Thin film spectral door lock management system and method based on data analysis

Abstract

The present invention discloses a thin film spectral door lock management system and method based on data analysis, relating to the field of data analysis technology. The method comprises the following steps: Step 1: Collecting intact thin film spectral data from the same batch, disturbed spectral samples, and corresponding environmental parameters; Step 2: Denoising and smoothing the spectral data, performing baseline correction, derivative calculations, and normalization; Step 3: Extracting key spectral features to construct vectors, which are then combined and decomposed into basis and coefficient matrices using a non-negative matrix; Step 4: Calculating intact and disturbed thresholds, respectively, and constructing a regression model using environmental parameters and coefficient offsets; Step 5: Collecting real-time spectra and environmental parameters, projecting them after preprocessing to solve the coefficient vector, and dynamically adjusting the judgment criteria and making decisions. This method can effectively improve the existing thin film spectral door lock management system, which primarily relies on fixed thresholds and is difficult to adapt to disturbances and environmental changes.

Description

Technical Field

[0001] The present invention relates to the technical field of data analysis, and in particular to a thin film spectral door lock management system and method based on data analysis. Background Art

[0002] Thin film spectroscopy technology accurately characterizes the composition, structure, and state of materials by analyzing the absorption, reflection, or scattering properties of light. It has wide applications in security, industrial inspection, and other fields. The core of the door lock management method based on spectral characteristics is to distinguish between legitimate thin films and illegal disturbances by extracting the characteristic parameters of the thin film spectrum and establishing a discrimination model. Traditional methods usually rely on fixed thresholds or simple machine learning models, using the statistical characteristics of spectral data (such as peak position and intensity distribution) as the basis for discrimination. However, when dealing with complex environmental interference or multiple types of disturbances, such methods often suffer from insufficient feature characterization capabilities or limited model adaptability, resulting in reduced discrimination accuracy.

[0003] Existing thin film spectral door lock management technology mainly faces two major bottlenecks: First, the environmental adaptability defect of fixed thresholds. Traditional solutions mostly set fixed discrimination thresholds based on spectral data under standard environments. When environmental parameters (such as temperature, humidity, and light intensity) fluctuate, the spectral data will drift due to changes in the physical and chemical properties of the material, resulting in the fixed threshold being unable to accurately distinguish normal fluctuations from illegal disturbances, which can easily lead to misjudgment or missed judgments; second, the dynamic characterization capability of disturbance characteristics is insufficient. Existing methods lack effective modeling of the spectral characteristics of various types of disturbances such as surface contamination, scratches, and chemical corrosion of thin films. They only use a single threshold to distinguish between intact and disturbed states, making it difficult to capture subtle differences between different types of disturbances, and unable to dynamically track the impact of environmental changes on disturbance characteristics, resulting in poor robustness of the system in complex scenarios. Summary of the Invention

[0004] The purpose of the present invention is to provide a thin film spectral door lock management system and method based on data analysis to solve the problems raised in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a thin film spectral door lock management method based on data analysis, the method comprising the following steps:

[0006] Step 1: Collect the spectral data of intact thin films, disturbed spectral samples and corresponding environmental parameters of the same batch;

[0007] Step 2: Denoising, smoothing, baseline correction, derivative calculation and normalization are performed on the spectral data;

[0008] Step 3: Extract the key features of the spectrum to construct a vector, merge it, and use non-negative matrix decomposition to form a basis and coefficient matrix;

[0009] Step 4: Calculate the intact and disturbed thresholds respectively, and build a regression model using environmental parameters and coefficient offsets;

[0010] Step 5: Collect real-time spectra and environmental parameters, project and solve coefficient vectors after preprocessing, dynamically adjust the judgment criteria and make decisions.

[0011] In step 1, the spectral data of P intact films of the same batch are collected under a set standard environment, which is expressed as S0: S0={S0 1 ,S0 2 ,…,S0 P}; Among them, S0 1 ,S0 2 ,…,S0 P Represent the 1st, 2nd, ..., Pth spectral data respectively, and each spectral data is an m-dimensional vector;

[0012] For Q types of disturbances, N data are collected for each type of disturbance. q samples, denoted as C q :C q ={S q 1 ,S q 2 ,…,S q Nq}; where q∈{1,2,…,Q}, N q Indicates the number of samples corresponding to the qth type of disturbance, S q 1 ,S q 2 ,…,S q Nq They represent the 1st, 2nd, ..., Nth disturbances corresponding to the qth type of disturbance. q The disturbance types can be bending, pollution, temperature mutation, etc. The disturbance samples need to cover the non-destructive interference scenarios that may be encountered in practical applications.

