the multiple measurement results of the two are used to train an artificial intelligence model that can effectively distinguish the two known samples. When the sample to be recognized is one of the two known samples, the artificial intelligence model can accurately identify the specific type of the sample to be recognized.

[0122] Optionally, the measurement module 1 includes a perturbation mechanism 11. Before each measurement of the reference sample and the test sample by the measurement module 1, the perturbation mechanism 11 introduces slight perturbations to create a rich-condition measurement environment, thereby increasing the noise observation dimension of the sample measurement and making the measurement results of each sample measurement contain different noise profiles.

[0123] Further, before each measurement of the reference sample and the test sample, the slight perturbations introduced by the perturbation mechanism 11 can be selected but are not limited to spatial perturbation, time perturbation, physical perturbation, and environmental perturbation.

[0124] Spatial perturbation includes but is not limited to: slightly displacing the measurement site, slightly rotating the measurement site; time perturbation includes but is not limited to: increasing the measurement duration, shortening the measurement duration, and changing the time interval between multiple measurements; physical perturbation includes but is not limited to: vibrating the measurement device or the sample during measurement, agitating the fluid sample; environmental perturbation includes but is not limited to: changing the environmental temperature during measurement, changing the environmental humidity during measurement, changing the electromagnetic field during measurement, and changing the air pressure during measurement.

[0125] Further, the processing module 2 includes a normalization module 21 and a posterior probability module 22;

[0126] Among them, the normalization module 21 performs normalization processing on the measurement results of the reference sample and the test sample, and respectively outputs the normalization results; the posterior probability module 22 builds a posterior probability model framework based on the normalization results, and respectively forms the training data of the reference sample and the test sample that meet the requirements for subsequent artificial intelligence model training.

[0127] Specifically, in the embodiments of the present invention, for the measurement results of the reference sample and the test sample, they are regarded as the measured values obtained by measuring the measurement target composed of a complex system.

[0128] Define the measurement density function as where S is the measurement space dimension; V is the measurement environment; then in the measurement target, the number of systems is N, and N is defined by Equation (1):

[0129]

[0130] Define B(V) as the measurement function, then the measured value There is:

[0131]

[0132] Wherein,

[0133]

[0134] Equation (3) is the normalization condition. In the embodiments of the present invention, in order to make the measurement results of the reference sample and the test sample satisfy the normalization condition of Equation (3), the normalization module 21 performs normalization processing on the measurement results of the reference sample and the test sample, and outputs the normalization result;

[0135] Since the measurements of the reference sample and the test sample are repeated, and the repetition process is represented in a discrete manner, Equation (2) is rewritten in the form of an ensemble:

[0136]

[0137] Define H as the ensemble density function, then there is:

[0138] <v> ＝H <n>(5)

[0139] The statistical fluctuations of complex systems are as follows:

[0140]

[0141] Among them, for repeated measurements, δS is the information entropy of the measurement, and δP is the environmental change amount of the measurement.

[0142] Regarding δP as the statistical space of the noise panorama and δS as the statistical space of the signal. Therefore, according to Bayes' formula, we have:

[0143]

[0144] In Equation (7), it is defined that is Equation (8), and Equation (8) is the posterior probability condition. In the embodiments of the present invention, in order to make the measurement results of the reference sample and the test sample satisfy the posterior probability condition of Equation (8), the posterior probability module 22 establishes a posterior probability model framework based on the normalization result.

[0145] Then the estimation δn of the statistical fluctuations of the complex system * is as follows:

[0146] δn * = argmax δn P(H <n>|δn)P(δn) (9)

[0147] In the embodiment of the present invention, the measurement results processed by the normalization module 21 and the posterior probability module 22 can satisfy the normalization condition of Equation (3) and the posterior probability condition of Equation (8). The measurement results that satisfy the above two conditions can be used to implement the estimation of the statistical fluctuations of a complex system in Equation (9). The measurement results that satisfy the above two conditions will be used as training data for subsequent artificial intelligence model training steps.

[0148] In the process that the processing module 2 processes the measurement results of the reference samples and the test samples to form training data, different noise profiles constitute a noise panorama or at least part of a noise panorama. At the same time, the overall measurement results of the two types of samples and the signals in the measurement results will respectively exhibit stable statistical characteristics; the statistical distribution pattern presented by the noise will also tend to be stable as the noise panorama is constructed.

[0149] Further, the artificial intelligence model can be selected but not limited to: artificial neural network, perceptron, support vector machine, Bayesian classifier, Bayesian network, random forest model or clustering model.

[0150] In the embodiment of the present invention, the estimation of the statistical fluctuations of a complex system in the above Equation (9) will be implemented by an artificial intelligence model.

