Multispectral data fusion intelligent monitoring method in reflective film production process

Abstract

The invention discloses a multi-spectral data fusion intelligent monitoring method in a reflective film production process, and relates to the technical field of reflective film production process monitoring, and the method specifically comprises the following steps: screening out all spectral band combinations with cross interference among different spectral bands based on preprocessed spectral data; calibration is carried out as a cross-band combination; establishing a wave band interference modeling matrix for each pair of spectral wave bands in the cross wave band combination, wherein the wave band interference modeling matrix is used for describing interference behavior characteristics between the spectral wave band pairs; determining the cross interference intensity of each cross band combination based on the established band interference modeling matrix, and classifying each cross band combination; and according to the classification result of each cross-band combination, respectively executing corresponding processing operations on different types of cross-band combinations. According to the invention, the problem of multispectral cross interference identification is solved, and interference quantitative evaluation and intelligent grading processing are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of reflective film production process monitoring, specifically to a multispectral data fusion intelligent monitoring method in the production process of reflective film. BACKGROUND

[0002] Reflective film is a film material used for reflecting light, widely used in optoelectronics, construction, automotive and other industries. Its main function is to improve energy efficiency or protect equipment by reflecting light or heat. The production process of reflective film usually involves multiple links, including film material synthesis, coating, curing, detection, etc. Each link needs to be accurately controlled to ensure the quality and performance of the film layer. In the production process of reflective film, multispectral data fusion technology plays an important role. Multispectral data fusion refers to the collection of images or data under different spectral bands (such as infrared, visible light, ultraviolet, etc.) and their combination for comprehensive analysis. This technology can provide more comprehensive physical information than single-band, allowing each stage of the production process to be more accurately monitored and controlled. In the production process of reflective film, through multispectral data fusion, the thickness, uniformity, reflectivity and other key indicators of the film layer can be monitored in real time, and problems or deviations in production can be found in time. Intelligent monitoring method uses artificial intelligence, machine learning and other technologies to automatically analyze and process multispectral data fusion information, thereby realizing real-time monitoring and optimization of the production process. This intelligent monitoring system can automatically identify potential risks in the production process, adjust production parameters, and even predict future production trends, thereby improving production efficiency, reducing waste, and ensuring the quality stability of reflective film. Through this intelligent monitoring method, the production process of reflective film not only gets more accurate control, but also reaches the level of automation and intelligence management.

[0003] The existing multispectral data fusion intelligent monitoring technology in the production process of reflective films achieves real-time monitoring and optimization of the production process through multiple links. First, in each key link of production, multiple sensors collect data in different spectral bands, such as visible light, infrared light, and ultraviolet light. These sensors are distributed at different positions on the production line, capturing information such as the optical properties and temperature distribution of the reflective film in real time. Then, the multispectral data collected is processed through data fusion technology, combining images or data from different bands to obtain more comprehensive film layer state information. These fused data are transmitted to an intelligent analysis system for real-time analysis using artificial intelligence algorithms to detect key quality parameters such as film layer uniformity, thickness, reflectivity, and surface defects. Once the system detects deviations or abnormalities in the production process, the intelligent monitoring system can automatically adjust relevant production parameters such as temperature, coating speed, or material ratio to ensure that the film layer quality meets the standards. In addition, the system can also perform predictive maintenance to identify potential equipment failures or production bottlenecks in advance, reducing downtime and ensuring efficient operation of the production line. Through multispectral data fusion and intelligent analysis, the production process of reflective films is more automated and precise, improving product consistency and production efficiency, and reducing human intervention and errors.

[0004] The existing technology has the following shortcomings: In the production process of reflective films, data from multiple spectral bands are collected simultaneously and used for multispectral data fusion to evaluate the quality of the film layer. However, there may be cross-interference between certain spectral bands, for example, the high absorption characteristics of the infrared band may affect the reflectivity measurement of the visible light band. Due to the mutual interference of different bands during sensor acquisition, especially when there are complex optical phenomena on the film layer surface (such as reflection, scattering, and absorption), this interference may be amplified. In practical applications, the existing multispectral data fusion intelligent monitoring technology in the production process of reflective films does not fully consider the mutual influence between bands, and the existing fusion algorithm usually assumes that the data of each band is independent, without designing corresponding modeling and correction methods for the cross-interference between bands. As a result, the interference between bands cannot be effectively eliminated or adjusted during the fusion process, resulting in errors in the final fused data, which affects the evaluation and monitoring of the film layer quality. When the fusion result deviates, the system may misjudge the optical performance of the film layer, leading to incorrect evaluation of important parameters such as reflectivity and transmittance of the reflective film. Such errors are not identified in time, which may allow unqualified film layers to pass through the production link, increasing the scrap rate and reducing product quality consistency, ultimately leading to increased production costs, increased product returns, and even affecting the satisfaction and trust of end customers.