[0013] Synchronously record the M-dimensional environment vector at each acquisition, expressed as E: E={E1,E2,…,E M}; where E1, E2, …, E M Represent the 1st, 2nd, ..., Mth dimensional environmental data respectively; environmental data include temperature, humidity, light, etc.;

[0014] Construct a multi-scenario dataset covering multiple states, including normal, disturbed, and illegally tampered, to improve model generalization. The introduction of environmental parameters provides a data foundation for subsequent dynamic calibration, preventing misjudgment of single spectral features due to environmental interference.

[0015] In step 2, the intact spectral data and the disturbed spectral data are normalized:

[0016] Including denoising and smoothing: using sliding window filtering, convolution operation is performed on the intact spectrum and the disturbed spectrum data, and the smoothed spectrum is output;

[0017] It also includes baseline correction: for the smoothed spectrum, the spectrum baseline is approximated by polynomial fitting, and the baseline is subtracted to obtain the baseline-corrected spectrum;

[0018] It also includes derivative operations: for the baseline-corrected spectrum, the finite difference method is used to calculate the first-order or second-order derivative, and the derivative spectrum is obtained by selecting the first-order or second-order derivative;

[0019] Normalization is also included: for derivative spectra, the spectral range is divided into different bands, and the spectral values ​​in each band are normalized by Z-score.

[0020] Sliding window filtering reduces random noise through neighborhood averaging; polynomial baseline correction assumes that the baseline is a slowly changing curve; derivative operation uses mathematical differences to highlight changes in spectral slope; Z-score normalization makes the data in each band obey the distribution with mean 0 and variance 1, avoiding the model being dominated by high-amplitude bands.

[0021] In step 3, key features are extracted from the preprocessed spectrum, including peak position, peak intensity, derivative features and statistical features, and the continuous spectrum curve is converted into a finite-dimensional feature vector D: D = [D1, D2, ..., D n ] T ; Where n represents the number of key features;

[0022] Matrix decomposition modeling: Combine the eigenvectors of the intact spectrum and the perturbed spectrum into an n×N matrix X; where N=P+Σ q=1 Q N q ;

[0023] Using the non-negative matrix factorization algorithm, it is decomposed into two parts: X≈W·H;

[0024] W=[w1,w2,…,w k ], represents the n×k basis matrix;

[0025] H represents a k×N coefficient matrix: each spectral sample corresponds to a set of coefficients, indicating the combination ratio of each basis.

[0026] NMF decomposes complex spectra into linear combinations of basic components through non-negative constraints (spectral absorbance is non-negative). Each column of W represents a spectral primitive (e.g., W1 corresponds to the characteristic peak combination of normal film, W2 corresponds to the characteristic peak combination of oxidative damage), and each row of H represents the contribution of each primitive in the sample.

[0027] In step 4, calculate the intact threshold R0: the coefficient vector {h j 0} j=1 P , calculate the centroid μ0 and covariance matrix Σ0, and define the intact threshold: R0=α·(λ max (Σ0)) 1 / 2 ;

[0028] Among them, α represents the empirical constant, λ max (Σ0) represents the maximum eigenvalue of the covariance matrix Σ0;

[0029] Calculate the disturbance threshold R1: Calculate the maximum distance between all disturbance sample coefficient vectors and μ0: R1=max j=P+1 N ||h j -μ0||2;

[0030] Using the environmental parameter set E and the corresponding coefficient offset Δh=h-μ0, a regression model is constructed: Δh=f(E);

[0031] Wherein, f represents the mapping function;

[0032] R0 evaluates the maximum discreteness of normal samples through the maximum eigenvalue of the covariance matrix. After amplification by α, it ensures that almost all normal samples are contained within the threshold; R1 is the farthest boundary of the perturbed sample, distinguishing between normal and abnormal; the regression model converts environmental interference into predictable coefficient offsets. For example, when the temperature increases, the molecular vibration intensifies, and the coefficient of a certain characteristic peak increases linearly.

[0033] In step 5, when the door lock triggers the identification process, the spectrum data S of the current film is collected. new , synchronously obtain real-time environmental parameters E new ;

[0034] For the collected spectral data S of the thin film new , execute the preprocessing process to obtain the feature vector D new ;

[0035] The preprocessed feature vector D new Projecting onto the basis matrix W, set the optimization problem: , solve to get the real-time coefficient vector h new ;

[0036] According to the real-time environmental parameter E new , prediction coefficient offset: Δh'=f(E new ), and dynamically adjust the judgment benchmark: μ new =μ0+Δh'; At the same time, the intact threshold and the disturbance threshold are adjusted respectively: ;

[0037] Among them, E0 is the standard environmental parameter, and β and γ are environmental sensitivity coefficients.

