Dual-wavelength differential signal adaptive compensation method for aerosol scattering interference

Abstract

The application relates to the technical field of adaptive compensation of photoelectric signals, and particularly discloses a dual-wavelength differential signal adaptive compensation method for aerosol scattering interference, which comprises the following steps: in response to a collection instruction of a multi-source sensing signal, synchronously acquiring an optical signal and an auxiliary sensing signal, and extracting an effective signal and a characteristic parameter thereof through preprocessing; and calculating an interference coupling coefficient based on the characteristic parameter, and determining the interference coupling state of a current environment according to the interference coupling coefficient. Through construction of a disturbance adaptive neural network model and a dual-mode hierarchical updating mechanism, the system can dynamically adjust a sensor fusion weight according to whether the environment is dominated by aerosol, disturbance or coupled interference, so that precise decoupling and classified compensation of complex interference sources are realized.

Description

Technical Field

[0001] This invention relates to the field of adaptive compensation technology for photoelectric signals, specifically to an adaptive compensation method for dual-wavelength differential signals to address aerosol scattering interference. Background Technology

[0002] In existing technologies, compensation for aerosol scattering interference mainly relies on dual-wavelength differential absorption radar or optical sensors, which perform baseline subtraction by assuming that the environmental interference is linear or statically distributed. However, in dynamic and complex application scenarios such as vehicle-mounted and airborne applications, high-frequency mechanical vibrations, abrupt attitude changes, and strong electromagnetic interference of the carrier often occur simultaneously with changes in aerosol concentration. This presents a significant challenge in frequency domain aliasing:

[0003] The low-frequency micro-vibrations of the carrier highly overlap with the turbulent changes of aerosols in the frequency domain. Traditional frequency domain filtering methods based on single-source optical signals cannot distinguish whether the signal fluctuations are caused by changes in external dust concentration or by internal mechanical vibrations. This makes the system prone to misinterpreting vibrations as changes in aerosol concentration, leading to the incorrect introduction of compensation amounts and resulting in spurious fluctuations in the final detection signal.

[0004] Furthermore, under extreme conditions of strong coupling interference, such as strong vibrations accompanied by high-concentration aerosol changes, there is also the problem of parameter update deadlock caused by correlation masking. Traditional adaptive algorithms usually include a safety mechanism to update model parameters only when the error signal is highly correlated with aerosol characteristics, in order to prevent model divergence.

[0005] However, under the masking effect of strong vibration noise, the true aerosol deviation characteristics are significantly weakened, causing the calculated correlation coefficient to remain consistently below the safety threshold. At this point, although the system faces a huge aerosol inversion error, it is judged as pure noise because the characteristics are not clearly visible, thus forcibly blocking the parameter update channel. This logical conflict—that the more compensation is needed, the less it can be updated—leads to the current technology's inability to maintain detection accuracy in harsh environments, and even complete failure. Summary of the Invention

[0006] This invention aims to at least partially solve one of the technical problems in related technologies. Therefore, the objective of this invention is to propose an adaptive compensation method for dual-wavelength differential signals to address aerosol scattering interference, thereby improving detection robustness and signal fidelity in dynamic environments.

[0007] To achieve the above objectives, a first aspect of the present invention proposes an adaptive compensation method for dual-wavelength differential signals to address aerosol scattering interference, comprising the following steps:

[0008] In response to the acquisition command of multi-source sensor signals, optical signals and auxiliary sensor signals are acquired simultaneously, and the effective signals and their characteristic parameters are extracted after preprocessing.

[0009] The interference coupling coefficient is calculated based on the aforementioned characteristic parameters, and the interference coupling state of the current environment is determined accordingly.

[0010] Based on the interference coupling state, compensation parameters and sensor fusion weights are dynamically output through a perturbation-adaptive neural network model.

[0011] Based on the compensation parameters, sensor fusion weights, and correction factors, comprehensive compensation is performed on the dual-wavelength differential signal to output the target signal.

[0012] The feedback optimization process of the method includes: classifying and identifying the deviation between the target signal and the reference signal, distinguishing between systematic deviation and random deviation; if it is determined to be a systematic deviation, further identifying the dominant type of the deviation, and iteratively updating the compensation parameters of the perturbation-adaptive neural network model in combination with the error threshold dynamically adjusted based on the signal-to-noise ratio.

[0013] To achieve the above objectives, a second aspect of the present invention proposes a dual-wavelength differential signal adaptive compensation system for aerosol scattering interference, the system comprising:

[0014] A multi-source signal acquisition module is used to simultaneously acquire optical signals and auxiliary sensing signals;

[0015] The interference analysis module is used to preprocess the acquired signals, extract feature parameters, and calculate the interference coupling coefficient to determine the interference coupling state.

[0016] The adaptive decision-making module is used to run the disturbance-adaptive neural network model and dynamically output compensation parameters and sensor fusion weights according to the disturbance coupling state.

[0017] The comprehensive compensation module is used to calculate the dual-wavelength differential signal by combining the compensation parameters and correction factors, and output the target signal;

[0018] The feedback optimization module is used to identify the type of deviation in the output signal and iteratively update the model parameters in the adaptive decision module when it is determined to be a system deviation.

[0019] To achieve the above objectives, a third aspect of the present invention provides an electronic device including a memory, a processor, and a computer program stored in the memory. When the computer program is executed by the processor, it implements the above-described adaptive compensation method for dual-wavelength differential signals against aerosol scattering interference.

[0020] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0021] The dual-wavelength differential signal adaptive compensation method for aerosol scattering interference in this invention effectively solves the problems of frequency domain aliasing and state misjudgment faced by single optical detection methods. Specifically:

[0022] By constructing a perturbation-adaptive neural network model and a dual-mode hierarchical update mechanism, the system can dynamically adjust the sensor fusion weights according to whether the environment is dominated by aerosols, perturbations, or coupling interference, thus achieving accurate decoupling and classification compensation for complex interference sources.

[0023] Especially in scenarios with strong coupling interference, the orthogonal subspace projection technology was used to successfully separate the weak aerosol deviations hidden under strong vibration noise, breaking the parameter update deadlock of traditional algorithms. This ensured that the system could maintain high-precision adaptive compensation capability under extreme conditions, and significantly improved the detection robustness and signal restoration accuracy in dynamic environments. Attached Figure Description

[0024] The disclosure of this invention is illustrated with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. In the drawings, the same reference numerals are used to refer to the same parts. Wherein:

[0025] Figure 1 This is a flowchart illustrating the adaptive compensation method for dual-wavelength differential signals against aerosol scattering interference provided by the present invention.

[0026] Figure 2 This is a three-dimensional mapping surface plot of the characteristic confidence C, signal-to-noise ratio, and statistical bias in the dual-wavelength differential signal adaptive compensation method for aerosol scattering interference provided by the present invention.

[0027] Figure 3 This is a schematic diagram of the time-varying curve and state partitioning of the interference coupling coefficient K in the dual-wavelength differential signal adaptive compensation method for aerosol scattering interference provided by the present invention.

[0028] Figure 4 This is a histogram of the dynamic allocation of neural network attention weights under different interference states in the dual-wavelength differential signal adaptive compensation method for aerosol scattering interference provided by the present invention.

[0029] Figure 5 This is a comparison diagram of the effects of dual-wavelength differential signals before and after comprehensive compensation in the adaptive compensation method for dual-wavelength differential signals against aerosol scattering interference provided by the present invention.

[0030] Figure 6 This is a comparison diagram of the deviation signal vectors before and after orthogonal subspace projection in the dual-wavelength differential signal adaptive compensation method for aerosol scattering interference provided by the present invention.

[0031] Figure 7 This is a schematic diagram illustrating the implementation of the dual-wavelength differential signal adaptive compensation system for aerosol scattering interference provided by the present invention.

[0032] Figure 8 This is a schematic diagram of the electronic device provided by the present invention. Detailed Implementation

[0033] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0034] The following describes, with reference to the accompanying drawings, an adaptive compensation method, system, and electronic device for dual-wavelength differential signals against aerosol scattering interference according to embodiments of the present invention.

[0035] Example 1:

[0036] This embodiment provides an adaptive compensation method for dual-wavelength differential signals to address aerosol scattering interference. This method is configured for use in a processing system with a high-performance computing unit capable of high-speed data interaction with front-end multi-source sensor hardware. The core logic of this method lies in using multi-dimensional sensor data fusion to solve the problem that single optical detection methods struggle to distinguish between aerosol concentration changes and carrier mechanical vibration or attitude disturbances in complex dynamic environments.