[0005] The above information disclosed in the BACKGROUND section merely to enhance the understanding of the background of the present disclosure, and therefore it can include information that does not constitute prior art known to those of ordinary skill in the art. SUMMARY

[0006] The object of the present application is to provide a multispectral data fusion intelligent monitoring method in the production process of reflective film to solve the problems in the background art.

[0007] In order to achieve the above object, the present application provides the following technical solution: a multispectral data fusion intelligent monitoring method in the production process of reflective film, specifically comprising the following steps: Real-time acquisition of spectral data of the film layer to be detected in the production process of reflective film at different spectral bands by a multi-channel spectral sensor, and pre-processing thereof; Based on the pre-processed spectral data, all spectral band combinations with cross interference between different spectral bands are screened out and labeled as cross band combinations; A band interference modeling matrix is established for each pair of spectral bands in the cross band combination, which is used to describe the interference behavior characteristics between the spectral band pairs; Based on the established band interference modeling matrix, the cross interference intensity of each cross band combination is determined, and each cross band combination is classified; According to the classification results of each cross band combination, corresponding processing operations are performed on different types of cross band combinations; Based on the historical fusion processing data and the film layer detection results, the determination method of cross interference intensity, the classification strategy of cross band combination and the processing operation are optimized, and the classification results and processing records are visualized and historically archived.

[0008] Preferably, based on the pre-processed spectral data, all spectral band combinations with cross interference between different spectral bands are screened out and labeled as cross band combinations, specifically: by performing statistical analysis operation on the pre-processed spectral data corresponding to each pair of spectral bands, calculating the change trend correlation index and signal interference coupling characteristic index between each pair of spectral bands, and according to the result whether any of the change trend correlation index and the signal interference coupling characteristic index exceeds the pre-set interference judgment threshold, identifying all spectral band combinations with cross interference and labeling them as cross band combinations.

[0009] Preferably, the band interference modeling matrix is established for each pair of spectral bands in the cross band combination, specifically comprising the following steps: constructing a time series matrix for representing the joint response characteristics of each pair of spectral bands in the calibrated cross-band combination based on the corresponding preprocessed spectral data within a preset time window; performing principal component analysis, covariance analysis and mutual information analysis operations on the time series matrix to extract a plurality of characteristic factors reflecting the interference characteristics between the pair of spectral bands; integrating and constructing the extracted characteristic factors into a band interference modeling matrix according to statistical behaviors, frequency coupling relationships and response change patterns, for describing the joint behavior characteristics of the pair of spectral bands in terms of spectral response intensity, change trend and interference coupling.

[0010] Preferably, based on the established band interference modeling matrix, the cross-interference intensity of each cross-band combination is determined, and each cross-band combination is classified, specifically including the following steps: extracting the cross-interference characteristic information of each cross-band combination from the established band interference modeling matrix, and preprocessing after extraction; extracting energy distribution information and response disturbance information from the preprocessed cross-interference characteristic information of each cross-band combination, and analyzing after extraction to generate the coupling concentration coefficient and the response disturbance index of each cross-band combination, respectively; establishing a weighted summation model for the generated coupling concentration coefficient and response disturbance index of each cross-band combination, and generating the cross-interference index of each cross-band combination through weighted summation; determining a pre-set cross-interference index threshold interval, and comparing with the generated cross-interference index of each cross-band combination after determination, evaluating the cross-interference intensity of each cross-band combination according to the comparison result, and dividing each cross-band combination into an effective fusion group, a limited fusion group and a rejection fusion group according to the evaluation result.