[0038] Calculate the adjusted distance: ; Determine the status according to the distance: when d ≤ R0′, it is determined as a normal thin film and unlocking is allowed; when R0′ < d ≤ R1′, it is determined as a disturbed thin film, triggering an alarm but allowing unlocking; when d > R1′, it is determined as an illegal thin film, rejecting unlocking and triggering an alarm;

[0039] Dynamically adjust the reference and threshold to compensate for the systematic influence of environmental changes on the spectrum (for example, when the temperature deviates from the standard value, the coefficient mean of the normal spectrum shifts accordingly), to avoid misjudgment; The three-level decision-making balances security and practicality, allowing use under slight disturbances (such as normal wear) and intercepting serious damage.

[0040] A thin-film spectrum door lock management system based on data analysis, which includes a data acquisition module, a spectrum processing module, a feature modeling module, a threshold modeling module, and a dynamic decision-making module;

[0041] The data acquisition module is used to collect the spectral data of intact thin films of the same batch, disturbed spectral samples, and corresponding environmental parameters; the spectrum processing module is used to perform denoising and smoothing, baseline correction, derivative operation, and normalization processing on the spectral data; the feature modeling module is used to extract the key features of the spectrum to construct vectors, and after merging, use non-negative matrix factorization to obtain the basis and coefficient matrices; the threshold modeling module is used to calculate the intact and disturbed thresholds respectively, and use environmental parameters and coefficient offsets to construct a regression model; the dynamic decision-making module is used to collect real-time spectra and environmental parameters, preprocess them, project and solve the coefficient vector, and dynamically adjust the decision-making reference and make decisions.

[0042] The data acquisition module includes an intact spectrum unit, a disturbed spectrum unit, and an environmental parameter unit;

[0043] The intact spectrum unit is used to collect the standard spectral data of intact thin films of the same batch; the disturbed spectrum unit is used to collect spectral samples under various disturbed states; the environmental parameter unit is used to synchronously record environmental parameters.

[0044] The spectrum processing module includes a denoising and smoothing unit, a baseline correction unit, a derivative processing unit, and a normalization unit;

[0045] The denoising and smoothing unit is used to achieve spectral denoising through sliding window convolution; the baseline correction unit is used to eliminate baseline drift through polynomial fitting; the derivative processing unit is used to calculate the derivative spectrum by the finite difference method; the normalization unit is used to perform normalization processing on the spectral data.

[0046] The feature modeling module includes a feature extraction unit and a matrix decomposition unit;

[0047] The feature extraction unit is used to extract key spectral features; the matrix decomposition unit is used to generate a basis matrix by non-negative matrix decomposition.

[0048] The threshold modeling module includes an intact threshold unit, a disturbance threshold unit and a regression modeling unit;

[0049] The intact threshold unit is used to calculate the decision boundary of the intact film sample; the disturbance threshold unit is used to determine the maximum deviation threshold of the disturbance sample; the regression modeling unit is used to construct the environmental coefficient offset mapping model;

[0050] The dynamic decision module includes a real-time projection unit, a reference adjustment unit and a state determination unit;

[0051] The real-time projection unit is used to collect new thin film spectrum data and solve the coefficient vector of the real-time spectrum; the reference adjustment unit is used to dynamically correct the judgment threshold according to the real-time environment; and the state judgment unit is used to make hierarchical decisions on the door lock state.

[0052] Compared with existing technologies, the present invention has the following advantages: by establishing a correlation model between environmental parameters and spectral feature offsets, the present invention analyzes the impact of the current environment on the spectrum in real time and dynamically adjusts the discrimination benchmark and threshold range. When environmental conditions change, the system predicts the reasonable fluctuation range of spectral features based on historical data and expands or contracts the threshold accordingly, thereby accurately distinguishing "normal changes caused by the environment" from "abnormalities caused by illegal disturbances", effectively improving the reliability of discrimination in complex environments and significantly reducing the risk of false positives and false negatives. The present invention uses feature extraction and dimensionality reduction techniques for spectral data to convert complex spectral curves into low-dimensional feature vectors, stripping away noise and highlighting the essential differences between different disturbances. Different types of disturbances will exhibit different distribution patterns in the feature space. The system can achieve early warning prompts for minor disturbances and precise interception of serious anomalies through a multi-layer threshold mechanism (such as dividing the three states into "normal allowed", "disturbance warning", and "illegal alarm"). The present invention uses data analysis and machine learning algorithms to dynamically update model parameters based on long-term sample data and continuously learn the changing patterns of thin film spectral features (such as spectral baseline drift caused by material aging). This data-driven adaptive capability enables the system to automatically adapt to long-term environmental changes and material performance degradation. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 Schematic diagram of the steps of the thin film spectral door lock management method based on data analysis of the present invention;