[0037] Specifically, such as Figure 1 As shown, the method in this embodiment includes the following steps:

[0038] S1. Synchronous acquisition and preprocessing of multi-source sensor signals.

[0039] Specifically, the execution flow of the method begins with responding to the acquisition command of multi-source sensor signals. Upon receiving the start command, the processing system sends synchronization trigger signals to all connected sensors to ensure that all data are strictly aligned on the time axis. This step forms the basis for subsequent heterogeneous data fusion, during which the processing system simultaneously acquires optical signals and auxiliary sensor signals.

[0040] For example, the hardware sources of the multi-source sensing signals include, but are not limited to, a dual-frequency, dual-wavelength laser system and auxiliary sensors. The dual-frequency, dual-wavelength laser system is used to acquire the core optical signals. This system can simultaneously emit and receive two laser pulses of different wavelengths, 532 nm and 1064 nm, to obtain information on the backscattering intensity of particulate matter in the atmosphere. Simultaneously, the auxiliary sensors are responsible for acquiring environmental and carrier state information. These auxiliary sensing signals specifically include distance signals acquired by millimeter-wave radar or similar ranging sensors, vibration and attitude signals acquired by a triaxial accelerometer and gyroscope within a high-precision inertial measurement unit, and electromagnetic interference signals acquired by a three-dimensional electric field sensor. These auxiliary signals provide a physical reference for subsequent elimination of signal fluctuations caused by non-aerosol factors.

[0041] It is important to note that the raw acquired signals often contain background noise, electronic thermal noise, and baseline drift. Therefore, preprocessing steps are necessary to extract the effective signals and their characteristic parameters. Preprocessing includes denoising and smoothing the raw optical signals, performing coordinate system transformation on the inertial sensor data, and filtering the electromagnetic signals. After basic data cleaning, the system extracts key characteristic parameters from the massive dataset. These parameters include, but are not limited to, the intensity ratio of the dual-wavelength signals, the rate of change of the signal, the vibration frequency, the vibration amplitude, the rate of change of the attitude angle, and the magnitude of the ambient electric field intensity.

[0042] Specifically, to ensure that the feature parameters input into the subsequent algorithm model have sufficient confidence, this embodiment introduces a feature confidence calculation mechanism during the feature parameter extraction process. Feature Confidence This is a quantitative indicator used to measure whether the currently extracted feature parameters are contaminated by instantaneous strong noise. The feature confidence score has a non-linear mapping relationship with the signal-to-noise ratio of the effective signal and the normalized statistical deviation of the feature parameters. This embodiment uses the following formula to calculate the feature confidence score. :

[0043] ;

[0044] In the formula, This represents the calculated feature confidence level, and its value is normalized to be between 0 and 1. The signal-to-noise ratio (SNR) represents the effective signal, and its unit is decibels (dB). The standard deviation of the characteristic parameter within the current sliding time window is used to characterize the dispersion of the data; The mean of the representative feature parameters within the current sliding time window is used to characterize the central tendency of the data; the constant 10 in the formula serves as a scaling factor for the signal-to-noise ratio (SNR), used to adjust the weight of the SNR on the confidence level; the function... is the base of the natural logarithm.

[0045] This formula shows that the higher the signal-to-noise ratio and the smaller the dispersion of the feature parameters relative to the mean, the higher the calculated confidence level. The closer the confidence level is to 1, the more reliable the characteristic parameter is; conversely, if the signal-to-noise ratio is low or the data fluctuates drastically, the confidence level will be low. It will drop rapidly.

[0046] like Figure 2 This diagram illustrates the nonlinear response characteristics and multidimensional parameter coupling relationship of the feature confidence calculation model in this embodiment. The horizontal axis represents the signal-to-noise ratio of the effective signal, the vertical axis represents the normalized statistical deviation of the feature parameters, and the vertical axis represents the calculated feature confidence value.

[0047] Figure 2 The three-dimensional smooth surface presented clearly distinguishes the data quality levels through color gradients. The flat warm-colored area in the upper right corner of the surface corresponds to a high signal-to-noise ratio and low statistical dispersion, indicating that the feature confidence level is close to 1 and the system will fully accept the feature parameter.

[0048] Conversely, as the signal-to-noise ratio decreases or the statistical bias increases, the surface exhibits a steep downward trend and transitions to the dark, cool-colored area in the lower left, intuitively reflecting the rapid decay of confidence under strong noise backgrounds or environments with drastic data fluctuations.

[0049] This unique nonlinear mapping mechanism ensures that when the signal-to-noise ratio falls below a certain level or the dispersion is too large, the calculation results can quickly drop below the preset confidence threshold. When the values ​​fall into the steep descent section of the curve or the low-color valley area in the graph, the algorithm will automatically trigger the weight reduction or removal operation of low-quality feature parameters in the preprocessing stage, thus strongly demonstrating that the present invention can dynamically adjust the inversion weights according to the signal quality and has excellent adaptive shielding capability and data cleaning accuracy for various interference noises in complex environments.

[0050] Optionally, when the calculated feature confidence level is lower than a preset confidence threshold, the processing system will automatically reduce the weight of the corresponding feature parameter in subsequent inversion processes. For example, if the optical signal-to-noise ratio drops sharply due to strong light interference at a certain moment, and the calculated confidence level is lower than the preset threshold of 0.6, then the weight of this optical feature will be forcibly reduced when calculating the interference coupling coefficient or inputting it into the neural network, in order to prevent low-quality data from misleading the system's judgment logic.

[0051] S2. Determination and analysis of interference coupling state.

[0052] Specifically, after obtaining the high-confidence feature parameters, the system enters the interference analysis stage. The core task of this stage is to calculate the interference coupling coefficient based on the feature parameters and determine the interference coupling state of the current environment accordingly. This is the key decision-making step that enables this method to achieve adaptive compensation.

[0053] For example, the interference coupling coefficient is defined as the ratio of the amplitude of the signal fluctuation dominated by the disturbance to the amplitude of the signal fluctuation dominated by the aerosol. This definition intuitively reflects the main energy source of the current signal fluctuation. If the carrier is in a state of violent vibration while the aerosol concentration is relatively stable, the disturbance amplitude in the numerator is much greater than the aerosol amplitude in the denominator, resulting in a larger coupling coefficient. Conversely, if the carrier is stationary while the dust concentration changes rapidly, the denominator is large and the coupling coefficient is small.

[0054] It should also be noted that, in order to achieve accurate classification of environmental states, this embodiment has formulated detailed rules for determining the interference coupling state of the current environment:

[0055] First, the system sets two key thresholds: a first coupling threshold and a second coupling threshold. Numerically, the first coupling threshold must be less than the second coupling threshold. In this embodiment, after extensive experimental calibration, the first coupling threshold is set to 0.5, and the second coupling threshold is set to 1.5.

[0056] Specifically, the decision logic is executed as follows:

[0057] When the calculated interference coupling coefficient is less than the first coupling threshold, i.e., less than 0.5, the system determines that it is currently in an aerosol-dominated interference state. This state usually corresponds to a situation where the carrier is running smoothly, but there is dense smoke, fog, or dust clusters in the external environment. At this time, the signal fluctuation mainly reflects the changes of the measured target, and the inversion of aerosol parameters should be the focus.

[0058] When the calculated interference coupling coefficient is between the first coupling threshold and the second coupling threshold, i.e., within the range of 0.5 to 1.5, the system determines that it is currently in a coupling interference state. This is the most complex operating condition, which usually occurs when the carrier is shaking to a certain extent, and the surrounding aerosol concentration is also changing significantly. The contributions of the two to the signal are of similar magnitude and difficult to separate simply, requiring a more refined weight allocation strategy.

[0059] When the calculated interference coupling coefficient is greater than the second coupling threshold, i.e., greater than 1.5, the system determines that it is currently in a disturbance-dominated interference state. This state is common when the carrier passes through a bumpy road, encounters strong airflow, or experiences mechanical resonance, while the surrounding air is relatively clean or uniform. In this case, the violent fluctuations in the signal are almost entirely caused by non-target factors, and strong vibration suppression compensation is necessary.