[0151] Embodiment 3

[0152] A signal analysis method based on obtaining and identifying a noise panorama distribution model, the method comprising the following steps:

[0153] S1: Under a rich-condition measurement environment, perform repeated measurements on a variety of known samples to respectively obtain a plurality of measurement results; wherein, each measurement result includes a signal and different noise profiles;

[0154] In the technical field of signal acquisition and analysis, maintaining the consistency of external conditions during the sample measurement process is a conventional means to reduce noise fluctuations and form a good signal-to-noise ratio, and repeated sample measurement is also recognized as an effective way to reduce random errors. However, in the embodiment of the present invention, the consistency setting of external conditions is not involved, but repeated measurements of a variety of known samples are carried out under a rich-condition measurement environment. Among them, rich condition means: a measurement condition that is not aimed at maintaining the consistency of external conditions, does not involve suppressing noise, is natural, and includes real complex noise factors. The purposes of "rich-condition measurement environment" and "repeated measurement" are both to obtain rich noise profiles sufficient to construct a noise panorama.

[0155] Specifically, in a rich conditional measurement environment, limited by the properties of the samples themselves, the signals in the measurement results of each type of sample will always remain statistically invariant, but the noise will vary due to environmental changes. That is, environmental changes will increase the observational dimension of the noise. Based on the noise observational dimensions with multi-directional, multi-angle, and multi-temporal characteristics, repeated measurements of each known sample will form rich noise profiles. The rich noise profiles are the basis for constructing the noise panorama and identifying noise based on data statistical laws in subsequent steps.

[0156] S2: Process the measurement results of multiple known samples to respectively form the training data of each known sample; wherein, the training data includes a noise panorama or at least part of the noise panorama composed of multiple noise profiles.

[0157] As shown in the Figure 4 specification appendix, in a single noise observational dimension, only the noise profiles reflecting the local part of the noise can be obtained, and a comprehensive noise observational result cannot be obtained, that is, the noise cannot present a complete data statistical law that conforms to its true distribution characteristics. However, in the embodiments of the present invention, repeated measurements of multiple known samples are carried out in a rich conditional measurement environment. The rich noise profiles obtained under different noise observational dimensions will be sufficient to construct the noise panorama or at least part of the noise panorama. While constructing the noise panorama, the data statistical law of the noise will tend to its true mathematical statistical law.

[0158] In the embodiments of the present invention, the noise panorama means that the distribution model of the noise can already comprehensively reflect its theoretically true distribution model; the partial noise panorama means that the distribution model of the noise cannot completely reflect its theoretically true distribution model, but the distribution model already has the accuracy that can be used for subsequent signal analysis.

[0159] S3: Based on the training data of multiple known samples, with the observability presentation of the noise as the convergence target, perform artificial intelligence model training so that the model can identify signals and noise from the measurement results and distinguish multiple known samples.

[0160] In the embodiments of the present invention, the training data of each known sample will be randomly assigned as learning data and detection data at a preset ratio. Use the learning data to train the artificial intelligence model, and input the detection data into the trained artificial intelligence model. Calculate the signal recognition result. If the signal recognition accuracy rate is lower than the preset threshold, continue to train with the learning data. If the signal recognition accuracy rate is higher than the preset threshold, it is considered that the artificial intelligence model has completed training.

[0161] S4: Input the measurement result of the sample to be recognized into the trained artificial intelligence model, and the output result of the artificial intelligence model is the specific type of the sample to be recognized.

[0162] In the embodiments of the present invention, the measurement results of multiple known samples are processed to respectively form training data, and the trained artificial intelligence model can effectively distinguish each known sample. When the sample to be recognized is one of the multiple known samples, the artificial intelligence model can accurately recognize the specific type of the sample to be recognized.

[0163] Optionally, in step S1, before each measurement of each known sample, by introducing slight perturbations, a rich-condition measurement environment is created, thereby increasing the noise observation dimension and making the measurement results of each measurement contain different noise profiles.

[0164] Furthermore, the slight perturbations introduced before each measurement of each known sample can be selected but are not limited to spatial perturbations, temporal perturbations, physical perturbations, and environmental perturbations.

[0165] Spatial perturbations include but are not limited to: slightly displacing the measurement site, slightly rotating the measurement site; temporal perturbations include but are not limited to: increasing the measurement duration, shortening the measurement duration, and changing the time interval between multiple measurements; physical perturbations include but are not limited to: vibrating the measurement device or the sample during measurement, agitating the fluid sample; environmental perturbations include but are not limited to: changing the environmental temperature during measurement, changing the environmental humidity during measurement, changing the electromagnetic field during measurement, and changing the air pressure during measurement.

[0166] In step S2, the steps of processing the measurement results of multiple known samples to respectively form the training data of each known sample include:

[0167] S21. Normalize the measurement results of each known sample;

[0168] S22. Based on the normalization result of step S21, establish a posterior probability model framework.