[0011] Preferably, the acquisition logic of the coupling concentration coefficient of each cross-band combination is as follows: extracting the energy distribution information from the preprocessed cross-interference characteristic information of each cross-band combination, specifically including the first principal component contribution rate, the frequency energy concentration and the information redundancy factor of each cross-band combination in the established band interference modeling matrix, and labeling them as , and , respectively, wherein represents the first principal component contribution rate of the th cross-band combination in the established band interference modeling matrix, Indicates the first The information redundancy factor of the cross-band combination, , is a positive integer; Calculate the coupling concentration coefficient for each cross-band combination The specific calculation logic is: multiply the first principal component contribution rate of each cross-band combination by the frequency energy concentration and square it, add one to the result and take the natural logarithm, then subtract the square root of the corresponding information redundancy factor. The resulting value is the coupling concentration coefficient of each cross-band combination.

[0012] Preferably, the logic for obtaining the response disturbance index of each cross-band combination is as follows: The response disturbance information is extracted from the cross-interference characteristic information of each cross-band combination after preprocessing, specifically including the trend offset angle, nonlinear variation coefficient and response rate fluctuation factor of each cross-band combination in the established band interference modeling matrix, and calibrated as 、 and , Indicates the first The trend deviation angle of the cross-band combination, Indicates the first The nonlinear variation coefficient of the cross-band combination, Indicates the first The response rate fluctuation factor of the cross-band combination, , is a positive integer; Calculate the response perturbation index for each cross-band combination The specific calculation logic is: after taking the absolute value of the sine value of the trend offset angle of each cross-band combination, add one to the exponential value corresponding to the product of its nonlinear change coefficient and the response rate fluctuation factor, and then take the natural logarithm. The sum of the two is the response disturbance index of each cross-band combination.

[0013] Preferably, the coupling concentration coefficient of each cross-band combination generated is and response disturbance index Establish a weighted summation model and generate the cross-interference index of each cross-band combination through weighted summation .

[0014] Preferably, a predetermined cross-interference index threshold interval is determined , and after determination, the cross-interference index of each cross-band combination generated The comparison is performed, the cross interference intensity of each cross waveband combination is evaluated according to the comparison result, and each cross waveband combination is divided into an effective fusion group, a limited fusion group and a rejected fusion group according to the evaluation result, and the specific comparison analysis and division are as follows: If the cross interference intensity of the cross waveband combination is low intensity, the cross waveband combination is divided into the effective fusion group; If the cross interference intensity of the cross waveband combination is low intensity, the cross waveband combination is divided into the effective fusion group; If the cross interference intensity of the cross waveband combination is medium intensity, the cross waveband combination is divided into the limited fusion group; If the cross interference intensity of the cross waveband combination is medium intensity, the cross waveband combination is divided into the limited fusion group; If the cross interference intensity of the cross waveband combination is high intensity, the cross waveband combination is divided into the rejected fusion group. If the cross interference intensity of the cross waveband combination is high intensity, the cross waveband combination is divided into the rejected fusion group.

[0015] Preferably, according to the classification results of each cross waveband combination, corresponding processing operations are respectively performed on cross waveband combinations of different types, and the specific operations are as follows: For the cross waveband combination divided into the effective fusion group, the processing operation performed is specifically: directly retaining the corresponding spectral data as the fusion input to participate in the subsequent spectral fusion calculation; For the cross waveband combination divided into the limited fusion group, the processing operation performed is specifically: applying a fusion weight reduction operation to the spectral data in the fusion calculation, and increasing a dynamic verification mechanism to limit the influence on the fusion result; For the cross waveband combination divided into the rejected fusion group, the processing operation performed is specifically: completely rejecting the corresponding spectral data and not including it in the fusion model calculation range.

[0016] In the above technical solution, the technical effects and advantages provided by the present application are as follows: 1. The present application realizes accurate identification and structured quantization of the interference behavior between wavebands in the process of monitoring the reflective film by constructing a cross interference modeling mechanism based on multi-spectral data. The technical solution extracts feature factors from multiple dimensions such as statistical behavior, frequency domain coupling characteristics and response trend for each pair of waveband combinations with cross interference, and further calculates the coupling concentration coefficient and the response disturbance index, effectively solving the interference distortion problem caused by regarding each waveband data as independent in the traditional fusion algorithm. Compared with the existing algorithm which cannot identify implicit interference, the present application can locate potential high-risk waveband combinations before data fusion, providing strong data support for subsequent interference control.