[0060] like Figure 3 The figure illustrates the dynamic process of the continuous change of the interference coupling coefficient over time in the system and its corresponding three-level state determination logic. The horizontal axis represents the time process of the system operation, and the vertical axis represents the interference coupling coefficient value calculated in real time.

[0061] Figure 3 The system clearly delineates the decision range of the interference state using three different background colors. The light green area at the bottom represents the aerosol-dominated interference area, which corresponds to the case where the coefficient value is less than the first coupling threshold of 0.5. At this time, the system determines that the signal fluctuation mainly originates from the change in smoke and dust concentration.

[0062] The light yellow area in the middle represents the coupling interference zone, which corresponds to the case where the coefficient value is between the first coupling threshold of 0.5 and the second coupling threshold of 1.5. At this time, the system determines that there is a mixed interference of aerosol and mechanical vibration.

[0063] The light red area at the top represents the disturbance-dominated interference zone, corresponding to a coefficient value greater than the second coupling threshold of 1.5. In this case, the system determines that the signal fluctuation is mainly caused by severe mechanical vibration.

[0064] Figure 3 The solid blue line waveform visually depicts the fluctuation trajectory of the interference coupling coefficient on the time axis. When the curve crosses different colored background regions as the external environment changes, the system can lock the environmental state in real time according to the region where the curve is currently located, and use this as the decision basis for the subsequent neural network model to dynamically adjust the attention weight and start different hierarchical update mechanisms. This proves that the present invention can accurately classify and adaptively decouple complex and ever-changing dynamic interference sources in real time.

[0065] S3. Operation and decision-making of perturbation-adaptive neural network models.

[0066] Specifically, based on the interference coupling state determined in the above steps, the processing system will invoke and run a specially designed disturbance-adaptive neural network model. This model dynamically outputs compensation parameters and sensor fusion weights according to the current environmental state. Unlike traditional static lookup table methods, this model possesses powerful nonlinear fitting capabilities and can adapt to changing battlefield or industrial environments.

[0067] For example, the perturbation-adaptive neural network model employs a hybrid network architecture, whose internal structure includes an input layer, hidden layers, and an output layer. The input layer receives a preprocessed eight-dimensional feature vector, including aerosol concentration, particle size distribution, pollutant dielectric properties, electric field strength, vibration amplitude, vibration frequency, and attitude angle change rate. The output layer is responsible for outputting the laser signal weights ultimately used for signal correction. Laser signal compensation coefficient And the fusion weights of auxiliary sensors such as LiDAR.

[0068] It is also worth noting that, to reflect the model's adaptability to different disturbance states, the hidden layer specifically includes a disturbance and aerosol-coordinated attention sublayer. This sublayer introduces an attention mechanism, whose weight allocation rules are dynamically adjusted in response to the disturbance coupling state determined in the preceding steps, thereby changing the model's attention level to different input features.

[0069] Specifically, the weighting rules are as follows:

[0070] When the disturbance is determined to be dominated by aerosols, the model will automatically adjust the attention mask so that the attention weight of aerosol features is significantly higher than that of disturbance features. For example, the attention weight of aerosol concentration and particle size distribution features is set to 0.7, while the weight of vibration and attitude features is reduced to 0.3, guiding the model to make more use of optical properties to calculate compensation parameters;

[0071] When the disturbance is determined to be dominant, the model performs the opposite operation, making the attention weight of the disturbance features significantly higher than that of the aerosol features. In this case, changes in vibration frequency and amplitude will dominate the model's output, ensuring that the model can respond quickly to mechanical disturbances and output large-scale suppression parameters.

[0072] When the system is identified as a coupled interference state, the model assigns balanced attention weights to aerosol features and perturbation features, for example, both of which are 0.5, forcing the network to take into account the nonlinear relationship between the two types of features and find the best balance point.

[0073] like Figure 4 The dynamic weight allocation mechanism of the collaborative attention sublayer inside the perturbation-adaptive neural network model in this embodiment is shown. The horizontal axis of the figure lists the three typical environmental states identified by the system: aerosol-dominated interference, coupling interference, and perturbation-dominated interference. The vertical axis represents the normalized attention weight values.

[0074] In the legend, the blue bars represent aerosol feature weights, and the red bars represent perturbation feature weights. The changes in their heights intuitively reflect the model's attention to different input features.

[0075] from Figure 4 As can be seen on the left side, when the environment is under aerosol-dominated interference, the blue bar is significantly higher than the red bar, which corresponds to the description of setting the aerosol feature weight to 0.7 and the disturbance feature weight to 0.3. This indicates that the algorithm focuses on using optical properties for inversion at this time.

[0076] The right side of the figure shows that under the disturbance-dominated interference state, the red bar surpasses the blue bar, the weight distribution is reversed, corresponding to the setting of the disturbance feature weight being increased to 0.7, which establishes the dominant position of vibration suppression.

[0077] In the intermediate coupled interference state, the heights of the red and blue bars are equal, both at the 0.5 level, verifying the model's balanced strategy that takes into account the nonlinear relationship between the two types of features under complex conditions. This visualization result of adaptively adjusting weights according to environmental conditions proves that the neural network model described in this invention is not a static black box, but possesses the intelligent characteristic of dynamically reconstructing internal attention in real time according to external physical conditions, thereby ensuring the system's high-precision decision-making capability under changing interference environments.

[0078] S4. Dynamic update mechanism for compensation parameters.

[0079] Specifically, the process of dynamically outputting compensation parameters by the neural network model is not static, but involves a rigorous dual-mode hierarchical update mechanism. This mechanism aims to resolve the contradiction between the real-time nature and stability of parameter updates, and is triggered based on the real-time monitored aerosol change rate.

[0080] For example, to define the rate of change, the system sets two rate thresholds: a first rate threshold and a second rate threshold, where the first rate threshold is less than the second rate threshold. In this embodiment, the first rate threshold is set to a concentration change rate of 5% per second, and the second rate threshold is set to a concentration change rate of 20% per second.

[0081] Optionally, when the monitored aerosol change rate exceeds the second rate threshold, it indicates a drastic change in the environment, such as a sudden dense smoke event. At this point, the system activates a first update strategy, also known as the fast tracking mode. In this mode, the system updates the parameters with a first update step size or a first update cycle. To keep up with changes, the first update cycle is set to be extremely short, for example, 5 milliseconds, and allows for larger parameter adjustment steps to ensure that the compensation parameters can quickly converge to the new operating point.

[0082] Optionally, when the monitored aerosol change rate is between the first rate threshold and the second rate threshold, it indicates that the environmental change is at a moderate level. At this time, the system activates a second update strategy, also known as the normal mode. In this mode, the system uses an adaptive learning rate to update the parameters. The learning rate automatically scales according to the magnitude of the current error gradient, ensuring both fast convergence and avoiding excessive oscillations near extreme points.

[0083] Optionally, when the monitored aerosol change rate is less than the first rate threshold, it indicates that the environment is relatively static or changing extremely slowly. At this point, the system initiates a third update strategy, also known as the stable convergence mode. In this mode, to prevent overfitting or parameter drift, the system introduces a regularization term and updates the parameters with a second update cycle. The second update cycle is significantly longer than the first update cycle, for example, set to 20 milliseconds. By reducing the update frequency and introducing L2 regularization constraints, the compensation parameters remain smooth and stable, reducing the consumption of system computational resources.

[0084] S5. Calculation of correction factors and execution of comprehensive compensation.

[0085] Specifically, after obtaining the basic compensation parameters, this method also needs to calculate a series of targeted correction factors to further eliminate specific errors introduced by the carrier motion. These correction factors specifically include vibration correction factors and attitude correction factors. These two factors directly affect the signal amplitude and represent physical-level corrections to the neural network output parameters.

[0086] For example, the vibration correction factor is calculated piecewise based on the vibration frequency and amplitude acquired in real time. This is because the mechanisms by which low-frequency large-amplitude vibrations and high-frequency micro-vibrations affect optical devices are different; the former mainly causes optical path shift, while the latter mainly causes speckle noise.

[0087] Specifically, when the detected vibration frequency is greater than a preset frequency threshold, such as 10Hz, it falls within the high-frequency interference zone. In this zone, the vibration correction factor exhibits a negative linear correlation with the vibration amplitude. The calculation formula is as follows: as the vibration amplitude increases, the correction factor decreases linearly to reduce signal gain and counteract the spurious signal enhancement caused by high-frequency jitter. Its mathematical expression can be:

[0088] ;

[0089] In the formula, This is a vibration correction factor; This represents the current vibration amplitude. This is the reference value for the maximum amplitude at high frequencies; For linear coefficients, such as 0.1.