[0169] The measurement results of multiple known samples, after being processed by steps S21-22, will respectively form qualified training data for subsequent training of the artificial intelligence model.

[0170] Specifically, in the embodiments of the present invention, for the measurement results of each known sample, they are regarded as the measured values obtained by measuring the measurement target composed of a complex system.

[0171] Define the measurement density function as where S is the measurement space dimension; V is the measurement environment; then in the measurement target, the number of systems is N, and N is defined by formula (1):

[0172]

[0173] Define B(V) as the measurement function, then the measured value There are:

[0174]

[0175] Among them,

[0176]

[0177] Equation (3) is the normalization condition. In the embodiments of the present invention, in order to make the measurement results of each known sample satisfy the normalization condition of Equation (3), step S21 is adopted to perform normalization processing on the measurement results of each known sample.

[0178] Since the measurement of each known sample is repeated, and the repetition process is represented in a discrete manner, Equation (2) is rewritten in the form of an ensemble:

[0179]

[0180] Define H as the ensemble density function, then there is:

[0181] <v> ＝H <n>(5)

[0182] The statistical fluctuations of a complex system are as follows:

[0183]

[0184] Among them, for repeated measurements, δS is the information entropy of the measurement, and δP is the environmental change amount of the measurement.

[0185] Taking δP as the statistical space of the noise panorama and δS as the statistical space of the signal. Therefore, according to Bayes' formula, we have:

[0186]

[0187] In Equation (7), it is defined that is Equation (8), and Equation (8) is the posterior probability condition. In the embodiments of the present invention, in order to make the measurement results of each known sample satisfy the posterior probability condition of Equation (8), Step S22 is adopted to establish a posterior probability model framework based on the normalization result obtained in Step S21.

[0188] Then, the estimation δn of the statistical fluctuations of the complex system * is as follows:

[0189] δn * = argmax δn P(H <n>|δn)P(δn) (9)

[0190] In the embodiment of the present invention, the measurement results processed through steps S21 - S22 can meet the normalization condition of formula (3) and the posterior probability condition of formula (8). The measurement results that meet the above two conditions can be used to implement the estimation of the statistical fluctuations of the complex system in formula (9). The measurement results that meet the above two conditions will be used as training data for subsequent artificial intelligence model training steps.

[0191] In the process of forming training data from the measurement results of each known sample, different noise profiles constitute a noise panorama or at least part of a noise panorama. At the same time, the overall measurement results of each known sample and the signals in the measurement results will respectively exhibit stable statistical characteristics; the statistical distribution pattern presented by the noise will also tend to be stable as the noise panorama is constructed.

[0192] In step S3, the artificial intelligence model can be selected but not limited to: artificial neural network, perceptron, support vector machine, Bayesian classifier, Bayesian network, random forest model or clustering model.

[0193] In the embodiment of the present invention, the estimation of the statistical fluctuations of the complex system in the above formula (9) will be implemented by the artificial intelligence model.

[0194] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Those skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, and all of them should be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope defined by the claims.< / n> < / n> < / v> < / n> < / n> < / v> < / n> < / n> < / v>

Claims

1. A signal analysis method based on obtaining and identifying a noise panoramic distribution model, characterized in that: It includes the following steps: S1: Under a rich-condition measurement environment, repeatedly measure the reference sample and the test sample to respectively obtain multiple measurement results; wherein, each measurement result includes a signal and different noise profiles; rich-condition refers to a measurement condition that is not aimed at maintaining external condition consistency, does not involve noise suppression, is natural, and includes real complex noise factors; wherein, before each measurement of the reference sample and the test sample, create a rich-condition measurement environment by introducing slight perturbations, thereby increasing the noise observation dimension and enabling different noise profiles to be included in the measurement results of each measurement; S2: Process the measurement results of the reference sample and the test sample to respectively form the training data of the reference sample and the test sample; wherein, the training data includes a noise panorama or at least a partial noise panorama composed of multiple noise profiles; wherein, a noise panorama means that the distribution model of the noise can already comprehensively reflect its theoretically true distribution model; a partial noise panorama means that the distribution model of the noise cannot completely reflect its theoretically true distribution model, but the distribution model already has the accuracy that can be used for subsequent signal analysis; S3: Based on the training data of the reference sample and the test sample, with the observability presentation of the noise as the convergence target, conduct artificial intelligence model training to enable the model to identify the signal and the noise from the measurement results and distinguish the reference sample and the test sample; S4: Input the measurement result of the sample to be identified into the trained artificial intelligence model, and the output result of the artificial intelligence model is the specific type of the sample to be identified.

