A method and system for detecting the condition of special vehicle lighting fixtures based on spectral analysis

Abstract

This invention discloses a method and system for detecting the status of special vehicle lamps based on spectral analysis, belonging to the field of optical fault detection. The method includes: simultaneously acquiring mixed spectral signals corresponding to multiple lamps operating simultaneously on a special vehicle, and establishing a linear mixed mathematical model for blind source separation; using an independent component analysis algorithm to blindly separate and independently extract the acquired mixed spectral signals, recovering the independent spectral signal of each lamp from the observed signals, and outputting N statistically independent lamp spectral signals. This invention improves detection accuracy and anti-interference capability in complex environments through steps such as adaptive acquisition of ambient light intensity and calculation of filtering parameters, temperature compensation correction, intelligent matching of detection strategy templates, and electromagnetic interference suppression. This invention solves the problem of difficult separation of mutually coupled spectral signals when multiple lamps are operating simultaneously, achieving real-time and accurate detection of the status of special vehicle lamps in all weather conditions, wide temperature ranges, and multiple scenarios.

Description

Technical Field

[0001] This application belongs to the field of optical fault detection, specifically relating to a method and system for detecting the condition of special vehicle lamps based on spectral analysis. Background Technology

[0002] Special vehicles, serving as mobile platforms for special missions such as public safety, roadside assistance, engineering operations, and fire emergency response, have their operational status directly impacting operational safety and efficiency. Special vehicles are typically equipped with various functional lights, including headlights, warning lights, work lights, fog lights, and turn signals. These lights perform crucial functions such as illumination, warning, and signal transmission in complex operating environments. The condition of vehicle lights directly affects the driving safety and operational effectiveness of special vehicles at night or under adverse conditions. Therefore, accurate and real-time detection of the operational status of special vehicle lights is of significant practical importance. With the rapid development of photoelectric detection technology and artificial intelligence algorithms, spectral analysis, as a non-contact, highly sensitive detection method, is gradually being introduced into the field of vehicle light condition detection. Spectral analysis technology collects spectral characteristic information of the light fixture during emission and compares it with a standard spectral template. This allows for effective determination of light fixture aging, spectral shift, and malfunctions. Compared to traditional image recognition or current detection methods, it has significant advantages in terms of rich information dimensions and strong anti-interference capabilities.

[0003] Among them, the vehicle lighting status detection method based on spectral analysis acquires the radiation spectrum of the target lighting fixture in real time and compares it with a preset standard spectral feature library to achieve quantitative assessment and fault diagnosis of the lighting fixture status. The core of this method lies in the accuracy of spectral feature extraction and the accuracy of the comparison algorithm. Currently published patent documents, such as CN119738036A, disclose a lighting fixture luminous status detection method, system, medium, program, and terminal, which establishes a lighting fixture spectral feature database and uses a pattern matching algorithm for status determination.

[0004] However, existing technologies have significant limitations in adaptability to the complex application scenarios of special vehicles: First, existing spectral detection schemes typically employ fixed-parameter spectral acquisition and filtering strategies, failing to consider the dynamic changes in ambient light intensity in the actual operating environment of special vehicles. During nighttime operations or under conditions of strong light interference, the superposition of ambient light and the target lamp's spectrum leads to a significant reduction in the signal-to-noise ratio of the acquired signal, affecting the accuracy of the detection results. Second, when multiple lamps on a special vehicle operate simultaneously, the spectral signals of different lamps spatially mix and superimpose, making existing time-division multiplexing detection methods inefficient and difficult to effectively separate coupled spectral components. Third, special vehicles typically require continuous operation in extreme temperature environments (-40℃ to +85℃), causing the luminous efficiency and spectral characteristics of the lamps to drift with temperature, while existing technologies lack effective temperature compensation mechanisms. Furthermore, the electromagnetic environment faced by special vehicles is complex; electromagnetic interference generated by motor drives, radio communications, etc., can be superimposed on the spectral signal, affecting the stability of the detection system.

[0005] In summary, existing vehicle lighting detection technologies based on spectral analysis have significant shortcomings in terms of anti-interference capability, environmental adaptability, and multi-target detection efficiency in complex application scenarios. They are unable to meet the stringent requirements of special vehicles for the accuracy and reliability of lighting status detection, and there is an urgent need for a new spectral detection solution for special vehicle lighting in complex application scenarios. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for detecting the status of special vehicle lighting based on spectral analysis, which can effectively solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A method for detecting the condition of lighting fixtures in special vehicles based on spectral analysis includes the following specific steps: Simultaneously collect mixed spectral signals corresponding to multiple lights operating simultaneously on special vehicles, and establish a linear mixed mathematical model for subsequent blind source separation; The independent component analysis algorithm is used to perform blind separation and independent extraction on the acquired mixed spectral signal. The blind separation refers to the process of recovering the source signal from the observed signal by calculating the separation matrix when the mixing matrix and the source signal are unknown, so as to recover the independent spectral signal of each lamp from the observed signal. The linear hybrid mathematical model is represented in matrix form as: X = A·S, where X is the observed signal vector, each element of which represents the hybrid spectral signal acquired by a spectral sensor channel; S is the source signal vector, each element of which corresponds to the independent spectral characteristics of a lamp; A is the hybrid matrix, whose elements represent the hybrid weight coefficients of the spectral signals of each lamp in each observation channel; the independent component analysis algorithm separates the statistically independent source signal vectors from the observed signal vector by finding the separation matrix and maximizing the non-Gaussianity of the independent components, so as to achieve the synchronous acquisition of the independent spectral signals of each lamp; The algorithm based on independent component analysis outputs N statistically independent luminaire spectral signals, where N is the number of luminaires operating simultaneously.

[0008] Furthermore, before synchronously acquiring the mixed spectral signal, the process also includes an adaptive acquisition of ambient light intensity and calculation of filtering parameters: Ambient light intensity signals are collected in real time using a broadband ambient light sensor, and a dynamic mapping relationship between ambient light intensity and integration time and spectral channel gain is established. According to the dynamic mapping relationship, the integration time is adaptively adjusted within the range of 1 millisecond to 1000 milliseconds, and the spectral channel gain is adaptively adjusted within the range of 1x to 100x. The acquired spectral signal is adaptively filtered using a bandwidth-adjustable bandpass filter bank. The bandwidth-adjustable bandpass filter bank includes six center wavelength channels, which respectively cover the violet spectral region from 380 nm to 420 nm, the blue spectral region from 450 nm to 495 nm, the green spectral region from 495 nm to 570 nm, the yellow spectral region from 570 nm to 590 nm, the orange spectral region from 590 nm to 620 nm, and the red spectral region from 620 nm to 750 nm. The filter bandwidth of each spectral channel is adjustable in the range of 5 nanometers to 30 nanometers. When the ambient light intensity is higher than the preset intensity threshold, the filter bandwidth of each spectral channel is automatically narrowed, and when the ambient light intensity is lower than the preset intensity threshold, the filter bandwidth of each spectral channel is automatically widened.

[0009] Furthermore, the filter coefficients of each channel in the bandwidth-adjustable bandpass filter bank are calculated and updated in real time by the main control processor according to the current ambient light intensity level; The filter coefficients for each channel are calculated using a second-order infinite impulse response (IRR) filter structure. The difference equation for the second-order IRR filter structure is expressed as follows: ;in, Let be the current input signal sequence value of the second-order infinite impulse response filter. The output signal sequence value of the second-order infinite impulse response filter at the current moment. The input signal sequence value at the previous time step. These are the input signal sequence values ​​from the previous two time points. The output signal sequence value of the previous time step. These are the output signal sequence values ​​from the first two time points. , , , , These are the coefficients of a second-order infinite impulse response filter; The coefficients of the second-order infinite impulse response filter are calculated using the bilinear transformation method based on the target center wavelength and the filter bandwidth.