[0090] Specifically, when the detected vibration frequency is less than or equal to a preset frequency threshold, such as 10Hz, it falls within the low-frequency coupling region, where the impact of the vibration is more complex. The vibration correction factor is negatively correlated with the product of the vibration amplitude and frequency. This means that even if the amplitude is small, if the frequency is close to a certain resonance point (reflected in the product effect of the frequency term), the correction strength will increase. Its mathematical expression can be:

[0091] ;

[0092] In the formula, The current vibration frequency; and These are the reference amplitude and reference frequency at low frequencies, respectively. This is a weighting factor, such as 0.15.

[0093] It is also important to note the attitude correction factor. It is based on the angular velocity measured by the gyroscope and is used to compensate for the deviation of the laser irradiation angle caused by the tilt of the carrier. Its calculation logic also follows the same principle: the larger the angular velocity, the smaller the correction factor, and the stronger the attenuation compensation of the original signal.

[0094] Specifically, after all parameters are prepared, the system enters the final signal generation stage. Based on the compensation parameters, sensor fusion weights, and correction factors, comprehensive compensation is performed on the dual-wavelength differential signal to output the target signal. The calculation process of the comprehensive compensation is as follows: the dual-wavelength signal is weighted and differentially processed using the compensation parameters, and the target signal is generated by combining the correction factors, pollutant correction factors, and environmental correction terms.

[0095] In this embodiment, the target signal The complete calculation formula is as follows:

[0096] ;

[0097] In the formula, This represents the final compensated target signal output. This represents the original laser signal with a wavelength of 532 nanometers that was collected; This represents the original laser signal acquired at a wavelength of 1064 nanometers; and These represent the laser signal weights and laser signal compensation coefficients output by the neural network model; these two parameters already include adaptive adjustments for aerosol scattering characteristics. This represents a contaminant correction factor, which is based on the dielectric properties of the contaminants. and pollutant thickness This is determined to eliminate the influence of substances adhering to the lens surface on light intensity; This represents the vibration correction factor obtained from the aforementioned calculation; Represents the attitude correction factor; the terms in parentheses Represents environmental correction items, among which Represents the relative humidity of the environment. This represents the laser wavelength.

[0098] The environmental correction term clearly shows a positive correlation with ambient humidity and an exponential decay with laser wavelength. This means that in high humidity environments, the signal will be moderately amplified and compensated due to the hygroscopic growth effect of water vapor aerosols. This compensation effect will decay exponentially for long-wavelength lasers, such as 1064nm, which is consistent with the physical law of the ratio of wavelength to particle size in Mie scattering theory.

[0099] like Figure 5 The figure demonstrates the final restoration effect of the integrated compensation module on the contaminated optical signal in this embodiment. The horizontal axis represents the sampling time process, and the vertical axis represents the normalized signal strength value.

[0100] Figure 5 The thin solid line waveform, which is grayish-black, represents the unprocessed raw dual-wavelength differential signal. The curve exhibits violent high-frequency oscillations accompanied by obvious low-frequency baseline drift. This corresponds to the optical path jitter interference caused by the mechanical vibration of the carrier and the angle deviation caused by the tilt of the attitude. This large fluctuation completely masks the true trend of aerosol concentration changes.

[0101] Figure 5 The green dashed line represents a real aerosol reference signal under ideal conditions, showing a smooth and clear contour of concentration changes.

[0102] Figure 5 The thick solid red line represents the output target signal processed by the method described in this invention. By comparison, it can be clearly seen that the red curve eliminates the high-frequency noise spikes in the black curve and corrects the low-frequency drift error. Its trajectory highly coincides with the green real reference signal. This proves that by combining the dynamic compensation parameters output by the perturbation-adaptive neural network and the calculated vibration correction factor and attitude correction factor, the system has successfully achieved accurate suppression and decoupling of multi-source interference in complex dynamic environments, ensuring that the final output signal can reflect the physical nature of the detected target with high fidelity.

[0103] S6, Disturbance-aware feedback optimization.

[0104] Specifically, outputting the target signal does not signify the end of the process. To ensure the long-term robustness of the system, this method also includes a closed-loop feedback optimization process. The core of this process lies in classifying and identifying the deviation between the target signal and the reference signal, distinguishing between systematic deviations and random deviations.

[0105] For example, the reference signal can be derived from a smoothed prediction value from the previous time step, or a standard value obtained under calibration conditions. The system first calculates the difference sequence between the current output signal and the reference signal, and then statistically analyzes the sequence over several consecutive frames. If the mean of the difference sequence fluctuates randomly around zero and passes the white noise test, it is determined to be a random bias. This bias is usually caused by electronic thermal noise or quantum shot noise, which is a physical limit that cannot be eliminated. The system will ignore this type of bias and maintain the current parameters unchanged.

[0106] Optionally, if the difference sequence exhibits a clear trend or a non-zero mean, it is identified as a systematic bias, indicating a mismatch between the current compensation model and the actual environment. In this case, the system will further identify the dominant type of the bias, i.e., determine whether the bias is caused by inaccurate estimation by the aerosol model or by strong noise that has not been filtered out.

[0107] Specifically, the process of identifying the dominant type of this bias includes calculating the correlation coefficient between the system bias and the aerosol inversion characteristics. The aerosol inversion characteristics mentioned here can be theoretical echo curves derived from Mie theory. The system sets an effective correlation threshold, which is set to 0.7 in this embodiment.

[0108] If the calculated correlation coefficient is greater than the effective correlation threshold, i.e., greater than 0.7, it indicates that the current deviation waveform highly matches the theoretical characteristics of aerosols, and is determined to be an aerosol-dominated deviation. This usually means that the parameters output by the neural network... or Accuracy errors exist. At this point, the system triggers adjustments to aerosol-related compensation parameters, iteratively updating the compensation parameters of the perturbation-adaptive neural network model by combining an error threshold dynamically adjusted based on the signal-to-noise ratio. The so-called dynamic error threshold means that when the signal-to-noise ratio is high, the allowable error threshold is smaller, and parameter adjustments are more sensitive; when the signal-to-noise ratio is low, the error threshold is relaxed to prevent noise from causing parameter over-adjustment.

[0109] If the calculated correlation coefficient is less than or equal to the effective correlation threshold, i.e., less than or equal to 0.7, it indicates that the current deviation waveform is chaotic or exhibits obvious mechanical vibration characteristics, unrelated to aerosol properties, and is determined to be a noise-dominated deviation. In this case, the system believes that the model parameters themselves are not the problem; the problem lies in the poor quality of the input signal. Therefore, the system does not trigger parameter adjustment, or only triggers noise suppression processing, such as increasing the filtering intensity, to avoid erroneous parameter updates that could cause model divergence. This mechanism constitutes a disturbance-aware feedback, ensuring that the system can maintain its composure in the face of extremely harsh, high-noise environments, and will not blindly change parameters due to unclear target visibility, thereby greatly improving the system's anti-interference capability and survivability.

[0110] In summary, this embodiment constructs a full-link, closed-loop adaptive compensation system, from heterogeneous fusion at the signal acquisition end, to confidence weighting at the feature processing end, to attention mechanisms and hierarchical updates at the model decision-making end, and finally to physical correction at the execution end and deviation classification at the feedback end. Through the detailed steps described above, this system can effectively extract high-fidelity aerosol detection signals in complex dynamic environments.

[0111] Example 2:

[0112] This embodiment, based on Embodiment 1, further refines and expands the design for scenarios where aerosol turbulence and low-frequency carrier vibration severely overlap in the frequency domain. In practical applications, such as the low-frequency chassis resonance generated by heavy engineering vehicles traversing bumpy roads, or the low-frequency disturbances generated by the rotors of large drones hovering, the frequency range is often concentrated between a few hertz and twenty hertz. Unfortunately, this frequency range highly coincides with the frequency of concentration changes caused by the rapid diffusion of dust clusters or turbulence in nature. If the conventional frequency domain filtering method in Embodiment 1 is used, it is very easy to mistake the mechanical vibration of the carrier for a sudden change in aerosol concentration, thus incorrectly calculating the interference coupling coefficient and leading to directional errors in the compensation strategy. To solve the problem of misjudgment of interference coupling state and critical oscillation caused by this frequency domain aliasing, this embodiment introduces an enhanced processing flow based on heterogeneous frequency domain coherence analysis and asymmetric hysteresis logic, including the following:

[0113] Step 1: Frequency domain coherence analysis based on heterogeneous signals.