[0054] Figure 2The figure is a flow chart of the thin film spectral door lock management system based on data analysis of the present invention. DETAILED DESCRIPTION

[0055] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0056] Example: Figure 1-Figure 2 As shown, the present invention provides a technical solution, a thin film spectral door lock management method based on data analysis, which includes the following steps:

[0057] Step 1: Collect the spectral data of intact thin films, disturbed spectral samples and corresponding environmental parameters of the same batch;

[0058] Step 2: Denoising, smoothing, baseline correction, derivative calculation and normalization are performed on the spectral data;

[0059] Step 3: Extract the key features of the spectrum to construct a vector, merge it, and use non-negative matrix decomposition to form a basis and coefficient matrix;

[0060] Step 4: Calculate the intact and disturbed thresholds respectively, and build a regression model using environmental parameters and coefficient offsets;

[0061] Step 5: Collect real-time spectra and environmental parameters, project and solve coefficient vectors after preprocessing, dynamically adjust the judgment criteria and make decisions.

[0062] In step 1, the spectral data of P intact films of the same batch are collected under a set standard environment, which is expressed as S0: S0={S0 1 ,S0 2 ,…,S0 P}; Among them, S0 1 ,S0 2 ,…,S0 P Represent the 1st, 2nd, ..., Pth spectral data respectively, and each spectral data is an m-dimensional vector;

[0063] For Q types of disturbances, N data are collected for each type of disturbance. q samples, denoted as C q :C q ={S q 1 ,S q 2 ,…,S q Nq}; where q∈{1,2,…,Q}, N qIndicates the number of samples corresponding to the qth type of disturbance, S q 1 ,S q 2 ,…,S q Nq They represent the 1st, 2nd, ..., Nth disturbances corresponding to the qth type of disturbance. q The disturbance types can be bending, pollution, temperature mutation, etc. The disturbance samples need to cover the non-destructive interference scenarios that may be encountered in practical applications.

[0064] Synchronously record the M-dimensional environment vector at each acquisition, expressed as E: E={E1,E2,…,E M}; where E1, E2, …, E M Represent the 1st, 2nd, ..., Mth dimensional environmental data respectively; environmental data include temperature, humidity, light, etc.;

[0065] Construct a multi-scenario dataset covering multiple states, including normal, disturbed, and illegally tampered, to improve model generalization. The introduction of environmental parameters provides a data foundation for subsequent dynamic calibration, preventing misjudgment of single spectral features due to environmental interference.

[0066] In step 2, the intact spectral data and the disturbed spectral data are normalized:

[0067] Including denoising and smoothing: using sliding window filtering, convolution operation is performed on the intact spectrum and the disturbed spectrum data, and the smoothed spectrum is output;

[0068] It also includes baseline correction: for the smoothed spectrum, the spectrum baseline is approximated by polynomial fitting, and the baseline is subtracted to obtain the baseline-corrected spectrum;

[0069] It also includes derivative operations: for the baseline-corrected spectrum, the finite difference method is used to calculate the first-order or second-order derivative, and the derivative spectrum is obtained by selecting the first-order or second-order derivative;

[0070] Normalization is also included: for derivative spectra, the spectral range is divided into different bands, and the spectral values ​​in each band are normalized by Z-score.

[0071] Sliding window filtering reduces random noise through neighborhood averaging; polynomial baseline correction assumes that the baseline is a slowly changing curve; derivative operation uses mathematical differences to highlight changes in spectral slope; Z-score normalization makes the data in each band obey the distribution with mean 0 and variance 1, avoiding the model being dominated by high-amplitude bands.

[0072] In step 3, key features are extracted from the preprocessed spectrum, including peak position, peak intensity, derivative features and statistical features, and the continuous spectrum curve is converted into a finite-dimensional feature vector D: D = [D1, D2, ..., Dn ] T ; Where n represents the number of key features;

[0073] Matrix decomposition modeling: Combine the eigenvectors of the intact spectrum and the perturbed spectrum into an n×N matrix X; where N=P+Σ q=1 Q N q ;

[0074] Using the non-negative matrix factorization algorithm, it is decomposed into two parts: X≈W·H;

[0075] W=[w1,w2,…,w k ], represents the n×k basis matrix;

[0076] H represents a k×N coefficient matrix: each spectral sample corresponds to a set of coefficients, indicating the combination ratio of each basis.

[0077] NMF decomposes complex spectra into linear combinations of basic components through non-negative constraints (spectral absorbance is non-negative). Each column of W represents a spectral primitive (e.g., W1 corresponds to the characteristic peak combination of normal film, W2 corresponds to the characteristic peak combination of oxidative damage), and each row of H represents the contribution of each primitive in the sample.