[0010] Furthermore, the multi-lamp spectral mixing model is expressed as: ;in, Let N be the observation signal vector, and the N-dimensional observation signal vector... Each element Representing the Mixed spectral signals acquired by each spectral sensor channel; It is an N×N dimensional mixture matrix. elements Indicates the first The spectral signal of the lamp in the first... Mixed weighting coefficients in each observation channel; Let N be the source signal vector, and the N-dimensional source signal vector... Each element Corresponding to the The independent spectral characteristics of each luminaire; The independent component analysis algorithm employs a fast fixed-point algorithm for iterative computation. During the initialization phase of the fast fixed-point algorithm, the initial value of the separation matrix is ​​set to a random orthogonal matrix. The gradient update formula for iteratively updating the separation matrix is: ;in, Let be the separation matrix for the current iteration step. The learning rate parameter, This represents the mathematical expectation operation. Let N be the observed signal vector. This is a nonlinear function used to measure the non-Gaussianity of a signal. nonlinear function The first derivative, Separation matrix Transpose of; When the separation matrix The difference norm between two adjacent iterations is less than The fast fixed-point algorithm is then determined to have converged, and the final separation matrix is ​​output. And through matrix operations Obtain the independent spectral signal for each lamp.

[0011] Furthermore, a preprocessing step for the highly correlated spectral signals is included before employing the independent component analysis algorithm: Calculate the correlation coefficient matrix between signals of each spectral channel. Two or more spectral channels with a correlation coefficient greater than the preset correlation coefficient threshold of 0.85 are determined to be highly correlated channels. The signals from the highly correlated channels are merged and reduced in dimension. The merging method used in the dimensionality reduction process is a weighted average algorithm, and the weights in the weighted average algorithm are determined based on the signal-to-noise ratio of each spectral channel signal. Independent component analysis was performed on the dimensionality-reduced signal to separate the independent spectral signal components of the lamps.

[0012] Furthermore, after obtaining the independent spectral signal for each luminaire, a luminaire operating temperature compensation and correction step is also included: The current operating temperature of the lamp is collected by a non-contact infrared temperature sensor, and the temperature measurement range of the non-contact infrared temperature sensor covers... to ; The database of pre-built temperature-spectral feature mapping relationships is queried. This database is established through multi-point calibration, and the calibration temperature points of the multi-point calibration cover... to The full temperature range, with calibration intervals of Each calibrated temperature point corresponds to the stored peak wavelength offset. Spectral bandwidth variation and changes in radiation intensity ; The spectral feature shift at the measured temperature is calculated using a piecewise linear interpolation algorithm, where the interpolation formula for the peak wavelength shift is: ;in, Measured temperature Peak wavelength shift under, The current operating temperature of the lamp is collected. Lower than the measured temperature And the closest calibration temperature point, Higher than the measured temperature And the closest calibration temperature point, For calibration temperature point Peak wavelength shift under, For calibration temperature point Peak wavelength shift; The compensated spectral features are equal to the measured spectral features minus the offset calculated by the piecewise linear interpolation algorithm.

[0013] Furthermore, before comparing the temperature-compensated spectral signal with the standard spectral feature library, an intelligent matching step for the detection strategy template is also included: The environmental condition sensing module collects environmental parameters in real time from three dimensions: ambient light intensity, weather conditions, and vehicle status. Based on a fuzzy logic reasoning mechanism, the automatic matching and detection strategy template is used, and the environmental condition perception module outputs the membership degree of the ambient light intensity level. Weather state level membership degree And vehicle status level membership The overall scene matching degree is calculated through weighted fusion: ;in, To ensure overall scene matching accuracy, This represents the weighting coefficient for the membership degree of the ambient light intensity level. This refers to the weighting coefficient for the membership degree of weather state levels. The weighting coefficients for the membership degree of the vehicle status level are: 0.4 for the membership degree of the ambient light intensity level, 0.3 for the membership degree of the weather status level, and 0.3 for the membership degree of the vehicle status level. The scene template with the highest matching degree is selected as the current detection strategy. The detection strategy template set includes daytime sunny stationary scene template, daytime sunny driving scene template, nighttime operation scene template, and severe weather scene template.

[0014] Furthermore, before the mixed spectral signal enters the adaptive filtering module after acquisition, an electromagnetic interference suppression processing step is included, as well as a spectral signal quality assessment step after adaptive filtering and temperature compensation are completed. The electromagnetic interference suppression process adopts a combination of wavelet transform and adaptive notch filter. The wavelet transform uses the Daubechies wavelet basis function db4 for 4-level decomposition and uses the soft thresholding method to denoise the high-frequency coefficients. The notch depth of the adaptive notch filter is 40dB and the notch bandwidth of the adaptive notch filter is adjustable from 2Hz to 10Hz. The spectral signal quality assessment step calculates the signal-to-noise ratio (SNR) and spectral feature completeness of the spectral signal. The SNR threshold is set to 15 dB, and the spectral feature completeness threshold is set to 80%. When the SNR of the spectral signal is lower than the SNR threshold or the spectral feature completeness is lower than the spectral feature completeness threshold, an abnormal alarm is triggered and the environmental parameters for the current time period are recorded to the local log.

[0015] A special vehicle lighting status detection system based on spectral analysis includes: a mixed spectral signal synchronous acquisition module, used to synchronously acquire mixed spectral signals corresponding to multiple lights operating simultaneously on a special vehicle; The multi-lamp spectral mixing model construction module is used to establish a linear mixed mathematical model of the observed signal vector and the source signal vector; The blind source separation processing module uses an independent component analysis algorithm to perform blind separation and independent extraction on the acquired mixed spectral signal, recovers the independent spectral signal of each lamp from the observed signal, and outputs N statistically independent lamp spectral signals, where N is the number of lamps working at the same time.

[0016] An adaptive ambient light intensity acquisition module is provided, which has a wide-spectrum ambient light sensor for real-time acquisition of ambient light intensity signals and establishment of a dynamic mapping relationship between integration time and spectral channel gain. The integration time is adaptively adjusted within the range of 1 millisecond to 1000 milliseconds, and the spectral channel gain is adaptively adjusted within the range of 1 to 100 times. A bandwidth-adjustable bandpass filter bank has six center wavelength channels, which respectively cover the violet spectral region from 380 nm to 420 nm, the blue spectral region from 450 nm to 495 nm, the green spectral region from 495 nm to 570 nm, the yellow spectral region from 570 nm to 590 nm, the orange spectral region from 590 nm to 620 nm, and the red spectral region from 620 nm to 750 nm. The filter bandwidth of each spectral channel is adjustable in the range of 5 nm to 30 nm. The temperature compensation and correction module has a non-contact infrared temperature sensor and a temperature-spectral feature mapping database. It is used to collect the working temperature of the lamp and query the pre-built temperature-spectral feature mapping database. It uses a piecewise linear interpolation algorithm to calculate the compensation value to correct the measured spectral features. The detection strategy template matching module has an environmental condition perception unit and a fuzzy logic reasoning unit. It is used to collect environmental parameters in three dimensions: ambient light intensity, weather status, and vehicle status in real time, and automatically match detection strategy templates based on fuzzy logic reasoning. The detection strategy template set includes daytime sunny stationary scene templates, daytime sunny driving scene templates, nighttime operation scene templates, and severe weather scene templates.

[0017] In summary, this application includes at least one of the following beneficial technical effects: 1. This invention acquires ambient light intensity signals in real time and establishes a dynamic mapping relationship. This invention achieves real-time adaptive adjustment of integration time and spectral channel gain. The bandwidth-adjustable adaptive filtering strategy can effectively filter ambient light noise in strong light interference environments, significantly improve the signal-to-noise ratio of spectral signals, and solve the technical problem of insufficient detection accuracy in nighttime operations and strong light interference scenarios.