[0017] 2、The application realizes continuous quantitative evaluation and hierarchical classification management of the cross interference intensity by constructing a cross interference index and setting multiple threshold intervals, divides all waveband combinations into effective fusion groups, restricted fusion groups and excluded fusion groups, and implements differentiated fusion processing strategies according to different levels. This scheme avoids the extensive data processing logic of 'all or nothing' in traditional fusion systems, especially for boundary interference combinations, provides a fusion weight weakening and dynamic verification mechanism to ensure the stability and reliability of the fusion information. Through this intelligent control based on intensity evaluation, the accuracy and consistency of the film optical performance evaluation are effectively improved.

[0018] 3、The application further introduces an adaptive optimization mechanism based on historical fusion data and film layer detection results, establishes a feedback channel between fusion quality and detection error. By continuously archiving each round of fusion classification and processing records and comparing them with the actual detection effect, the system can dynamically optimize the interference index calculation parameters, classification boundaries and fusion strategies, thereby realizing intelligent adaptation to environmental changes, material differences and equipment performance fluctuations. At the same time, the system visualizes all processing processes and archives them historically, improving data traceability and strategy transparency, and providing an intuitive management interface for operators. Overall, this scheme has strong robustness, good scalability and real-time optimization capability, and can be long-term and stable applied in complex and dynamic reflective film production environments. BRIEF DESCRIPTION OF DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments described in the present application, and other drawings can also be obtained by those skilled in the art based on these drawings.

[0020] Figure 1 A flowchart of the multispectral data fusion intelligent monitoring method in the reflective film production process of the present application; Figure 2 A method mind map of the multispectral data fusion intelligent monitoring method in the reflective film production process of the present application. DETAILED DESCRIPTION

[0021] Example implementations will now be described more fully with reference to the accompanying drawings. Example implementations, however, can be implemented in many different ways and should not be construed as limited to the examples set forth herein; rather, these examples are provided so that this disclosure will be thorough and complete, and will fully convey the inventive concept to those skilled in the art.

[0022] The present application provides a method for multispectral data fusion intelligent monitoring in a reflective film production process, which comprises the following steps: Figure 1 and Figure 2The multispectral data fusion intelligent monitoring method in the reflective film production process shown specifically comprises the following steps: The multispectral data fusion intelligent monitoring method in the reflective film production process shown specifically comprises the following steps: In order to realize real-time acquisition of the spectral data of the to-be-detected film layer in different spectral bands during the reflective film production process, the system can adopt a multi-channel spectral sensor array and combine high-frequency data acquisition control logic to realize synchronous monitoring. The multi-channel spectral sensor is composed of multiple detection channels with different preset central wavelengths and wavelength ranges. Each channel receives a specific wavelength band of reflected light signals through a light splitting device and converts them into corresponding spectral intensity digital signals. By configuring the sampling frequency and channel reading order of the software controller, parallel acquisition of all wavelength band data at the same time is realized. At the software level, the sensor parameters can be set, the drive interface can be initialized, and the synchronous acquisition thread can be started to receive the original spectral data stream of each channel in real time and store it in the cache matrix with a unified timestamp, thereby ensuring the spatiotemporal alignment of the reflected signals of each wavelength band and avoiding cross-frame interference caused by delayed sampling. In addition, the system can also set a linkage mechanism between the film layer moving speed and the sampling frequency to ensure that the acquired data is representative and covers the entire production area.

[0023] Pretreatment of the acquired spectral data is a key step to ensure the accuracy of subsequent interference identification, modeling and fusion judgment. The reason is that the original data may be affected by factors such as device noise, environmental light changes, and acquisition angle deviation, resulting in problems such as intensity drift, signal-to-noise ratio drop, and time misalignment. If not handled, it will seriously interfere with the input basis of the fusion algorithm. In software implementation, the pretreatment includes the following operations: first, time alignment processing is performed to calibrate the timestamps of each channel data through software control, ensuring that different wavelength band data comes from the same reflection time; second, noise filtering operation is performed to remove high-frequency interference through low-pass filtering, wavelet denoising or sliding window averaging algorithm; third, normalization processing is performed to standardize the intensity values of each wavelength band to a unified numerical range for subsequent comparison and weighted calculation; finally, spatial registration is implemented, i.e. adjusting the spatial correspondence of each channel data according to the relative position of the film layer in the sensor array. These pretreatment processes are automatically executed in software through data processing functions or pipeline processing framework (such as data stream buffer + filtering module), ensuring that the multispectral data has uniformity, comparability and high-quality structured expression, laying a solid data foundation for subsequent cross-interference modeling and fusion calculation.