[0078] In step 4, calculate the intact threshold R0: the coefficient vector {h j 0} j=1 P , calculate the centroid μ0 and covariance matrix Σ0, and define the intact threshold: R0=α·(λ max (Σ0)) 1 / 2 ;

[0079] Among them, α represents the empirical constant, λ max (Σ0) represents the maximum eigenvalue of the covariance matrix Σ0;

[0080] Calculate the disturbance threshold R1: Calculate the maximum distance between all disturbance sample coefficient vectors and μ0: R1=max j=P+1 N ||h j -μ0||2;

[0081] Using the environmental parameter set E and the corresponding coefficient offset Δh=h-μ0, a regression model is constructed: Δh=f(E);

[0082] Wherein, f represents the mapping function;

[0083] R0 evaluates the maximum dispersion of normal samples through the maximum eigenvalue of the covariance matrix, and α is amplified to ensure that almost all normal samples are included within the threshold; R1 is the farthest boundary of the perturbed samples, distinguishing normal from abnormal; the regression model converts environmental interference into predictable coefficient offsets. For example, when the temperature rises, the molecular vibration intensifies, and the coefficient of a certain characteristic peak increases linearly.

[0084] In step 5, when the door lock triggers the recognition process, the spectral data S of the current thin film is collected new , and the real-time environmental parameters E are obtained synchronously new ;

[0085] For the spectral data S of the collected thin film new , a preprocessing process is performed to obtain the feature vector D new ;

[0086] The preprocessed feature vector D new is projected onto the basis matrix W, and the optimization problem is set: , and the real-time coefficient vector h is obtained by solving new ;

[0087] According to the real-time environmental parameters E new , the coefficient offset is predicted: Δh’ = f(E new ), and the decision criterion μ new is dynamically adjusted: μ = μ0 + Δh’; at the same time, the intact threshold and the perturbation threshold are adjusted respectively: ;

[0088] where E0 is the standard environmental parameter, and β and γ are environmental sensitivity coefficients.

[0089] Calculate the adjusted distance: ; The status is determined according to the distance: when d ≤ R0′, it is determined as a normal thin film and unlocking is allowed; when R0′ < d ≤ R1′, it is determined as a perturbed thin film, and an alarm is triggered but unlocking is allowed; when d > R1′, it is determined as an illegal thin film, and unlocking is refused and an alarm is triggered;

[0090] The reference and threshold are dynamically adjusted to compensate for the systematic influence of environmental changes on the spectrum (for example, when the temperature deviates from the standard value, the coefficient mean of the normal spectrum shifts accordingly), avoiding misjudgment; the three-level decision-making balances security and practicality, allowing use under slight perturbations (such as normal wear) and intercepting serious damage.

[0091] A thin film spectral door lock management system based on data analysis, which includes a data acquisition module, a spectral processing module, a feature modeling module, a threshold modeling module, and a dynamic decision-making module;

[0092] The data acquisition module is used to collect intact film spectral data, disturbed spectral samples and corresponding environmental parameters of the same batch; the spectral processing module is used to perform denoising and smoothing, baseline correction, derivative operation and normalization on the spectral data; the feature modeling module is used to extract key spectral features to construct vectors, and after merging, decompose them into basis and coefficient matrices using non-negative matrices; the threshold modeling module is used to calculate intact and disturbed thresholds respectively, and construct a regression model using environmental parameters and coefficient offsets; the dynamic decision module is used to collect real-time spectra and environmental parameters, project and solve the coefficient vector after preprocessing, dynamically adjust the judgment benchmark and make decisions.

[0093] The data acquisition module includes an intact spectrum unit, a disturbed spectrum unit and an environmental parameter unit;

[0094] The intact spectrum unit is used to collect standard spectrum data of intact films of the same batch; the disturbed spectrum unit is used to collect spectrum samples under various disturbance states; and the environmental parameter unit is used to synchronously record environmental parameters.

[0095] The spectrum processing module includes a denoising and smoothing unit, a baseline correction unit, a derivative processing unit and a normalization unit;

[0096] The denoising and smoothing unit is used for sliding window convolution to achieve spectral noise reduction; the baseline correction unit is used for polynomial fitting to eliminate baseline drift; the derivative processing unit is used for calculating derivative spectrum by finite difference method; and the normalization unit is used for normalizing spectral data.

[0097] The feature modeling module includes a feature extraction unit and a matrix decomposition unit;

[0098] The feature extraction unit is used to extract key spectral features; the matrix decomposition unit is used to generate a basis matrix by non-negative matrix decomposition.