[0018] 2. This invention establishes a multi-lamp spectral mixing model and uses an independent component analysis algorithm for blind source separation. This invention enables the independent extraction and separation of spectral signals from multiple lamps operating simultaneously, solving the problem of difficulty in judgment caused by the mutual coupling of spectral signals when multiple lamps are operating simultaneously. It can simultaneously acquire the spectral characteristics of each lamp without time-division multiplexing detection, thus greatly improving detection efficiency.

[0019] 3. By pre-establishing a temperature-spectral feature mapping database and performing multi-point calibration across the entire temperature range, this invention achieves online spectral feature compensation across the entire operating temperature range, solving the problem of spectral feature drift caused by the wide operating temperature range of special vehicle lamps and ensuring detection stability under harsh climatic conditions.

[0020] 4. This invention automatically matches a predefined set of detection strategy templates through an environmental condition perception module. This invention enables dynamic and intelligent switching of detection parameters and algorithm combinations. It can adapt to different application scenarios such as daytime, nighttime, sunny days, rainy days, stationary days, and driving days without manual intervention, thus solving the problem of low efficiency caused by the need for manual switching of detection parameters in different application scenarios. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the overall technical solution for a special vehicle lighting condition detection method based on spectral analysis; Figure 2 This is a schematic diagram illustrating the core principle of blind source separation and adaptive filtering for mixed spectral signals from multiple lamps. Figure 3 This is a flowchart of the logic for calculating the adaptive filtering parameters for ambient light intensity and acquiring the spectrum. Figure 4 This is a schematic diagram of the interaction and data flow between temperature compensation and detection strategy template matching; Figure 5 This is a flowchart illustrating the process of electromagnetic interference suppression and spectral signal quality assessment. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this invention clearer, the following description is provided in conjunction with the appendix. Figure 1 To be continued Figure 5 The present invention will be further described in detail below with reference to specific embodiments.

[0023] The first aspect is the method for detecting the status of special vehicle lamps based on spectral analysis proposed in this invention, which covers core technical aspects such as ambient light intensity adaptive filtering parameter calculation, multi-lamp mixed spectral signal acquisition and mixed model construction, blind source separation processing of mixed spectral signals, lamp operating temperature compensation correction, intelligent matching of detection strategy templates, and lamp status determination. The specific implementation of each technical aspect is described in detail below.

[0024] The first step is S1, the adaptive acquisition and filtering parameter calculation step for ambient light intensity. The spectral acquisition and filtering parameters are dynamically adjusted according to the real-time ambient light intensity to suppress ambient light interference and improve the signal-to-noise ratio. The specific steps are as follows.

[0025] Step S101: Ambient light intensity signal acquisition. The ambient light intensity signal is acquired in real time using a broadband ambient light sensor. The broadband ambient light sensor has a spectral response range covering 300nm to 1100nm, enabling it to sense ambient light radiation energy in the ultraviolet to near-infrared region. The sampling frequency of the broadband ambient light sensor is set to 100Hz to capture rapid changes in ambient light conditions.

[0026] In engineering implementation, a silicon photodiode is used as the photoelectric conversion element. The photosensitive area of ​​the silicon photodiode is set to 5mm × 5mm, and the dark current is controlled below 10 nanoamps to ensure measurement accuracy in low-light environments. The analog signal output from the silicon photodiode is digitized by a 24-bit analog-to-digital converter, and the digitized signal is sent to the main control processor for subsequent calculations.

[0027] Step S102: Setting the dynamic update cycle for the filter threshold. The update cycle for the main control processor to dynamically calculate the filter threshold is set to 50 milliseconds. Every 50 milliseconds, the main control processor recalculates the filter threshold parameter based on the latest ambient light intensity measurement value.

[0028] Step S103: Establish a dynamic mapping relationship between ambient light intensity and acquisition parameters, and establish a dynamic mapping relationship between the filter threshold and ambient light intensity. The dynamic mapping relationship is described by a piecewise linear function: When the ambient light intensity is lower than the first preset intensity threshold T1, the integration time is set to the first preset integration time Tint1, and the spectral channel gain is set to the first preset gain factor G1.

[0029] When the ambient light intensity is between the first preset intensity threshold T1 and the second preset intensity threshold T2, the integration time is dynamically adjusted to the second preset integration time range Tint2, and the spectral channel gain is adjusted to the second preset gain factor range G2.

[0030] When the ambient light intensity is higher than the second preset intensity threshold T2, the integration time is set to the third preset integration time Tint3, and the spectral channel gain is adjusted to the third preset gain multiple range G3.

[0031] Step S104: Smoothing adjustment of integration time and gain. The main control processor receives the current ambient light intensity measurement value I_env. The main control processor compares I_env with the first preset intensity threshold T1 and the second preset intensity threshold T2, and determines the current light intensity level range based on the comparison result.

[0032] Based on the current interval, the main control processor calls the corresponding parameter mapping table to obtain the initial values ​​of the target integration time and spectral channel gain.

[0033] To avoid the impact of parameter abrupt changes on spectral acquisition, a gradual adjustment strategy is implemented for the integration time: when the integration time is adjusted from a steady-state value to the target value, the adjustment range does not exceed 20% of the current value each time, and the adjustment interval is an integer multiple of the acquisition period.

[0034] The spectral channel gain is adjusted via a programmable gain amplifier. The programmable gain amplifier supports seven gain settings: 1x, 2x, 5x, 10x, 20x, 50x, and 100x. The adjustment response time of the programmable gain amplifier is less than 1 millisecond.

[0035] Step S105: Adaptive filtering is implemented using a bandwidth-adjustable bandpass filter bank. The adaptive filtering is achieved using a bandwidth-adjustable bandpass filter bank structure. The bandwidth-adjustable bandpass filter bank contains 6 center wavelength channels, covering: The violet spectral region from 380 nm to 420 nm; The blue spectral region from 450nm to 495nm; The green spectral region from 495nm to 570nm; The yellow spectral region from 570nm to 590nm; The orange spectral region from 590 nm to 620 nm; The red spectral region from 620nm to 750nm.

[0036] The bandwidth of each spectral channel is adjustable through digital filter technology, with the filter bandwidth ranging from 5nm to 30nm. When the ambient light intensity is high, the filter bandwidth automatically narrows to 5nm to 10nm to exclude more ambient light noise from entering the signal channel; when the ambient light intensity is low, the filter bandwidth is appropriately widened to 15nm to 30nm to ensure sufficient signal throughput.

[0037] Step S106: Real-time calculation and updating of filter coefficients. The filter coefficients of each channel in the bandwidth-adjustable bandpass filter bank are calculated and updated in real time by the main control processor based on the current ambient light intensity level. The filter coefficients are calculated using a second-order infinite impulse response filter structure. The difference equation of the second-order infinite impulse response filter structure is expressed as: in: This represents the current input signal sequence value of the filter; The output signal sequence value of the filter at the current moment; This represents the input signal sequence value from the previous time step. These are the input signal sequence values ​​from the previous two time points; This represents the output signal sequence value from the previous time step. These are the output signal sequence values ​​from the previous two time points; , , , , These are the filter coefficients.

[0038] Filter coefficients , , , , The calculation is based on the target center wavelength and the filter bandwidth, and is obtained from the analog filter prototype through the bilinear transformation method.

[0039] Step S107: Ambient light intensity feature vector generation. The output signals of each channel of the bandwidth-adjustable bandpass filter bank are integrated by a multiplexer to form an ambient light intensity feature vector containing information from six wavelength channels. The ambient light intensity feature vector is used for subsequent fine-tuning of adaptive filtering parameters.