[0114] Specifically, the method in this embodiment first requires performing frequency domain coherence analysis based on heterogeneous signals. Heterogeneous signals here refer to signals with completely different physical generation mechanisms; one is a dual-wavelength differential optical signal based on the photoelectric effect, and the other is a mechanical vibration accelerometer signal based on the piezoelectric or capacitive effect. The system utilizes the high-speed parallel computing capabilities of the processing system to simultaneously capture the data streams of the dual-wavelength differential signal and the triaxial accelerometer signal within a preset sliding time window. To prevent spectral leakage, the system applies a Hanning or Hamming window to the captured data segments for weighting, and then converts the time-domain signal to the frequency domain using a Fast Fourier Transform.

[0115] For example, the core of the analysis lies in calculating the amplitude-squared coherence spectrum between two signals. This metric is used to quantify whether a causal linear relationship exists between the two signals at each specific frequency component. In this embodiment, the amplitude-squared coherence spectrum... The calculation strictly follows the following mathematical definition:

[0116] ;

[0117] In the formula, The representative frequency is The amplitude squared coherence function value at time, its range is between zero and one; Represents a two-wavelength differential signal With accelerometer signal The cross-power spectral density between two signals reflects their frequency response. The degree of interrelation between the parts; symbols This represents the modulo operation for complex numbers; The power spectral density represents the self-power of the dual-wavelength differential signal and is used to characterize the energy distribution of the optical signal at different frequencies. The power spectral density represents the accelerometer signal itself and is used to characterize the energy distribution of mechanical vibration at different frequencies.

[0118] It should also be noted that, as can be seen from the above formula, when When the frequency approaches one, it indicates that at that frequency point, there is a very strong linear correlation between the fluctuation of the optical signal and the vibration sensed by the accelerometer, meaning that the fluctuation of the optical signal is most likely caused by physical optical path jitter due to mechanical vibration. Conversely, when... When the frequency approaches zero, it indicates that the two are uncorrelated at that frequency point, and the fluctuation of the optical signal should be attributed to the change in aerosol concentration outside the optical path. Based on this physical fact, the system further calculates the frequency domain overlap factor. The aforementioned Defined within the effective signal bandwidth The percentage of energy from frequency components with a coherence degree greater than a preset coherence threshold is a macroscopic statistic used to determine whether severe frequency domain aliasing exists. For example, if the preset coherence threshold is 0.8, the system will count all frequency points with a coherence degree greater than 0.8, calculate the proportion of the sum of the energy at these frequency points to the total signal energy, and use this as the frequency domain overlap factor. .

[0119] Step 2: Interference source separation and coefficient correction based on coherence.

[0120] Specifically, when the system calculates the frequency domain overlap factor Then, it will be compared with a preset overlap threshold. When the frequency domain overlap factor... When the overlap exceeds a preset threshold, such as 30%, the system determines that there is a severe frequency domain aliasing phenomenon. At this point, directly calculating the interference coupling coefficient using the original waveform amplitude is no longer accurate. Therefore, the system enters the source separation correction process, using the amplitude squared coherence spectrum to decompose the dual-wavelength differential signal into coherent and incoherent components.

[0121] For example, the decomposition process is not a simple filtering process, but rather based on the reconstruction of the energy spectrum. For each frequency point... The system will output its corresponding optical signal energy. The data is split according to coherence. Multiply The portion is considered as the coherent component of energy caused by vibration, which is physically a disturbance; while Multiply The portion of energy is considered as an incoherent component of energy unrelated to vibration, which is physically a product of aerosol changes.

[0122] Optionally, after completing the full-band energy decomposition, the system performs a reclassification operation. The system classifies the energy of the coherent components as the disturbance-dominated signal fluctuation amplitude, using it as the numerator for calculating the interference coupling coefficient; simultaneously, it classifies the energy of the incoherent components as the aerosol-dominated signal fluctuation amplitude, using it as the denominator for calculating the interference coupling coefficient. Based on this amplitude value after physical source separation, the system recalculates the corrected interference coupling coefficient. At this time No longer limited by frequency, it directly points to the physical source of the signal, thus accurately reflecting which fluctuations are vibration interference and which are real dust even in the low-frequency band.

[0123] Step 3: State locking and smoothing of asymmetric hysteresis logic.

[0124] Specifically, after obtaining the corrected interference coupling coefficient This embodiment further addresses the critical oscillation problem in state determination. In Embodiment 1, the state switching threshold is singular, such as being set to 1.5, which leads to... When the voltage fluctuates slightly around 1.5, the system repeatedly jumps between coupled interference and disturbance-dominated interference at high frequency. This ping-pong effect causes severe jitter in the neural network weights, leading to step-like noise in the output signal. To completely eliminate this potential problem, this embodiment uses asymmetric hysteresis logic to determine the interference coupling state.

[0125] For example, the core of this logic lies in introducing two separate thresholds: setting a state escalation threshold. and state degradation threshold To construct an effective hysteresis loop, these two thresholds must satisfy a strict size relationship: the state escalation threshold must be greater than the second coupling threshold in Embodiment 1 (set to 1.5), while the state degrade threshold must be less than the second coupling threshold in Embodiment 1 (set to 1.5). In this embodiment, the state escalation threshold is... Set the threshold to 1.6 to reduce the state degradation threshold. It is set to 1.4, thus creating a buffer band or dead zone between 1.4 and 1.6.

[0126] It's also important to note that the triggering conditions for state transitions have become more stringent and directional. The specific judgment logic is as follows:

[0127] First, regarding the state upgrade process, only when the corrected interference coupling coefficient... Greater than the state escalation threshold At that time, that is Only when the high threshold of 1.6 is breached does the system determine that the state has switched from coupled interference to disturbance-dominated interference. This means that if the interference only increases slightly to 1.55, the system will consider it to be merely instantaneous measurement noise, insufficient to fundamentally alter the assessment of the current environmental properties. Therefore, it maintains the original state, avoiding the loss of effective signals caused by abruptly entering a strong vibration suppression mode.

[0128] Secondly, regarding the state degradation process, only when the corrected interference coupling coefficient... Less than the state degradation threshold At that time, that is The system only determines its state as switching from disturbance-dominated to coupled disturbance when it falls below the low-level line of 1.4. This means that once the system enters the disturbance-dominated mode, it tends to maintain this mode until the disturbance intensity decreases significantly and it is certain that the environment has stabilized before exiting. This asymmetric design, which makes it easy to enter but difficult to exit, or difficult to enter but easy to exit, greatly enhances the inertia and stability of the system's state determination.

[0129] Specifically, if the corrected interference coupling coefficient It falls within the middle buffer area, which satisfies... Less than or equal to and Less than or equal to At this time, the system executes a special steady-state maintenance strategy. In this case, the system forcibly maintains the interference coupling state from the previous moment. For example, if the previous moment was a coupled interference, then even if... Even if it rises to 1.58, it is still considered coupling interference; if the previous moment was dominated by disturbance interference, even... It dropped to 1.42, but it is still considered to be disturbance-dominated interference.

[0130] Optionally, in this buffered state, to further prevent output jitter caused by minor parameter changes, the system freezes the dynamic updates of the sensor fusion weights. That is, the LiDAR weights output by the neural network at this time... The values ​​from the previous frame will remain fixed and will not change with minor fluctuations in the input features. Meanwhile, the weights for the laser signal... and compensation coefficient For these core compensation parameters, the system only performs smoothing filtering, such as using moving average filtering or first-order hysteresis filtering, to filter out parameter spikes caused by computational noise under critical conditions.

[0131] As can be seen from the detailed description of this embodiment, this method successfully achieves accurate identification of low-frequency vibrations and aerosol turbulence in deep water regions with aliasing in the frequency domain by introducing amplitude squared coherence spectrum analysis; and successfully achieves steady-state locking of system control in the critical region of state determination by introducing asymmetric hysteresis logic. The combination of these two key technologies enables the compensation system to not only work in static or simple dynamic environments, but also adapt to extreme conditions such as battlefield reconnaissance and post-disaster relief, which involve severe mechanical vibrations and complex dust environments, ensuring the continuity, smoothness, and authenticity of the detection data.