[0099] The threshold modeling module includes an intact threshold unit, a disturbance threshold unit and a regression modeling unit;

[0100] The intact threshold unit is used to calculate the decision boundary of the intact film sample; the disturbance threshold unit is used to determine the maximum deviation threshold of the disturbance sample; the regression modeling unit is used to construct the environmental coefficient offset mapping model;

[0101] The dynamic decision module includes a real-time projection unit, a reference adjustment unit and a state determination unit;

[0102] The real-time projection unit is used to collect new thin film spectrum data and solve the coefficient vector of the real-time spectrum; the reference adjustment unit is used to dynamically correct the judgment threshold according to the real-time environment; and the state judgment unit is used to make hierarchical decisions on the door lock state.

[0103] In this embodiment, an office building uses coated glass as an access control identification medium. The glass surface is coated with a specific metal oxide film (for example, indium tin oxide), and its spectral characteristics can reflect the integrity of the film.

[0104] When glass is subjected to violent damage (such as scratches or knocks), chemical contamination (such as corrosive liquids), or environmental changes (such as sudden temperature fluctuations causing film expansion and contraction), its spectral characteristics will change regularly. The system must identify the glass status in real time and implement hierarchical management of "normal unlocking, disturbance warning, and illegal alarm."

[0105] Implementation steps and technical realization:

[0106] Data collection phase:

[0107] Intact samples: Spectral data of 100 pieces of standard glass from the same batch (P=100) were collected, covering the visible light to near-infrared band (400-1000nm). Each spectral data is a 200-dimensional vector (m=200) corresponding to a standard environment (temperature 25°C, humidity 50%RH, light intensity 300lux).

[0108] Perturbation samples: Simulate three typical perturbations (Q=3): Scratch perturbation: Use tools of different hardness to make 50 samples (N1=50);

[0109] Chemical contamination: Apply acidic solution to generate 30 samples (N2=30);

[0110] Thermal stress damage: 20 samples were produced by high temperature baking (N3=20).

[0111] Environmental parameters: Synchronously record the environmental data collected each time, including temperature, humidity, and light intensity (M=3).

[0112] Spectral preprocessing:

[0113] Denoising and smoothing: 5-point sliding window filtering is used to remove random noise (such as instrument fluctuations) in the spectrum;

[0114] Baseline correction: fitting the spectral baseline with a cubic polynomial to deduct baseline drift (such as background absorption of the glass substrate);

[0115] Derivative operation: Calculate the first-order derivative to enhance the recognition of characteristic peaks (for example, the slope change of the characteristic absorption peak of indium tin oxide at 800nm);

[0116] Normalization: The spectrum is divided into four bands (400-550nm, 550-700nm, 700-850nm, and 850-1000nm). Z-score normalization is performed on each band to eliminate the influence of amplitude differences between different bands.

[0117] Feature extraction and matrix decomposition: Key features: Extract 8-dimensional features (n=8) including peak position (800nm ​​absorption peak position), peak intensity (peak intensity), derivative peak (maximum value of first-order derivative), and band energy mean (average value of spectral values ​​in each band) to construct feature vectors;

[0118] Non-negative matrix factorization (NMF): The feature vectors of 100 intact samples and 100 perturbed samples are combined into an 8×200 matrix, which is decomposed into an 8×3 basis matrix (k=3), corresponding to the three basic spectral modes and a 3×200 coefficient matrix. The coefficient vector of each sample represents the combination ratio of the three basis modes (for example, a scratch perturbation may be mainly dominated by the "structural defect basis").

[0119] Threshold calculation and regression modeling:

[0120] Intact threshold: Calculate the centroid and covariance matrix of the coefficient vectors of 100 intact samples, and set the normal fluctuation range based on the maximum eigenvalue (for example, allow the coefficient vector to fluctuate within a certain range around the centroid);

[0121] Perturbation threshold: Calculate the maximum distance between the coefficient vectors of all perturbed samples and the intact centroid to determine the "abnormal" boundary;

[0122] Environmental regression model: Establish a linear regression model between environmental parameters (such as temperature, humidity, and light) and coefficient shifts. For example, if it is found that for every 10°C increase in temperature, a certain base coefficient increases by an average of 5%, the spectral feature shift under different environments can be predicted based on this.

[0123] Real-time identification and dynamic adjustment:

[0124] Data collection: When the user swipes the card to trigger recognition, the door lock's built-in spectral sensor (for example, a micro-fiber spectrometer) collects the glass spectrum in real time and obtains the current environmental parameters through the temperature and humidity sensor.

[0125] Preprocessing and projection: De-noising and baseline correction are performed on the real-time spectrum. After extracting the 8-dimensional eigenvector, it is projected onto the basis matrix to solve the current coefficient vector (i.e., the combination ratio of the spectrum composed of the three basis matrixes).