[0024] Based on the pretreated spectral data, all spectral band combinations with cross-interference between different spectral bands are screened out and labeled as cross-band combinations; In this embodiment, all spectral band combinations with cross interference between different spectral bands are screened out based on the pre-processed spectral data, and are labeled as cross band combinations. Specifically, by performing statistical analysis on the pre-processed spectral data corresponding to each pair of spectral bands, the change trend correlation index and the signal interference coupling characteristic index between each pair of spectral bands are calculated, and according to the result of whether any of the change trend correlation index and the signal interference coupling characteristic index exceeds the pre-set interference judgment threshold, all spectral band combinations with cross interference are identified and labeled as cross band combinations.

[0025] In order to effectively identify the cross interference relationship that may exist between different spectral bands, the system performs pairwise analysis on all spectral band combinations based on the pre-processed spectral data. At the software level, a cross combination iteration matrix can be constructed, each band combination is taken as an analysis unit, and a statistical analysis function is called to process the spectral data of the corresponding band. First, the spectral response sequence corresponding to each pair of bands is processed by a sliding time window to capture the spectral change trend in a local time range, and the differential characteristics such as change slope, fluctuation amplitude and response rate are calculated. Then, the system further calculates the dynamic correlation index (such as Pearson coefficient, Spearman rank correlation coefficient or mutual information entropy, etc.) of the trend sequence of the two bands, so as to judge whether there is trend synchronization or behavior consistency between the two bands. At the same time, the system also extracts the signal coupling characteristics of the two bands under high frequency or specific interference mode, such as through cross power spectral density (CPSD) analysis, frequency domain coherence detection or wavelet resonance spectrum coefficient calculation, to measure whether the interference of one band signal produces resonance or offset response in another band signal. These statistical indexes can be completed through embedded signal processing library or customized time series analysis module, and a set of structured evaluation results will be generated for each band pair in the execution process, which will be used as the basis for subsequent judgment of whether it constitutes “cross interference”. The necessity of this analysis method lies in that it can reveal the potential interference mode through the data itself without additional hardware intervention, and provide theoretical support for cross band identification and subsequent processing.

[0026] The "trend correlation index" refers to an index for measuring the consistency of the spectral response values of two spectral bands in a unit time window or a continuous observation period in terms of their trend, and the core purpose is to determine whether two bands exhibit similar rising, falling, oscillating or sudden change characteristics in the same process stage. Common implementation methods include calculating the correlation (such as the Pearson correlation coefficient) of the first derivative sequence of the spectral response curves of the two bands in a sliding window, or using a nonlinear sequence similarity algorithm (such as dynamic time warping, DTW) to analyze the trend alignment degree. The "signal interference coupling feature index" is mainly used to evaluate the energy coupling or interference transmission relationship between two bands in the time-frequency domain, and the purpose is to determine whether the fluctuations of one band affect the response behavior of another band in terms of certain frequency, amplitude or phase characteristics. It can be specifically implemented by spectral overlap rate analysis, cross wavelet transform (XWT) coefficient extraction or signal covariance offset rate calculation. These two types of indexes evaluate the interaction between bands from the "macro trend" and "micro interference behavior" dimensions respectively, and have complementary and discriminative properties, which are important quantitative means for identifying real cross interference phenomena.