[0040] Through the aforementioned adaptive operations, ambient light interference is suppressed at the source, ensuring the acquisition of high signal-to-noise ratio spectral signals. The generated ambient light intensity feature vector serves as the basis for further fine-tuning of the filtering parameters, ensuring that the parameter configuration of the acquisition link matches the current lighting conditions.

[0041] Step S2 involves synchronously acquiring mixed spectral signals emitted by multiple simultaneously operating lamps using a spectral sensor, preprocessing the original signals, and establishing a linear mixed mathematical model for subsequent blind source separation. The specific steps are as follows.

[0042] Step S201: Configure the spectral sensor, using a back-illuminated scientific-grade complementary metal-oxide-semiconductor sensor as the core photosensitive element. The back-illuminated scientific-grade complementary metal-oxide-semiconductor sensor has a pixel array of 2048×2048, a pixel size of 13 μm×13 μm, and a quantum efficiency exceeding 90% in the visible to near-infrared band.

[0043] The optical system of the spectral sensor employs an achromatic lens group design. The achromatic lens group design has an effective focal length of 50 mm, a relative aperture of F / 2.0, and a field of view of 25 degrees, which can cover the imaging requirements of typical special vehicle lighting distribution areas.

[0044] The spectral resolution of the spectral sensor is set to 2 nanometers. The spectral dispersion function is achieved by integrating an adjustable slit and a diffraction grating in front of the spectral sensor. The diffraction grating is a blazed grating with 1200 lines per millimeter and a blaze wavelength of 500 nanometers.

[0045] Step S202: Synchronous acquisition of mixed spectral signals. The mixed spectral signals corresponding to multiple simultaneously operating lamps on the special vehicle are synchronously acquired using the spectral sensor configured in step S201. During the acquisition process, the spectral sensor images the target area, and each lamp forms a corresponding spectral pixel area on the spectral sensor.

[0046] This embodiment extracts the region of interest (ROI) for each lamp and performs spectral calculations on the pixels within the ROI to obtain the spectral radiance distribution curve for each lamp. The integration time for spectral acquisition is determined based on the adaptive filtering parameters in step S1, ranging from 1 millisecond to 1000 milliseconds, and supports microsecond-level fine adjustment. The spectral channel gain is also controlled by the output of the adaptive filtering module in step S1, with a gain adjustment range of 1 to 100 times.

[0047] Step S203: Establish a multi-lamp spectral mixing model based on the independent component analysis algorithm. Let N be the number of lamps operating simultaneously on the special vehicle, and each lamp emits a light signal with unique spectral characteristics. The spectral mixing model is expressed as: in: Let N be the observed signal vector. Each element Representing the The first spectral sensor channel or the first Mixed spectral signals collected from each lighting fixture area; The mixture is an N×N dimensional mixture matrix. elements Indicates the first The spectral signal of the lamp in the first... Mixed weighting coefficients in each observation channel; Let N be the source signal vector, and let the source signal vector be... Each element Corresponding to the The independent spectral characteristics of each luminaire are determined by the properties of the luminescent material, the parameters of the driving circuit, and the operating status of the luminaire.

[0048] Mixed matrix This reflects the linear superposition relationship between the light signals from each lamp during spatial propagation and spectral sensor acquisition. Within a short acquisition window, the mixing matrix... Maintaining this constant provides a mathematical basis for subsequent blind source separation processing.

[0049] In this embodiment, the number N of lamps working simultaneously is the total number of lamps pre-configured in the special vehicle controller, or the number of lamps currently in the powered-on state can be read in real time through the vehicle controller local area network.

[0050] Step S204: Preprocessing of the original spectral signal. In constructing the spectral mixing model, this embodiment first preprocesses the acquired original spectral signal. The preprocessing process includes three steps: dark current subtraction, flat field correction, and bad pixel correction.

[0051] Dark current subtraction operation: Under the same integration time conditions, the output of the spectral sensor when there is no light signal is acquired, the dark current image is calculated, and the dark current image is subtracted from the original acquired image to eliminate the interference of dark current on weak light signals.

[0052] Flat field correction operation: The reflectance spectrum of a uniform diffuse reflection standard white board is collected to obtain the spectral response uniformity curve of this embodiment. The spectral response uniformity curve is used to normalize and correct the spectral response of each pixel to eliminate the brightness non-uniformity problem caused by the optical system.

[0053] Defect correction operation: By comparing the signal values ​​of adjacent pixels, fixed defective pixels and noise anomalies on the spectral sensor are identified and corrected.

[0054] Step S205, Spectral Downsampling and Feature Compression: The second step in preprocessing is spectral downsampling and feature compression. To reduce the computational complexity of the subsequent blind source separation algorithm, this embodiment performs downsampling processing on the high-resolution spectral data, reducing the original spectral data from 2 nanometer resolution to 5 nanometer resolution. Downsampling is implemented using a third-order spline interpolation algorithm.

[0055] The feature compression operation uses principal component analysis to compress the spectral feature dimensions, retaining principal components with a cumulative contribution rate of over 95% as input features for subsequent blind source separation.

[0056] The data processed in step S2 eliminates interference from sensor noise and optical system inhomogeneity, and compresses high-dimensional spectral data into a low-dimensional feature space. This provides a high-quality, low-complexity input signal for the blind source separation algorithm in step S3, ensuring that the independent spectral features of each lamp can be effectively separated in the subsequent steps.

[0057] Step S3 uses an independent component analysis algorithm to blindly separate and independently extract the mixed spectral signal acquired in step S2, recover the independent spectral signal of each lamp from the observed signal, and extract the feature parameters used for state determination. This is carried out according to the following steps.

[0058] Step S301, Blind source separation problem definition: The goal of the independent component analysis algorithm is to separate the blind source from the observed signal vector. Recover the source signal vector The recovery process involves finding the separation matrix. To achieve, making Separation matrix It is a mixed matrix The estimated value of the inverse matrix. In the mixture matrix Under the condition of full rank, the separation matrix It exists and is unique.

[0059] Step S302: Fast Fixed-Point Algorithm Initialization. The independent component analysis algorithm uses the Fast Fixed-Point Algorithm for iterative calculations. The Fast Fixed-Point Algorithm is also known as the FastICA algorithm. The core idea of ​​the FastICA algorithm is to achieve a measure of statistical independence by maximizing the non-Gaussianity of independent components. In blind separation applications of spectral signals, the FastICA algorithm handles the problem of maximizing the negative entropy of spectral data.

[0060] During the algorithm initialization phase, the separation matrix is... The initial value is set to a random orthogonal matrix.

[0061] Step S303: Iteratively update the separation matrix. During the iterative calculation, first calculate the linear combination of the separation matrix and the observed signal. Then through a nonlinear function right A transformation is performed to extract non-Gaussian features. Nonlinear function. Choose hyperbolic tangent function ,in A constant factor, The value range is from 1 to 2.

[0062] The gradient update formula is expressed as: in: This is the separation matrix for the current iteration step; The learning rate parameter. Control the step size of each update; Represents the mathematical expectation operation; The N-dimensional observation signal vector defined in step S2; nonlinear function nonlinear functions Used to measure the non-Gaussianity of a signal; nonlinear function The first derivative; Separation matrix The transpose of .

[0063] Step S304, convergence determination: The iterative convergence condition is that the statistical independence metric of independent components is less than a preset convergence threshold. The statistical independence of independent components is determined by the separation matrix. The orthogonality of column vectors is measured. When the separation matrix... The difference norm between two adjacent iterations is less than At that point, it is determined that the algorithm has converged.

[0064] Step S305: Output independent spectral signals; the algorithm outputs the final separation matrix. Through matrix operations The independent spectral signal of each luminaire is obtained. The output of the blind source separation process is N independent luminaire spectral signals, each of which contains the radiation intensity distribution curve of the corresponding luminaire in the visible light band from 380 nm to 750 nm.