[0126] Dynamic adjustment: Based on the current environmental parameters, the regression model predicts the coefficient offset, updates the judgment benchmark (for example, when the temperature rises, the center of mass adjusts in the direction of the predicted offset), and scales the threshold range according to the amplitude of environmental fluctuations (for example, a larger normal fluctuation range is allowed in a high temperature environment).

[0127] Hierarchical decision making:

[0128] If the distance between the coefficient vector and the adjusted centroid is within the intact threshold, the glass is judged to be normal and unlocking is allowed;

[0129] If the distance exceeds the intact threshold but is within the disturbance threshold: it is determined to be disturbed (such as a minor scratch), triggering a background warning (prompting property management to check), but temporary unlocking is allowed;

[0130] If the distance exceeds the disturbance threshold: it is judged as illegal damage (such as serious cracks), unlocking is refused and an audible and visual alarm is triggered.

[0131] 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 invention can be embodied in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered in all respects as illustrative and non-restrictive, and the scope of the invention is defined by the appended claims, not the foregoing description, and all variations within the meaning and range of equivalents of the claims are intended to be included therein. Any reference sign in a claim should not be construed as limiting the claim to which it relates.

Claims

1. A thin film spectral door lock management method based on data analysis, characterized by: The method comprises the following steps: Step 1: Collect the spectral data of intact thin films, disturbed spectral samples and corresponding environmental parameters of the same batch; Step 2: Denoising, smoothing, baseline correction, derivative calculation and normalization are performed on the spectral data; Step 3: Extract the key features of the spectrum to construct a vector, merge it, and use non-negative matrix decomposition to form a basis and coefficient matrix; Step 4: Calculate the intact and disturbed thresholds respectively, and build a regression model using environmental parameters and coefficient offsets; Step 5: Collect real-time spectrum and environmental parameters, project and solve coefficient vectors after pre-processing, dynamically adjust the judgment criteria and make decisions; In step 4, calculate the intact threshold R0: the coefficient vector {h j 0 } j=1 P , calculate the centroid μ0 and covariance matrix Σ0, and define the intact threshold: R0=α·(λ max (Σ0)) 1 / 2 ; Where P represents the number of spectral data of intact films in the same batch under the set standard environment; α represents the empirical constant, λ max (Σ0) represents the maximum eigenvalue of the covariance matrix Σ0; Calculate the disturbance threshold R1: Calculate the maximum distance between all disturbance sample coefficient vectors and μ0: R1=max j=P+1 N ||h j -μ0||2; where N=P+Σ q=1 Q N q ; Q represents the number of perturbation categories; q∈{1,2,…,Q}, N q Indicates the number of samples corresponding to the qth type of disturbance; Using the environmental parameter set E and the corresponding coefficient offset Δh=h-μ0, a regression model is constructed: Δh=f(E); Wherein, f represents the mapping function; In step 5, when the door lock triggers the identification process, the spectrum data S of the current film is collected. new , synchronously obtain real-time environmental parameters E new ; For the collected spectral data S of the thin film new , execute the preprocessing process to obtain the feature vector D new ; The preprocessed feature vector D new Projected onto the basis matrix W, set the optimization problem: h new =arg min h ∥D new −W⋅h∥2 2 , solve to get the real-time coefficient vector h new ; According to the real-time environmental parameter E new , prediction coefficient offset: Δh'=f(E new ), and dynamically adjust the judgment benchmark: μ new =μ0+Δh'; At the same time, the intact threshold and the disturbance threshold are adjusted respectively: R0′=R0⋅(1+β⋅∥E new −E0∥), R1′=R1⋅(1+γ⋅∥E new −E0∥); Among them, E0 is the standard environmental parameter, β and γ are environmental sensitivity coefficients; Calculate the adjusted distance: d = ∥h new −μ new ∥2; Determine the status based on the distance: When d ≤ R0′, it is determined as a normal film and unlocking is allowed; when R0′ < d ≤ R1′, it is determined as a perturbed film, triggering a warning but allowing unlocking; when d > R1′, it is determined as an illegal film, rejecting unlocking and triggering an alarm.

2. The thin film spectral door lock management method based on data analysis according to claim 1 is characterized in that: In step 1, the spectral data of P intact films of the same batch are collected under a set standard environment, which is expressed as S0: S0={S0 1 ,S0 2 ,…,S0 P }; Among them, S0 1 ,S0 2 ,…,S0 P Represent the 1st, 2nd, ..., Pth spectral data respectively, and each spectral data is an m-dimensional vector; For Q types of disturbances, N data are collected for each type of disturbance. q samples, denoted as C q :C q ={S q 1 ,S q 2 ,…,S q Nq }; Among them, S q 1 ,S q 2 ,…,S q Nq They represent the 1st, 2nd, ..., Nth disturbances corresponding to the qth type of disturbance. q Spectral data; Synchronously record the M-dimensional environment vector at each acquisition, expressed as E: E={E1,E2,…,E M }; where E1, E2, …, E M Represent the 1st, 2nd,…, M-dimensional environmental data respectively.