[0132] Example 3:

[0133] Based on Embodiment 1 and Embodiment 2, this embodiment further explores the technology and designs a solution for an extreme but unavoidable working condition in practical applications. This working condition is called the correlation masking state under strong coupling interference.

[0134] In Example 1, the system employs a correlation coefficient-based gating mechanism, allowing model parameter updates only when the system bias is highly correlated with the theoretical inversion characteristics of the aerosol. This mechanism effectively prevents parameter divergence caused by noise in most cases. However, when the carrier is in a severely vibrating environment, such as a helicopter experiencing strong airflow during hovering observation missions, or an off-road vehicle traveling at high speed on uneven roads, the signals collected by the sensors will be mixed with extremely high-energy mechanical vibration noise.

[0135] Specifically, under this state of strong coupling interference, although the total error signal generated by the detection system does contain effective deviation components due to inaccurate estimation of aerosol model parameters, the energy of mechanical noise is far higher than that of aerosol deviation. This causes the waveform characteristics of the total error signal to macroscopically manifest primarily as random fluctuations or periodic oscillations of mechanical vibration. This results in a serious consequence: the correlation coefficient between the calculated total error signal and the theoretical characteristics of aerosols is forcibly lowered, easily falling below the preset effective correlation threshold.

[0136] According to the logic of Example 1, the system will determine that the current situation is dominated by noise bias, thereby cutting off the parameter update channel. This creates a logical paradox: the harsher the environment, the more the system needs to adjust its parameters to adapt to the environment; however, precisely because of the harsh environment, the system cannot identify effective features and refuses to adjust, ultimately causing the compensation model to gradually fail. To solve the problem of strong vibration noise masking aerosol bias features under coupled interference conditions, leading to a decrease in correlation coefficient and thus mistakenly blocking parameter updates, this example introduces a set of orthogonal subspace projection decoupling steps, specifically:

[0137] Phase 1: Determination of the initiation of orthogonal subspace projection decoupling.

[0138] Specifically, the method in this embodiment first requires precise determination of the initiation conditions. The processing system monitors the current interference coupling state and the preliminarily calculated correlation coefficient in real time. When the interference coupling state is determined to be coupled interference, and the preliminarily calculated correlation coefficient is less than or equal to the effective correlation threshold, the system will not simply determine it as noise-dominated deviation and abandon processing as in Embodiment 1. Instead, it will identify that the system may be in a masked state and trigger the decoupling process. In this embodiment, the effective correlation threshold is set as described above, i.e., 0.7. This means that when the system finds that although it is in a coupled state, the error signal looks completely like noise, the system will start a deep mining mode to try to extract the weak effective signal from the background noise.

[0139] Phase 2: Construction of the vibration disturbance matrix and orthogonal projection operator.

[0140] For example, once the decoupling process is triggered, the core task of the system is to construct a mathematical filter that can accurately filter out vibrational components while retaining aerosol components as much as possible. To this end, the system constructs a vibration interference matrix using synchronously acquired carrier vibration signals. The processing system retrieves inertial measurement unit data synchronized with the current optical signal from memory; this data includes acceleration information acquired by the triaxial accelerometer and angular velocity information acquired by the triaxial gyroscope.

[0141] Assuming the current sliding time window contains N sampling points, the system arranges and combines these six dimensions of vibration data to construct an N-row, six-column matrix, which is the vibration disturbance matrix, denoted in the formula as follows. Each column of this matrix represents a sequence of changes in a specific degree of vibration along the time axis, and the six column vectors span a six-dimensional vibration subspace. Physically, any signal component falling within this subspace is considered by the system to be interference caused by the mechanical motion of the carrier and should be eliminated.

[0142] Specifically, to perform the removal operation, it is necessary to calculate the orthogonal projection operator corresponding to the vibration disturbance matrix. Mathematically, we need to find an operator that can project any vector into the orthogonal complement of the aforementioned vibration subspace. This orthogonal projection operator is denoted in the formula as follows: The calculation formula is as follows:

[0143] ;

[0144] In the formula, Let represent the orthogonal projection operator we need to solve for; it is an N x N square matrix. I represents the N-dimensional identity matrix, with its diagonal elements being 1 and all other elements being zero, representing the full-space projection. The superscript T represents the vibration disturbance matrix constructed above; the superscript T represents the matrix transpose operation, which interchanges the rows and columns of the matrix; the superscript -1 represents the matrix inversion operation; the second term in the formula In linear algebra, it is called a hat matrix or projection matrix, and its function is to project a vector onto a matrix formed by the projection matrix. The projection matrix is ​​the space spanned by the column vectors of the identity matrix. Therefore, subtracting this projection matrix from the identity matrix yields an operator that projects the vectors onto the null space, which is completely perpendicular (or orthogonal) to the vibrational subspace.

[0145] It is also important to note that in actual calculations, to ensure the stability of matrix inversion and prevent matrix singularities caused by collinear data, it is usually necessary to perform a certain step before inversion. Add a small regularization term, or use singular value decomposition to compute the generalized inverse. This step ensures that the system can construct a numerically stable projection operator regardless of the complexity of the carrier's vibration modes.

[0146] Phase 3: Extraction and decorrelation of net aerosol bias.

[0147] Specifically, after obtaining the orthogonal projection operator, the system performs the operation of extracting the net aerosol bias. At this point, the system denotes the original system bias vector, which includes strong noise, in the formula as follows: Left-multiply the orthogonal projection operator The mathematical operation involves projecting the system deviation onto the null space of the vibration disturbance matrix. Physically, this means forcibly eliminating all components in the system deviation that are linearly related to mechanical vibration, regardless of their magnitude.

[0148] For example, the calculation formula for this step is expressed as follows:

[0149] ;

[0150] In the formula, This represents the net aerosol bias obtained after decoupling; The orthogonal projection operator obtained from the aforementioned calculation; This is the original, unprocessed system deviation vector. Through this transformation, the minute deviations caused by changes in aerosol concentration, which were originally submerged in the large-amplitude vibration waveform, are preserved. This is because the variation law of aerosols is physically independent of, and orthogonal to, the mechanical vibration of the carrier, so it will not be filtered out by the projection operator.

[0151] like Figure 6 The intuitive comparison of time-domain waveforms vividly demonstrates the significant effect of using orthogonal subspace projection technology to decouple the deviation signal under strong coupling interference in this embodiment. The horizontal axis in the figure represents the sampling time process, and the vertical axis represents the normalized signal deviation amplitude.

[0152] Figure 6 The large-amplitude red dashed line represents the unprocessed raw system deviation signal, which is mixed with high-energy mechanical vibration noise. Its waveform exhibits violent periodic fluctuations, which completely mask the true aerosol deviation characteristics. If the correlation coefficient is calculated directly at this time, it will be far below the effective threshold.

[0153] In comparison, Figure 6 The smooth blue solid line in the middle represents the net aerosol deviation signal after processing by the orthogonal projection operator. It can be seen that by projecting the original signal onto the null space of the vibration interference matrix, the large vibration interference components are successfully filtered out, and the signal amplitude converges from the original 1.5 to a small range of about 0.2, successfully restoring the linear trend term that was originally hidden in the background of strong noise.

[0154] This comparison demonstrates that the orthogonal subspace projection step can completely remove the interference energy related to mechanical vibration from a physical mechanism perspective, thereby extracting a pure and effective feedback signal. This provides crucial data support for the system to resolve parameter update deadlock and maintain high-precision inversion under extreme conditions.

[0155] Phase 4: Rectify the calculation of correlation coefficients and force the unlocking of parameter updates.

[0156] Specifically, after extracting the net aerosol deviation, the system needs to reassess whether the parameters should be updated. At this point, the system no longer uses the original system deviation, but instead calculates the net aerosol deviation. The corrected correlation coefficient with aerosol inversion characteristics is denoted in the formula as follows. Aerosol inversion characteristics are typically based on theoretical residual trends calculated using current model parameters. Because... Most of the irrelevant vibration noise has been removed, and its waveform characteristics will more purely reflect the model's fit to the aerosol.

[0157] Optionally, the system will calculate the corrected correlation coefficient. A second comparison is performed with the previously set effective correlation threshold of 0.7. If the corrected correlation coefficient is greater than the effective correlation threshold, then... A value greater than 0.7 indicates that, after removing the outer shell containing strong vibration and noise, the system successfully confirmed that the core error source was indeed the mismatch of aerosol model parameters. At this point, the system determined that the update conditions were met, thus forcibly removing the parameter update blocking.