[0065] Step S306: Processing highly correlated spectral signals. When the spectral signals of multiple lamps are highly correlated, i.e., when multiple lamps use the same luminescent material or have similar spectral peak wavelengths, the independent component analysis algorithm cannot effectively separate the signal sources. To address this issue, a principal component analysis preprocessing step is added before independent component analysis.

[0066] First, calculate the correlation coefficient matrix between the signals of each spectral channel. When the correlation coefficient is greater than a preset correlation coefficient threshold... Two or more spectral channels are identified as highly correlated channels. A preset correlation coefficient threshold is used. The value is 0.85.

[0067] Then, the signals from these highly correlated channels are merged and their dimensions reduced. The merging method uses a weighted average algorithm, with the weights determined based on the signal-to-noise ratio of each channel.

[0068] Finally, independent component analysis was performed on the dimension-reduced signal to separate the independent luminaire spectral signal components.

[0069] Step S307: Extract spectral feature parameters. In this embodiment, feature parameters of each independent spectral signal are extracted. These feature parameters include spectral peak wavelength, spectral center wavelength, spectral bandwidth, spectral radiant intensity integral value, and spectral shape feature vector. These feature parameters are used for lamp status determination in subsequent step S5.

[0070] Step S3 separates the pure spectral signal of each lamp from the mixed spectral signal acquired in step S2, and extracts key features that can characterize the working status of the lamp. These features will be used as input for comparison with the standard spectral feature library in step S5, thereby determining the lamp status.

[0071] Step S4 involves collecting the current operating temperature of the lamp using a temperature sensor, querying a pre-built temperature-spectral feature mapping database to obtain the spectral shift at the corresponding temperature, calculating the compensation value using a piecewise linear interpolation algorithm, correcting the measured spectral features, and eliminating the influence of temperature drift on spectral detection. This is carried out according to the following steps.

[0072] Step S401: Acquisition of lamp operating temperature. A non-contact infrared temperature sensor is used to acquire the current operating temperature of the lamp. The non-contact infrared temperature sensor has a temperature measurement range of -40°C to +150°C, a measurement accuracy of ±1°C, and a response time of 100ms. The non-contact infrared temperature sensor is installed near the lamp housing, and its installation position accurately reflects the actual operating temperature of the lamp.

[0073] The output signal of the non-contact infrared temperature sensor is amplified by a low-noise amplifier and then sent to an analog-to-digital converter for digital processing. The sampling frequency of the analog-to-digital converter is set to 10Hz to obtain stable temperature measurement values.

[0074] Step S402: Query the temperature-spectral feature mapping relationship database. The temperature-spectral feature mapping relationship database is established through multi-point calibration, with calibration temperature points covering the entire temperature range from -40°C to +85°C, and calibration intervals of 10°C. At each calibration temperature point, this embodiment performs no fewer than 100 spectral sample acquisitions for the currently detected lamp type to obtain statistically reliable spectral feature shifts.

[0075] The temperature-spectral feature mapping database stores the spectral feature offset corresponding to each calibrated temperature point. The spectral feature offset includes the peak wavelength offset. Spectral bandwidth variation and changes in radiation intensity The temperature-spectral feature mapping database is stored using a hash table data structure, with the calibrated temperature value as the key for fast lookup.

[0076] Step S403: Calculate the spectral feature shift using piecewise linear interpolation; temperature compensation employs a piecewise linear interpolation algorithm. When the measured temperature... Between two adjacent calibration temperature points and In this embodiment, the two calibration temperature points closest to the measured temperature are first located, i.e. and Then, the measured temperature is calculated using linear interpolation. The corresponding spectral feature offset.

[0077] The interpolation formula for peak wavelength offset is: in: Measured temperature Peak wavelength shift; For calibration temperature point Peak wavelength shift; For calibration temperature point Peak wavelength shift; The current operating temperature of the lamp is collected in step S401; Lower than the measured temperature And the closest calibration temperature point; Higher than the measured temperature And the closest calibration temperature point. Spectral bandwidth variation. and changes in radiation intensity The same piecewise linear interpolation algorithm is used for calculation.

[0078] Step S404: Perform spectral characteristic temperature compensation. The compensated spectral characteristic is equal to the measured spectral characteristic minus the offset calculated by interpolation. The specific calculation process is as follows: First, the measured spectral feature parameters are extracted from the independent spectral signal output in step S3. The measured spectral feature parameters include peak wavelength, spectral bandwidth and radiance.

[0079] Then, the estimated peak wavelength shift at the current temperature is calculated according to the interpolation formula in step S403. Estimates of spectral bandwidth variation and estimated value of change in radiation intensity .

[0080] Finally, each offset is subtracted from the measured spectral characteristics to obtain the temperature-compensated spectral characteristic values. The compensated peak wavelength is equal to the measured peak wavelength minus... The compensated spectral bandwidth is equal to the measured spectral bandwidth minus The compensated radiation intensity is equal to the measured radiation intensity minus .

[0081] Step S4 eliminates the measurement error caused by the drift of the spectral characteristics of the luminaire in the wide temperature range of -40°C to +85°C for special vehicles, so that the compensated spectral characteristics can accurately reflect the inherent state of the luminaire, providing an accurate data basis for the comparison with the standard spectral feature library in step S5.

[0082] Finally, step S5, the detection strategy template matching and lamp status determination step, is carried out by collecting information on three dimensions in real time: ambient light intensity, weather status, and vehicle status through the environmental condition perception module. Based on fuzzy logic reasoning, the detection strategy template is automatically matched, the temperature-compensated spectral signal is compared with the standard spectral feature library, the working status of each lamp is determined, and electromagnetic interference suppression processing is performed on the collected signal. The following steps are followed.

[0083] Step S501: Multi-dimensional environmental parameter acquisition. The environmental condition perception module acquires environmental parameters in real time, including three dimensions: ambient light intensity, weather conditions, and vehicle conditions.

[0084] The ambient light intensity parameter is obtained by the broadband ambient light sensor in step S1.

[0085] Weather condition parameters are acquired through onboard meteorological sensors or external communication interfaces. Weather conditions are categorized into six levels: sunny, cloudy, overcast, rainy, snowy, and foggy.

[0086] Vehicle status parameters are obtained through the vehicle controller's local area network. Vehicle status includes five levels: stationary, idling, low-speed driving, medium-speed driving, and high-speed driving.

[0087] Step S502: Configure the detection strategy template set. The detection strategy template set contains at least 4 preset scene templates.

[0088] The daytime sunny static scene template is suitable for special vehicles in stationary conditions with good ambient lighting. The integration time of the daytime sunny static scene template is set to 50ms to 100ms, the spectral channel gain is 10x to 20x, the filtering bandwidth is 20nm to 30nm, and the comparison algorithm uses Euclidean distance metric.

[0089] The daytime clear weather driving scenario template is suitable for special vehicles in operation. The integration time of the daytime clear weather driving scenario template is set to 20ms to 50ms, the spectral channel gain is 20x to 50x, the filtering bandwidth is 15nm to 20nm, and the comparison algorithm uses Mahalanobis distance to eliminate the influence of differences in the dimensions of each spectral feature.

[0090] The nighttime operation scenario template is suitable for working conditions at night or in low light intensity environments. The integration time of the nighttime operation scenario template is set to 500ms to 1000ms, the spectral channel gain is 50x to 100x, the filtering bandwidth is 10nm to 15nm, and the comparison algorithm uses weighted cosine similarity measurement.

[0091] The severe weather scenario template is suitable for low visibility conditions such as rain, snow, and fog. The integration time of the severe weather scenario template is set to 200ms to 500ms, the spectral channel gain is 30x to 80x, the filtering bandwidth is 10nm to 20nm, and the comparison algorithm uses a dynamic time warping algorithm to compensate for the time distortion of the spectral signal.