3. The thin film spectral door lock management method based on data analysis according to claim 2 is characterized in that: In step 2, the intact spectral data and the disturbed spectral data are normalized: Including denoising and smoothing: using sliding window filtering, convolution operation is performed on the intact spectrum and the disturbed spectrum data, and the smoothed spectrum is output; It also includes baseline correction: for the smoothed spectrum, the spectrum baseline is approximated by polynomial fitting, and the baseline is subtracted to obtain the baseline-corrected spectrum; It also includes derivative operations: for the baseline-corrected spectrum, the finite difference method is used to calculate the first-order or second-order derivative, and the derivative spectrum is obtained by selecting the first-order or second-order derivative; Normalization is also included: for derivative spectra, the spectral range is divided into different bands, and the spectral values ​​in each band are normalized by Z-score.

4. The thin film spectral door lock management method based on data analysis according to claim 3 is characterized in that: In step 3, key features are extracted from the preprocessed spectrum, including peak position, peak intensity, derivative features and statistical features, and the continuous spectrum curve is converted into a finite-dimensional feature vector D: D = [D1, D2, ..., D n ] T ; Where n represents the number of key features; Matrix decomposition modeling: merge the eigenvectors of the intact spectrum and the perturbed spectrum into an n×N matrix X; Using the non-negative matrix factorization algorithm, it is decomposed into two parts: X≈W·H; W=[w1,w2,…,w k ], represents the n×k basis matrix; H represents a k×N coefficient matrix: each spectral sample corresponds to a set of coefficients, indicating the combination ratio of each basis.

5. A thin film spectral door lock management system based on data analysis, applied to the thin film spectral door lock management method based on data analysis according to any one of claims 1 to 4, characterized in that: The system includes a data acquisition module, a spectrum processing module, a feature modeling module, a threshold modeling module and a dynamic decision module; The data acquisition module is used to collect intact film spectral data, disturbed spectral samples and corresponding environmental parameters of the same batch; the spectral processing module is used to perform denoising and smoothing, baseline correction, derivative operation and normalization on the spectral data; the feature modeling module is used to extract key spectral features to construct vectors, and after merging, decompose them into basis and coefficient matrices using non-negative matrices; the threshold modeling module is used to calculate intact and disturbed thresholds respectively, and construct a regression model using environmental parameters and coefficient offsets; the dynamic decision module is used to collect real-time spectra and environmental parameters, project and solve the coefficient vector after preprocessing, dynamically adjust the judgment benchmark and make decisions.

6. The thin film spectral door lock management system based on data analysis according to claim 5 is characterized by: The data acquisition module includes an intact spectrum unit, a disturbed spectrum unit and an environmental parameter unit; The intact spectrum unit is used to collect standard spectrum data of intact films of the same batch; the disturbed spectrum unit is used to collect spectrum samples under various disturbance states; and the environmental parameter unit is used to synchronously record environmental parameters.

7. The thin film spectral door lock management system based on data analysis according to claim 6 is characterized in that: The spectrum processing module includes a denoising and smoothing unit, a baseline correction unit, a derivative processing unit and a normalization unit; The denoising and smoothing unit is used for sliding window convolution to achieve spectral noise reduction; the baseline correction unit is used for polynomial fitting to eliminate baseline drift; the derivative processing unit is used for calculating derivative spectrum by finite difference method; and the normalization unit is used for normalizing spectral data.

8. The thin film spectral door lock management system based on data analysis according to claim 7 is characterized in that: The feature modeling module includes a feature extraction unit and a matrix decomposition unit; The feature extraction unit is used to extract key features of the spectrum; The matrix decomposition unit is used for generating a basis matrix by non-negative matrix decomposition.

9. The thin film spectral door lock management system based on data analysis according to claim 8 is characterized in that: The threshold modeling module includes an intact threshold unit, a disturbance threshold unit and a regression modeling unit; The intact threshold unit is used to calculate the decision boundary of the intact film sample; the disturbance threshold unit is used to determine the maximum deviation threshold of the disturbance sample; the regression modeling unit is used to construct the environmental coefficient offset mapping model; The dynamic decision module includes a real-time projection unit, a reference adjustment unit and a state determination unit; The real-time projection unit is used to collect new thin film spectrum data and solve the coefficient vector of the real-time spectrum; the reference adjustment unit is used to dynamically modify the judgment threshold according to the real-time environment; The state determination unit is used to determine the door lock state in a hierarchical manner.

Images