[0158] It is also important to note that although we extracted the effective deviation, the projection operation actually removed a portion of the signal energy, namely the energy of the vibration component. Directly using this energy... Gradient descent may lead to inaccurate estimation of the update step size. Furthermore, parameter updates in noisy environments are always risky. To ensure safety, this embodiment introduces a specific energy scaling factor into the update logic. The system uses the net aerosol deviation and a preset projected energy attenuation factor to iteratively update the parameters.

[0159] Specifically, the projection energy attenuation factor is denoted in the formula as follows: It is defined as the ratio of the L2 norm of net aerosol deviation to the L2 norm of systematic deviation. The calculation formula is as follows:

[0160] ;

[0161] In the formula, The Euclidean norm (2-norm) of a vector is the square root of the sum of the squares of the vector's elements, and is used to characterize the energy of a signal. The value ranges from 0 to 1. When the vibration noise is extremely strong, Very large and Very small Approaching 0; when the vibration noise is very weak, and near, Approaching 1.

[0162] For example, the final parameter update formula can be expressed as:

[0163] ;

[0164] In the formula, Represents the updated model parameters; The model parameters represent the current time step. Represents the base learning rate; This is the projection energy attenuation factor calculated above; This represents the gradient of the loss function calculated based on the net aerosol bias.

[0165] By introducing The system implements an adaptive, cautious update strategy: when extracting signals from strong noise, updates are allowed, but the update magnitude is automatically reduced. The stronger the noise, the more energy is removed during projection, and the more cautious the update becomes. The smaller the value, the less likely the parameters will diverge due to over-adjustment, thus solving the deadlock problem.

[0166] In summary, this embodiment, by introducing orthogonal subspace projection technology based on linear algebra theory, mathematically constructs a hyperplane that completely separates mechanical vibration from aerosol changes. This technique not only solves the potential failure risks faced by Embodiments 1 and 2 under extremely strong coupling interference, but also endows the system with the ability to recover in environments with almost unusable signal-to-noise ratios. It ensures that no matter how severe the carrier vibration, as long as the physical characteristics of aerosols remain in the optical signal, the system can accurately extract them and use them to correct the model, thereby greatly expanding the applicability and engineering value of the technical solution of this invention.

[0167] Example 4:

[0168] like Figure 7 As shown, this embodiment provides a dual-wavelength differential signal adaptive compensation system for aerosol scattering interference. This system is a specific implementation of the methods described in Embodiments 1 to 3 at the hardware and logic architecture level. The system can be deployed in airborne, vehicle-mounted, or fixed high-performance computing platforms, achieving high-fidelity restoration of the detected signal through coordinated hardware and software operations.

[0169] Specifically, this dual-wavelength differential signal adaptive compensation system for aerosol scattering interference mainly comprises five core functional modules: a multi-source signal acquisition module, an interference analysis module, an adaptive decision-making module, a comprehensive compensation module, and a feedback optimization module. These modules are connected via a high-speed data bus, collaboratively completing the entire process from physical signal sensing to intelligent algorithm decision-making and closed-loop feedback optimization.

[0170] First, the multi-source signal acquisition module constitutes the system's sensing front end, responsible for establishing a connection with the physical world. This module is configured to respond to the system's acquisition commands, simultaneously acquiring optical signals and auxiliary sensing signals. In terms of hardware connectivity, this module communicates with the dual-frequency, dual-wavelength laser system via an optoelectronic interface, acquiring real-time dual-wavelength echo data at 532 nm and 1064 nm. Simultaneously, this module connects to the auxiliary sensor array via a serial communication interface or CAN bus. These auxiliary sensors include a millimeter-wave radar for ranging, a high-precision inertial measurement unit for sensing subtle movements of the carrier, and a three-dimensional electric field sensor for monitoring the electromagnetic environment. The core function of the multi-source signal acquisition module lies in the unification of the time reference. It utilizes hardware triggering or a precise time protocol to ensure that all heterogeneous signals are strictly aligned on the time axis, thereby providing a precise data foundation for the heterogeneous frequency domain coherence analysis mentioned in Example 2 and preventing phase errors caused by time slippage.

[0171] Secondly, the interference analysis module is the primary processing hub of the system, used to preprocess the acquired signals, extract feature parameters, and calculate the interference coupling coefficient to determine the interference coupling state. This module integrates signal cleaning and feature extraction algorithms, capable of separating key features such as the ratio of dual-wavelength intensity, signal rate of change, vibration frequency, and amplitude from the original noise. It also calculates feature confidence levels according to the rules described in Example 1, eliminating low-quality data. More importantly, this module incorporates the frequency domain coherence analysis logic described in Example 2. When facing complex conditions where low-frequency vibrations overlap with aerosol turbulence frequencies, the interference analysis module does not simply rely on time-domain amplitude but identifies the frequency domain overlap factor by calculating the squared coherence spectrum of the amplitudes of the dual-wavelength differential signal and the accelerometer signal. Based on this in-depth analysis, the module can accurately decompose signal fluctuations into coherent disturbance components and incoherent aerosol components, thereby calculating the corrected interference coupling coefficient and accurately classifying the current environmental state as aerosol-dominated interference, disturbance-dominated interference, or coupled interference.

[0172] Furthermore, the adaptive decision-making module runs a perturbation-adaptive neural network model, dynamically outputting compensation parameters and sensor fusion weights based on the perturbation coupling state. This module implements the lightweight hybrid network architecture detailed in Embodiment 1. When the perturbation analysis module determines that aerosols dominate, the adaptive decision-making module automatically adjusts the attention mechanism within the network, increasing the weight of aerosol features; when it determines that perturbation dominates, it emphasizes the weight of vibration features. In addition, this module also executes the dual-mode hierarchical update mechanism in Embodiment 1, intelligently selecting a parameter update strategy of rapid tracking, conventional adaptation, or stable convergence based on the rate of aerosol change. Through this dynamic adjustment, the adaptive decision-making module can output the optimal laser signal weights, compensation coefficients, and fusion weights of each sensor, ensuring that the system can make optimal compensation decisions under any operating condition.

[0173] Subsequently, the integrated compensation module, as the system's execution unit, calculates the dual-wavelength differential signal by combining the compensation parameters and correction factors, and outputs the target signal. This module receives dynamic parameters from the adaptive decision module and, in conjunction with real-time calculated vibration correction factors, attitude correction factors, and contaminant correction factors, constructs the final signal compensation equation. As described in the formula of Example 1, this module performs weighted differential processing on the original dual-wavelength signal and introduces an environmental correction term to offset the influence of environmental factors such as humidity. The integrated compensation module's operations have extremely high real-time requirements and typically run on hardware acceleration units such as FPGAs or DSPs to ensure that complex floating-point operations are completed within milliseconds, outputting a smooth and realistic target signal.

[0174] Finally, the feedback optimization module forms a closed-loop guarantee for the system, used to identify the type of deviation in the output signal and iteratively update the model parameters in the adaptive decision module when it is determined to be a system deviation. This module not only performs the deviation classification and identification described in Embodiment 1, distinguishing between random noise and system deviation, but also integrates the orthogonal subspace projection decoupling technique described in Embodiment 3. In the extreme state of strong coupling interference, if the feedback optimization module detects that the correlation coefficient of the error signal is masked by strong noise, it will initiate the decoupling process, construct the vibration interference matrix, and project the system deviation into the orthogonal complement space of the vibration subspace. Through this mathematical transformation, the module can extract the weak net aerosol deviation from the background of strong mechanical vibration and forcibly unlock the parameter update channel based on this deviation. This mechanism breaks the deadlock of traditional systems being unable to self-evolve in harsh environments, giving the system extremely strong robustness.

[0175] In summary, the dual-wavelength differential signal adaptive compensation system for aerosol scattering interference provided in this embodiment perfectly implements all the technical processes described in the aforementioned method embodiments through the organic combination of five functional modules. This system not only solves the problem of state misjudgment caused by frequency domain aliasing but also overcomes the challenge of parameter updates under strong interference, demonstrating significant beneficial effects and greatly improving the adaptability and measurement accuracy of photoelectric detection equipment in complex dynamic environments.