2. The signal analysis method based on obtaining and identifying the noise panorama distribution model according to claim 1, wherein: The slight perturbations are selected from spatial perturbations, temporal perturbations, physical perturbations, and environmental perturbations; wherein, spatial perturbations include: slightly displacing the measurement site, slightly rotating the measurement site; temporal perturbations include: increasing the measurement duration, shortening the measurement duration, changing the time interval between multiple measurements; physical perturbations include: vibrating the measurement device or the sample during measurement, agitating the fluid sample; environmental perturbations include: changing the environmental temperature during measurement, changing the environmental humidity during measurement, changing the electromagnetic field during measurement, changing the air pressure during measurement.

3. The signal analysis method based on obtaining and identifying the noise panorama distribution model according to claim 1, wherein: In step S2, the steps of processing the measurement results of the reference sample and the test sample to respectively form the training data of the reference sample and the test sample include: S21. Normalize the measurement results of the reference sample and the test sample; S22. Based on the normalization result of step S21, establish a posterior probability model framework; After the measurement results of the reference sample and the test sample are processed through steps S21 - 22, the qualified training data will be respectively formed for subsequent artificial intelligence model training.

4. The signal analysis method based on obtaining and identifying the noise panorama distribution model according to claim 1, wherein: In step S3, the artificial intelligence model is: an artificial neural network, a support vector machine, a Bayesian network, a random forest model, or a clustering model.

5. The signal analysis method based on obtaining and identifying a noise panoramic distribution model according to claim 1, characterized in that: In step S3, during the training process of the artificial intelligence model, the model will, in an iterative manner, perform a large number of empirical learning, induction, and convergence on the features contained in the training data that can achieve signal and noise identification, as well as the features that can distinguish between reference samples and test samples, and learn the relationship between the features and the preset labels; Among them, the features that can achieve signal identification include the statistical distribution patterns presented after processing multiple measurement results and conforming to the true mathematical statistical laws of the signal; the features that can achieve noise identification include the statistical distribution patterns presented by the noise panorama constructed by diverse noise profiles and approaching the true mathematical statistical laws of the noise; the features that can distinguish between reference samples and test samples include the statistical distribution patterns presented respectively after processing multiple measurement results of reference samples and test samples.

6. The signal analysis method based on obtaining and identifying a noise panoramic distribution model according to claim 5, characterized in that: The preset labels include output labels and input labels; among them, the output labels include two labels representing reference samples and test samples respectively; the input labels are two sets of coupled labels of the training data related to reference samples and test samples respectively, and each coupled label is respectively associated with the rich-condition measurement environment when the sample is measured; each coupled label in different groups respectively represents: the coupling of the measurement results of reference samples or test samples and the noise panorama under each independent measurement environment in the rich-condition measurement environment; among them, the noise profile included in the measurement result is the noise profile obtained under this independent measurement environment.

7. A signal analysis system based on obtaining and identifying a noise panoramic distribution model, characterized in that: It includes a measurement module, a processing module, a training module, and an analysis module; the measurement module includes a perturbation mechanism; Under the rich-condition measurement environment, the measurement module repeatedly measures reference samples and test samples to respectively obtain multiple measurement results; among them, each measurement result contains signals and different noise profiles; Before each measurement of the reference sample and the test sample by the measurement module, the perturbation mechanism introduces slight perturbations to create a rich-condition measurement environment, thereby increasing the noise observation dimension of the sample measurement and making the measurement results of each sample measurement contain different noise profiles; The processing module processes the measurement results of reference samples and test samples to respectively form the training data of reference samples and test samples; among them, the training data includes a noise panorama or at least a partial noise panorama composed of multiple noise profiles; where the noise panorama means that the distribution model of the noise can already comprehensively reflect its theoretically true distribution model; the partial noise panorama means that the distribution model of the noise cannot completely reflect its theoretically true distribution model, but the distribution model already has the accuracy that can be used for subsequent signal analysis; Based on the training data of reference samples and test samples, the training module conducts artificial intelligence model training with the observable presentation of noise as the convergence target, enabling the model to identify signals and noise from measurement results and distinguish between reference samples and test samples; For the measurement results of the sample to be identified, the analysis module inputs them into the trained artificial intelligence model, and the output result of the artificial intelligence model is the specific type of the sample to be identified.

8. The signal analysis system based on obtaining and identifying a noise panoramic distribution model according to claim 7, wherein: Before each measurement of the reference sample and the test sample, the slight perturbation introduced by the perturbation mechanism is selected from spatial perturbation, temporal perturbation, physical perturbation, and environmental perturbation.

9. The signal analysis system based on obtaining and identifying the noise panoramic distribution model according to claim 7, wherein: The processing module includes a normalization module and a posterior probability module; Among them, the normalization module normalizes the measurement results of the reference sample and the test sample and outputs the normalization results respectively; the posterior probability module, based on the normalization results, establishes a posterior probability model framework and forms the training data of the reference sample and the test sample that meet the requirements respectively for subsequent artificial intelligence model training.

Images