[0092] Step S503: Scene template matching based on fuzzy logic. Automatic scene template matching is achieved based on a fuzzy logic reasoning mechanism. The environmental condition perception module outputs three membership values: ambient light intensity level membership value. Weather state level membership degree And vehicle status level membership The membership function adopts the triangular membership function definition, and the membership function parameters of each level are determined based on statistical data from typical application conditions.

[0093] The overall scene matching degree is calculated through weighted fusion: in: To ensure overall scene matching accuracy; Membership degree of ambient light intensity level; The degree of membership to the weather state level; The membership degree of the vehicle status level; , , These are weighting coefficients, which are set based on the degree of influence of each environmental parameter on the detection performance. , , The values ​​are 0.4, 0.3, and 0.3 respectively.

[0094] In this embodiment, the scene template with the highest matching degree is selected as the current detection strategy, and the detection parameters and algorithm parameters corresponding to the scene template with the highest matching degree are loaded into the spectrum acquisition and processing module.

[0095] Step S504: Compare the spectral signal with the standard spectral feature library. The processed spectral signal is compared with the standard spectral feature library to determine the operating status of each lamp. The standard spectral feature library stores standard spectral feature templates for various special vehicle lamps. These templates include typical spectral features for normal, slightly aged, severely aged, and faulty states.

[0096] The comparison process first normalizes the temperature-compensated spectral signal, normalizing the spectral intensity values ​​to the range of 0 to 1. Then, it calculates the similarity distance between the normalized spectral signal and each standard spectral feature template. The calculation method for the similarity distance is determined based on the current scene template, and the output is a similarity vector with each standard spectral feature template.

[0097] The lamp status determination adopts the maximum membership principle, selecting the status corresponding to the standard spectral feature template with the smallest similarity distance as the current lamp status determination status.

[0098] Step S505, Electromagnetic Interference Suppression Processing: Electromagnetic interference suppression processing is performed after the spectral signal is acquired and before it enters the adaptive filtering module. The electromagnetic interference suppression processing uses a combination of wavelet transform and adaptive notch filter.

[0099] Wavelet transform uses the Daubechies wavelet basis function db4 for 4-level decomposition. Wavelet decomposition decomposes the spectral signal into low-frequency approximation coefficients and high-frequency detail coefficients. The high-frequency coefficients are denoised using a soft thresholding method. Finally, the denoised signal is recovered through wavelet reconstruction.

[0100] An adaptive notch filter is used to suppress periodic electromagnetic interference pulses. The notch depth of the adaptive notch filter is 40dB, the notch center frequency is dynamically determined based on the actual electromagnetic interference spectrum analysis results, and the notch bandwidth is adjustable from 2Hz to 10Hz.

[0101] Step S5 enables dynamic adaptive switching of detection parameters and algorithm combinations, adapting to different application scenarios such as daytime, nighttime, sunny days, rainy days, stationary days, and driving days without manual intervention. Finally, it outputs the working status judgment results of each lamp, completing the entire process of special vehicle lamp status detection.

[0102] In addition, a quality assessment of the spectral signal is included. After adaptive filtering and temperature compensation, this embodiment calculates the signal-to-noise ratio (SNR) and spectral feature completeness of the spectral signal. The SNR is calculated as the ratio of signal power to noise power. The signal power is obtained by integrating the energy within a 20nm wide peak region centered on the spectral peak wavelength. The noise power is estimated by estimating the signal standard deviation within a baseline region in the 700nm to 750nm band, far from any lamp characteristic peak. Spectral feature completeness is obtained by calculating the energy coverage of the spectral signal within predefined key characteristic bands for each lamp type. A coverage rate below 80% is considered a feature missing. The quality assessment thresholds are: a SNR_th threshold of 15dB and a spectral feature completeness threshold of 80%. When the SNR is lower than the SNR_th threshold or the spectral feature completeness is lower than the spectral feature completeness threshold of 80%, an anomaly alarm is triggered, and the environmental parameters for the current time period are recorded to the local log for subsequent analysis.

[0103] To facilitate understanding, the actual operation process of the method of the present invention will be explained in detail below with reference to specific application scenarios.

[0104] Assume a special vehicle is performing a nighttime operation, simultaneously activating four lights: headlights, fog lights, side marker lights, and brake lights. In a low-light environment at night, the measured ambient light intensity is approximately 0.1 lux, lower than a first preset intensity threshold T1, which is set to 10 lux. This embodiment determines that the current operating condition is low light intensity and automatically adjusts the integration time to a first preset integration time Tint1, set to 800 ms. The spectral channel gain is also adjusted to a first preset gain factor G1, set to 80 times. The adaptive filtering module broadens the bandwidth of each spectral channel to 25 nm to obtain more signal energy.

[0105] A spectral sensor synchronously acquires the mixed spectral signals from four lamps. Since the four lamps operate simultaneously and are spatially close, the signal acquired by the spectral sensor is a linear mixture of the spectral signals from the four lamps. In this embodiment, a four-dimensional spectral mixing model X equal to A·S is constructed, where the four elements of the observed signal vector X correspond to the mixed spectra acquired by the four sensor channels, the mixing matrix A is a 4×4 matrix, and the four elements of the source signal vector S correspond to the independent spectral characteristics of the four lamps.

[0106] The blind source separation module uses the FastICA algorithm to separate the mixed spectra. After initialization, the algorithm performs iterative calculations. In each iteration, the update gradient of the separation matrix W is calculated and the W matrix is ​​updated. The algorithm is considered to have converged when the difference norm between two consecutive iterations of the separation matrix W is less than 10 to the power of -6. The algorithm outputs the separation matrix W, and the independent spectral signals of the four lamps are recovered through matrix operations. Assume that the brake lights use a red LED array with a peak wavelength in the range of 620nm to 630nm; the headlights use white LEDs with a spectrum that is a composite spectrum of blue light chip excited yellow phosphor, exhibiting a bimodal characteristic; the fog lights use yellow LEDs with a peak wavelength in the range of 580nm to 590nm; and the marker lights use red LEDs with a peak wavelength in the range of 630nm to 640nm. Since there is some overlap in the spectra of the brake lights and marker lights, this embodiment adds principal component analysis preprocessing before independent component analysis. The spectral channels of the two lamps are correlated, and when the correlation coefficient exceeds 0.85, merging and dimensionality reduction processing is performed to improve the separation accuracy.

[0107] The temperature sensor detected the brake lamp's operating temperature as 65°C, which falls between the calibration temperatures of 60°C and 70°C. In this embodiment, the temperature-spectral feature mapping database was consulted to obtain the spectral feature offsets at the two calibration temperatures of 60°C and 70°C. The offset at 65°C was calculated using piecewise linear interpolation: the peak wavelength offset Δλ_peak was -2.3 nm, the spectral bandwidth change ΔFWHM was +4.1 nm, and the radiant intensity change ΔI was -15.2%. Subtracting the interpolated offsets from the measured spectral features yielded the temperature-compensated spectral features.

[0108] The environmental condition perception module integrates the membership values ​​of ambient light intensity level, weather condition level, and vehicle condition level. Under nighttime operation conditions, the membership degree μ_I for ambient light intensity level is 0.9 for nighttime and 0.7 for clear weather; the membership degree μ_W for weather condition level is 0.8 for clear weather; and the membership degree μ_V for vehicle condition level is 0.3 for stationary and 0.7 for moving. Weighted fusion calculation yields the highest matching degree for the nighttime operation scenario. In this embodiment, a nighttime operation scenario template is loaded, the integration time is adjusted to 800ms, the spectral channel gain is adjusted to 80 times, the filter bandwidth is adjusted to 12nm, and the comparison algorithm uses a weighted cosine similarity metric.