[0176] Example 5:

[0177] Corresponding to the above embodiments, the present invention also proposes an electronic device.

[0178] like Figure 8The diagram shows a structural schematic of an electronic device according to the present invention. The electronic device 100 includes a processor 101 and a memory 103. The processor 101 and the memory 103 are connected, for example, via a bus 102. Optionally, the electronic device 100 may further include a transceiver 104. It should be noted that in practical applications, the transceiver 104 is not limited to one unit, and the structure of this electronic device 100 does not constitute a limitation on the embodiments of the present invention.

[0179] Processor 101 may be a CPU, a general-purpose processor, a DSP, an ASIC, an FPGA, or other programmable logic device, transistor logic device, hardware component, or any combination thereof. It may implement or execute the various exemplary logic blocks, modules, and circuits described in connection with this disclosure. Processor 101 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0180] Bus 102 may include a pathway for transmitting information between the aforementioned components. Bus 102 may be a PCI bus or an EISA bus, etc. Bus 102 may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 8 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0181] The memory 103 stores a computer program corresponding to the dual-wavelength differential signal adaptive compensation method for aerosol scattering interference in the above embodiments of the present invention. This computer program is executed by the processor 101. The processor 101 executes the computer program stored in the memory 103 to implement the content shown in the aforementioned method embodiments.

[0182] Among them, electronic devices 100 include, but are not limited to: mobile terminals such as laptops and PADs (tablet computers) and fixed terminals such as desktop computers. Figure 8 The electronic device 100 shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments of the present invention.

[0183] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. An adaptive compensation method for dual-wavelength differential signals to address aerosol scattering interference, characterized in that, Includes the following steps: In response to the acquisition command of multi-source sensor signals, optical signals and auxiliary sensor signals are acquired simultaneously, and the effective signals and their characteristic parameters are extracted after preprocessing. The interference coupling coefficient is calculated based on the aforementioned characteristic parameters, and the interference coupling state of the current environment is determined accordingly. Based on the interference coupling state, compensation parameters and sensor fusion weights are dynamically output through a perturbation-adaptive neural network model. Based on the compensation parameters, sensor fusion weights, and correction factors, comprehensive compensation is performed on the dual-wavelength differential signal to output the target signal. The feedback optimization process of the method includes: classifying and identifying the deviation between the target signal and the reference signal, distinguishing between systematic deviation and random deviation; if it is determined to be a systematic deviation, further identifying the dominant type of the deviation, and iteratively updating the compensation parameters of the perturbation-adaptive neural network model in combination with the error threshold dynamically adjusted based on the signal-to-noise ratio.

2. The method according to claim 1, characterized in that, The multi-source sensing signals include optical signals acquired by a dual-frequency dual-wavelength laser system, as well as distance, vibration, attitude, and electromagnetic interference signals acquired by auxiliary sensors. The extraction process of the feature parameters includes calculating the feature confidence level, which has a non-linear mapping relationship with the signal-to-noise ratio of the effective signal and the normalized statistical deviation of the feature parameters. When the confidence level of a feature is lower than a preset confidence threshold, the weight of the corresponding feature parameter is reduced in the subsequent inversion process.

3. The method according to claim 1, characterized in that, The interference coupling coefficient is defined as the ratio of the amplitude of the disturbance-dominated signal fluctuation to the amplitude of the aerosol-dominated signal fluctuation. The specific rule for determining the interference coupling state of the current environment is as follows: a first coupling threshold and a second coupling threshold are set, and the first coupling threshold is less than the second coupling threshold; When the interference coupling coefficient is less than the first coupling threshold, it is determined to be aerosol-dominated interference; When the interference coupling coefficient is between the first coupling threshold and the second coupling threshold, it is determined to be coupling interference; When the interference coupling coefficient is greater than the second coupling threshold, it is determined to be disturbance-dominated interference.

4. The method according to claim 3, characterized in that, The method further includes: Perform frequency domain coherence analysis based on heterogeneous signals, calculate the amplitude squared coherence spectrum of the dual-wavelength differential signal and the accelerometer signal, and calculate the frequency domain overlap factor accordingly. When the frequency domain overlap factor is greater than the preset overlap threshold, the dual-wavelength differential signal is decomposed into coherent and incoherent components using the amplitude squared coherence spectrum. The energy of the two components is assigned to the disturbance-dominated amplitude and the aerosol-dominated amplitude, respectively, and the corrected interference coupling coefficient is recalculated. The interference coupling state is determined by applying asymmetric hysteresis logic, and a state escalation threshold and a state degradation threshold are set, wherein the state escalation threshold is greater than the second coupling threshold and the state degradation threshold is less than the second coupling threshold. The switching of the interference coupling state is only performed when the corrected interference coupling coefficient exceeds the state upgrade threshold or the state downgrade threshold respectively; otherwise, the state determination of the previous moment is maintained.

5. The method according to claim 1, characterized in that, The perturbation-adaptive neural network model adopts a hybrid network architecture, including an input layer, a hidden layer, and an output layer; The hidden layer includes a perturbation-aerosol cooperative attention sublayer, whose weight allocation rule is dynamically adjusted in response to the perturbation coupling state: Under aerosol-dominated disturbance conditions, the attention weight of aerosol features is higher than that of perturbation features; Under perturbation-dominated interference conditions, the attention weight of perturbation features is higher than that of aerosol features; Under coupled interference conditions, aerosol features and perturbation features are assigned balanced attention weights.

6. The method according to claim 1, characterized in that, The process of dynamically outputting compensation parameters includes a dual-mode hierarchical update mechanism, which is triggered based on the aerosol change rate: A first rate threshold and a second rate threshold are set, wherein the first rate threshold is less than the second rate threshold; When the aerosol change rate is greater than the second rate threshold, the first update strategy is activated to update the parameters with a first update step size or a first update cycle. When the aerosol change rate is between the first rate threshold and the second rate threshold, the second update strategy is activated, and the parameters are updated using an adaptive learning rate. When the aerosol change rate is less than the first rate threshold, a third update strategy is initiated, introducing a regularization term and updating the parameters with a second update cycle, wherein the second update cycle is greater than the first update cycle.

7. The method according to claim 1, characterized in that, The correction factors include vibration correction factors and attitude correction factors; The vibration correction factor is calculated in segments based on the vibration frequency and vibration amplitude: When the vibration frequency is greater than the preset frequency threshold, the vibration correction factor has a negative linear relationship with the vibration amplitude. When the vibration frequency is less than or equal to a preset frequency threshold, the vibration correction factor is negatively correlated with the product of the vibration amplitude and frequency.

8. The method according to claim 7, characterized in that, The calculation process for the comprehensive compensation is as follows: The dual-wavelength signal is weighted and differentially processed using the compensation parameters, and the target signal is generated by combining the correction factor, pollutant correction factor and environmental correction term. The pollutant correction factor is determined based on the dielectric properties and thickness of the pollutant, and the environmental correction term is positively correlated with the ambient humidity and decreases exponentially with the laser wavelength.

9. The method according to claim 1, characterized in that, The dominant types for identifying this deviation specifically include: Calculate the correlation coefficient between the system bias and the aerosol inversion characteristics; Set an effective threshold for correlation; If the correlation coefficient is greater than the effective threshold of the correlation, it is determined to be an aerosol-dominant bias, triggering an adjustment of the aerosol-related compensation parameters; If the correlation coefficient is less than or equal to the effective correlation threshold, it is determined to be a noise-dominated bias, and parameter adjustment is not triggered or only noise suppression processing is triggered.

10. The method according to claim 9, characterized in that, The method further includes an orthogonal subspace projection decoupling step: When the interference coupling state is determined to be coupling interference, and the preliminarily calculated correlation coefficient is less than or equal to the effective correlation threshold, the decoupling process is triggered; A vibration interference matrix is ​​constructed using synchronously acquired carrier vibration signals, and the orthogonal projection operator corresponding to the vibration interference matrix is ​​calculated. The system bias is multiplied on the left by the orthogonal projection operator, and the system bias is projected onto the null space of the vibration disturbance matrix to obtain the net aerosol bias after decorrelation. Calculate the corrected correlation coefficient between the net aerosol bias and the aerosol inversion characteristics; If the corrected correlation coefficient is greater than the effective correlation threshold, the parameter update blocking is forcibly lifted, and the parameters are iteratively updated using the net aerosol deviation and the preset projection energy attenuation factor.

Images