[0109] The temperature-compensated spectral signals of the four lamps were compared with a standard spectral feature library. The weighted cosine similarity of the brake lamp's spectral signal with the normal state template was 0.97, and with the slightly aged state template was 0.82; the weighted cosine similarity of the headlight's spectral signal with the normal state template was 0.95, and with the slightly aged state template was 0.79; the weighted cosine similarity of the fog lamp's spectral signal with the normal state template was 0.94; and the weighted cosine similarity of the marker lamp's spectral signal with the normal state template was 0.93. In this embodiment, all four lamps were determined to be in normal condition, and a detection result report was output.

[0110] Suppose a sudden strong light interference occurs during the detection process, such as a vehicle behind turning on its high beams. At this time, the ambient light intensity measurement rapidly rises to 500 lux, exceeding the second preset intensity threshold T2, which is set to 200 lux. This embodiment determines that the current condition is high light intensity and reduces the integration time to the third preset integration time Tint3, set to 10 ms. The spectral channel gain is also reduced to the third preset gain factor G3, set to 5 times. The adaptive filtering module narrows the bandwidth of each spectral channel to 8 nm to filter out strong light interference signals. The electromagnetic interference suppression module detects periodic electromagnetic interference pulses generated by the high beams and activates an adaptive notch filter to suppress interference, with the notch center frequency set to 50 Hz and the notch depth to 40 dB. The spectral signal quality assessment module monitors the signal-to-noise ratio (SNR) changes in real time, outputting an alarm message when the SNR falls below 15 dB. The above detection process continues to run under strong light interference conditions to determine the operating status of each lamp.

[0111] On the other hand, the special vehicle lighting condition detection system based on spectral analysis proposed in this invention includes: The mixed spectral signal synchronous acquisition module is used to synchronously acquire the mixed spectral signals corresponding to multiple lights that are working simultaneously on special vehicles; The multi-lamp spectral mixing model construction module is used to establish a linear mixed mathematical model of the observed signal vector and the source signal vector; The blind source separation processing module uses an independent component analysis algorithm to perform blind separation and independent extraction on the acquired mixed spectral signals, recovers the independent spectral signal of each lamp from the observed signal, and outputs N statistically independent lamp spectral signals, where N is the number of lamps working at the same time.

[0112] An adaptive ambient light intensity acquisition module is equipped with a wide-spectrum ambient light sensor, which is used to acquire ambient light intensity signals in real time and establish a dynamic mapping relationship between integration time and spectral channel gain. The integration time is adaptively adjusted within the range of 1 millisecond to 1000 milliseconds, and the spectral channel gain is adaptively adjusted within the range of 1 to 100 times. The bandwidth-adjustable bandpass filter bank has six center wavelength channels, which cover the violet spectral region from 380 nm to 420 nm, the blue spectral region from 450 nm to 495 nm, the green spectral region from 495 nm to 570 nm, the yellow spectral region from 570 nm to 590 nm, the orange spectral region from 590 nm to 620 nm, and the red spectral region from 620 nm to 750 nm, respectively. The filter bandwidth of each spectral channel is adjustable in the range of 5 nm to 30 nm. The temperature compensation and correction module has a non-contact infrared temperature sensor and a temperature-spectral feature mapping database. It is used to collect the operating temperature of the lamp and query the pre-built temperature-spectral feature mapping database. It uses a piecewise linear interpolation algorithm to calculate the compensation value to correct the measured spectral features. The detection strategy template matching module has an environmental condition perception unit and a fuzzy logic reasoning unit. It is used to collect environmental parameters in three dimensions: ambient light intensity, weather status, and vehicle status in real time. Based on fuzzy logic reasoning, it automatically matches detection strategy templates. The detection strategy template set includes daytime sunny stationary scene templates, daytime sunny driving scene templates, nighttime operation scene templates, and severe weather scene templates.

[0113] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention. Therefore, the embodiments should be regarded as exemplary and non-limiting in all respects.

[0114] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A method for detecting the state of a special vehicle lamp based on spectral analysis, characterized in that, Includes the following steps: Simultaneously collect mixed spectral signals corresponding to multiple lights operating simultaneously on special vehicles, and establish a linear mixed mathematical model for subsequent blind source separation; The independent component analysis algorithm is used to perform blind separation and independent extraction on the acquired mixed spectral signal. The blind separation refers to the process of recovering the source signal from the observed signal by calculating the separation matrix when the mixing matrix and the source signal are unknown, so as to recover the independent spectral signal of each lamp from the observed signal. The linear hybrid mathematical model is represented in matrix form as: X = A·S, where X is the observed signal vector, each element of which represents the hybrid spectral signal acquired by a spectral sensor channel; S is the source signal vector, each element of which corresponds to the independent spectral characteristics of a lamp. A is a mixing matrix, whose elements represent the mixing weight coefficients of the spectral signals of each lamp in each observation channel; the independent component analysis algorithm separates the statistically independent source signal vectors from the observed signal vector by finding the separation matrix and maximizing the non-Gaussianity of the independent components, so as to realize the synchronous acquisition of the independent spectral signals of each lamp. The algorithm based on independent component analysis outputs N statistically independent luminaire spectral signals, where N is the number of luminaires operating simultaneously.

2. The method for detecting the state of a special vehicle lamp based on spectral analysis according to claim 1, characterized in that, Before synchronously acquiring the mixed spectral signal, the process also includes an adaptive acquisition of ambient light intensity and calculation of filtering parameters. Ambient light intensity signals are collected in real time using a broadband ambient light sensor, and a dynamic mapping relationship between ambient light intensity and integration time and spectral channel gain is established. According to the dynamic mapping relationship, the integration time is adaptively adjusted within the range of 1 millisecond to 1000 milliseconds, and the spectral channel gain is adaptively adjusted within the range of 1x to 100x. The acquired spectral signal is adaptively filtered using a bandwidth-adjustable bandpass filter bank. The bandwidth-adjustable bandpass filter bank includes six center wavelength channels, which respectively cover the violet spectral region from 380 nm to 420 nm, the blue spectral region from 450 nm to 495 nm, the green spectral region from 495 nm to 570 nm, the yellow spectral region from 570 nm to 590 nm, the orange spectral region from 590 nm to 620 nm, and the red spectral region from 620 nm to 750 nm. The filter bandwidth of each spectral channel is adjustable in the range of 5 nanometers to 30 nanometers. When the ambient light intensity is higher than the preset intensity threshold, the filter bandwidth of each spectral channel is automatically narrowed, and when the ambient light intensity is lower than the preset intensity threshold, the filter bandwidth of each spectral channel is automatically widened.

3. The method for detecting the state of a special vehicle lamp based on spectral analysis according to claim 2, characterized in that, The filter coefficients of each channel in the bandwidth-adjustable bandpass filter bank are calculated and updated in real time by the main control processor according to the current ambient light intensity level. The filter coefficients for each channel are calculated using a second-order infinite impulse response (IRR) filter structure. The difference equation for the second-order IRR filter structure is expressed as follows: ;in, Let be the current input signal sequence value of the second-order infinite impulse response filter. The output signal sequence value of the second-order infinite impulse response filter at the current moment. The input signal sequence value at the previous time step. These are the input signal sequence values ​​from the previous two time points. The output signal sequence value of the previous time step. These are the output signal sequence values ​​from the first two time points. , , , , These are the coefficients of a second-order infinite impulse response filter; The coefficients of the second-order infinite impulse response filter are calculated using the bilinear transformation method based on the target center wavelength and the filter bandwidth.

4. The method for detecting the state of a special vehicle lamp based on spectral analysis according to claim 1, characterized in that, The multi-lamp spectral mixing model is expressed as follows: ;in, Let N be the N-dimensional observation signal vector. Each element Representing the Mixed spectral signals acquired by each spectral sensor channel; It is an N×N dimensional mixture matrix. elements Indicates the first The spectral signal of the lamp in the first... Mixed weighting coefficients in each observation channel; Let N be the source signal vector, and the N-dimensional source signal vector... Each element Corresponding to the The independent spectral characteristics of each luminaire; The independent component analysis algorithm employs a fast fixed-point algorithm for iterative computation. During the initialization phase of the fast fixed-point algorithm, the initial value of the separation matrix is ​​set to a random orthogonal matrix. The gradient update formula for iteratively updating the separation matrix is: ;in, Let be the separation matrix for the current iteration step. The learning rate parameter, This represents the mathematical expectation operation. Let N be the observed signal vector. This is a nonlinear function used to measure the non-Gaussianity of a signal. nonlinear function The first derivative, Separation matrix Transpose of; When the separation matrix The difference norm between two adjacent iterations is less than The fast fixed-point algorithm is then determined to have converged, and the final separation matrix is ​​output. And through matrix operations Obtain the independent spectral signal for each lamp.

5. The method for detecting the state of a special vehicle lamp based on spectral analysis according to claim 1, characterized in that, Before employing the independent component analysis algorithm, a preprocessing step for the highly correlated spectral signals is also included: Calculate the correlation coefficient matrix between signals of each spectral channel. Two or more spectral channels with a correlation coefficient greater than the preset correlation coefficient threshold of 0.85 are determined to be highly correlated channels. The signals from the highly correlated channels are merged and reduced in dimension. The merging method used in the dimensionality reduction process is a weighted average algorithm, and the weights in the weighted average algorithm are determined based on the signal-to-noise ratio of each spectral channel signal. Independent component analysis was performed on the dimensionality-reduced signal to separate the independent spectral signal components of the lamps.

6. The method for detecting the state of a special vehicle lamp based on spectral analysis according to claim 1, characterized in that, After obtaining the independent spectral signal for each luminaire, a luminaire operating temperature compensation and correction step is also included: The current working temperature of the lamp is collected by a non-contact infrared temperature measurement sensor, and the temperature measurement range of the non-contact infrared temperature measurement sensor covers to ; The database of pre-built temperature-spectral feature mapping relationships is queried. This database is established through multi-point calibration, and the calibration temperature points of the multi-point calibration cover... to The full temperature range, with calibration intervals of Each calibrated temperature point corresponds to the stored peak wavelength offset. Spectral bandwidth variation and changes in radiation intensity ; The spectral feature shift at the measured temperature is calculated using a piecewise linear interpolation algorithm, where the interpolation formula for the peak wavelength shift is: ;in, Measured temperature Peak wavelength shift under, The current operating temperature of the lamp is collected. Lower than the measured temperature And the closest calibration temperature point, Higher than the measured temperature And the closest calibration temperature point, For calibration temperature point Peak wavelength shift under, For calibration temperature point Peak wavelength shift; The compensated spectral features are equal to the measured spectral features minus the offset calculated by the piecewise linear interpolation algorithm.

7. The method for detecting the state of a special vehicle lamp based on spectral analysis according to claim 1, characterized in that, Before comparing the temperature-compensated spectral signal with the standard spectral feature library, an intelligent matching step for the detection strategy template is also included: The environmental condition sensing module collects environmental parameters in real time from three dimensions: ambient light intensity, weather conditions, and vehicle status. Based on a fuzzy logic reasoning mechanism, the detection strategy template is automatically matched, and the environmental condition perception module outputs the membership degree of the ambient light intensity level. Weather state level membership degree And vehicle status level membership The overall scene matching degree is calculated through weighted fusion: ;in, To ensure overall scene matching accuracy, This represents the weighting coefficient for the membership degree of the ambient light intensity level. This refers to the weighting coefficient for the membership degree of weather state levels. The weighting coefficients for the membership degree of the vehicle status level are: 0.4 for the membership degree of the ambient light intensity level, 0.3 for the membership degree of the weather status level, and 0.3 for the membership degree of the vehicle status level. The scene template with the highest matching degree is selected as the current detection strategy. The detection strategy template set includes daytime sunny stationary scene template, daytime sunny driving scene template, nighttime operation scene template, and severe weather scene template.

8. The method for detecting the state of a special vehicle lamp based on spectral analysis according to claim 1, characterized in that, Before entering the adaptive filtering module after acquiring the mixed spectral signal, the process also includes an electromagnetic interference suppression step, and a spectral signal quality assessment step after completing adaptive filtering and temperature compensation. The electromagnetic interference suppression process adopts a combination of wavelet transform and adaptive notch filter. The wavelet transform uses the Daubechies wavelet basis function db4 for 4-level decomposition and uses the soft thresholding method to denoise the high-frequency coefficients. The notch depth of the adaptive notch filter is 40dB and the notch bandwidth of the adaptive notch filter is adjustable from 2Hz to 10Hz. The spectral signal quality assessment step calculates the signal-to-noise ratio (SNR) and spectral feature completeness of the spectral signal. The SNR threshold is set to 15 dB, and the spectral feature completeness threshold is set to 80%. When the SNR of the spectral signal is lower than the SNR threshold or the spectral feature completeness is lower than the spectral feature completeness threshold, an abnormal alarm is triggered and the environmental parameters for the current time period are recorded to the local log.

9. A special vehicle lamp state detection system based on spectral analysis, characterized in that A method for detecting the condition of special vehicle lighting fixtures based on spectral analysis, as described in any one of claims 1 to 8, wherein the system for detecting the condition of special vehicle lighting fixtures based on spectral analysis comprises: The mixed spectral signal synchronous acquisition module is used to synchronously acquire the mixed spectral signals corresponding to multiple lights that are working simultaneously on special vehicles; The multi-lamp spectral mixing model construction module is used to establish a linear mixed mathematical model of the observed signal vector and the source signal vector; The blind source separation processing module uses an independent component analysis algorithm to perform blind separation and independent extraction on the acquired mixed spectral signal, recovers the independent spectral signal of each lamp from the observed signal, and outputs N statistically independent lamp spectral signals, where N is the number of lamps working at the same time.

10. The special vehicle lamp state detection system based on spectral analysis according to claim 9, characterized in that, Also includes: An adaptive ambient light intensity acquisition module is provided, which has a wide-spectrum ambient light sensor for real-time acquisition of ambient light intensity signals and establishment of a dynamic mapping relationship between integration time and spectral channel gain. The integration time is adaptively adjusted within the range of 1 millisecond to 1000 milliseconds, and the spectral channel gain is adaptively adjusted within the range of 1 to 100 times. A bandwidth-adjustable bandpass filter bank has six center wavelength channels, which respectively cover the violet spectral region from 380 nm to 420 nm, the blue spectral region from 450 nm to 495 nm, the green spectral region from 495 nm to 570 nm, the yellow spectral region from 570 nm to 590 nm, the orange spectral region from 590 nm to 620 nm, and the red spectral region from 620 nm to 750 nm. The filter bandwidth of each spectral channel is adjustable in the range of 5 nm to 30 nm. The temperature compensation and correction module has a non-contact infrared temperature sensor and a temperature-spectral feature mapping database. It is used to collect the working temperature of the lamp and query the pre-built temperature-spectral feature mapping database. It uses a piecewise linear interpolation algorithm to calculate the compensation value to correct the measured spectral features. The detection strategy template matching module has an environmental condition perception unit and a fuzzy logic reasoning unit. It is used to collect environmental parameters in three dimensions: ambient light intensity, weather status, and vehicle status in real time, and automatically match detection strategy templates based on fuzzy logic reasoning. The detection strategy template set includes daytime sunny stationary scene templates, daytime sunny driving scene templates, nighttime operation scene templates, and severe weather scene templates.

Images