[0027] In order to identify spectral band combinations with cross interference based on the trend correlation index and the signal interference coupling feature index, the system needs to set the corresponding interference judgment threshold through software after completing the index extraction of each band combination, and perform automatic comparison and classification recognition process. Specifically, the system first loads the threshold interval corresponding to each type of index in the database or configuration file. The threshold can be obtained by historical sample training or set by manual experience. For example, the threshold of the correlation index can be set to 0.8 (high positive correlation) or -0.8 (high negative correlation), and the threshold of the coupling feature index can be set to a certain percentage of power spectrum overlap or resonance band amplitude limit. After the system completes the index calculation, the index values of the current spectral band combination are compared with the corresponding threshold one by one using a decision function. Once a certain index exceeds the set threshold, the "interference established" label of the combination is triggered through Boolean logic. Subsequently, the system automatically writes the spectral band combination into the calibration list and marks it with the "cross band combination" identifier in the memory structure. In order to ensure the efficiency and accuracy of the entire recognition process, the system can use vectorized operations, threshold table matching mechanisms and result cache indexing to ensure good real-time performance in large-scale multi-band combination analysis scenarios. The core significance of this process is that through quantitative evaluation and threshold judgment of key interference features, the system can accurately select combinations with real coupling interference from a large number of normal band relationships, avoid redundant modeling and misjudgment, and ensure that subsequent interference modeling and fusion calculation focus on the most critical interference risk points, thereby improving the accuracy and response efficiency of the monitoring system.

[0028] a band interference modeling matrix is established for each pair of spectral bands in the cross-band combination, and the band interference modeling matrix is used to describe interference behavior characteristics between the pair of spectral bands; In this embodiment, a band interference modeling matrix is established for each pair of spectral bands in the cross-band combination, specifically including the following steps: Based on the corresponding pre-processed spectral data of each pair of spectral bands in the calibrated cross-band combination within a preset time window, a time series matrix is constructed for representing joint response characteristics of the pair of spectral bands; In order to construct a time series matrix for representing joint response characteristics of each pair of spectral bands in the cross-band combination, the system needs to organize and process spectral data in a time sequence driven manner at the software level. Specifically, after identifying the cross-band combination, the system calls the pre-processed spectral data in the cache or database, extracts the spectral response sequence of each pair of spectral bands within the time period according to the set time window size (such as 5 seconds, 10 frames or any logical sampling section). For each pair of spectral bands, the system constructs the corresponding spectral reflectance or absorption data into two independent vectors, and then aligns them by timestamp to combine them into a two-dimensional matrix in row direction structure, where each row corresponds to a time sampling point and each column corresponds to the response value of the two bands at the time point. This time series matrix not only retains the time evolution information, but also explicitly reflects the synchronous fluctuation behavior of the two bands in the time sequence dimension. The necessity of this operation is that in the actual production environment, the spectral response is not static, but changes subtly with the change of film thickness, the enhancement of interference effect or the accumulation of materials, and this "joint response trend" is the key basis for subsequent judgment of the interference behavior between bands. By constructing the time series matrix, the dynamic interaction between bands can be converted into a structured input form for subsequent feature extraction and statistical modeling algorithms, so as to realize accurate quantification and modeling of cross interference characteristics.

[0029] Performing principal component analysis, covariance analysis and mutual information analysis operations on the time series matrix to extract a plurality of feature factors for reflecting interference characteristics between the pair of spectral bands; To extract multiple characteristic factors from the constructed time series matrix that can represent the interference characteristics between spectral bands, the system needs to perform statistical modeling operations such as principal component analysis (PCA), covariance analysis, and mutual information analysis on the matrix at the software level. In specific implementation, the system first calls the matrix processing module to standardize the time series matrix, ensuring that the response values of the two bands are within a comparable range. Subsequently, the system applies the three types of analysis methods in turn: using PCA to extract the principal component vectors in the matrix, identifying the dominant contribution direction of the response change; using covariance analysis to calculate the joint fluctuation intensity of the response values of the two bands on the time axis; using mutual information analysis to evaluate the information sharing degree and nonlinear coupling relationship between the two bands. The software system extracts characteristic indicators such as "first principal component proportion", "covariance coefficient value", and "mutual information entropy" by constructing statistical models, and integrates them into a characteristic factor vector, which serves as the basis for subsequent interference strength judgment, band classification, and fusion weight allocation. The significance of this composite analysis method lies in its comprehensive modeling of the interaction between bands from multiple mathematical dimensions, considering both linear synchronicity and nonlinear interference behavior, thereby achieving accurate characterization and data layer modeling of band coupling phenomena in complex spectral environments.

[0030] Principal component analysis (PCA) is a dimensionality reduction technique based on eigenvalue decomposition, used to identify the main change direction in the time series matrix and its contribution to the overall response. In this context, PCA can help extract the "joint main dynamics" between the two bands and determine whether there is a high degree of synchronization or driving response. Covariance analysis is used to measure the joint shift degree of the response values of the two bands at each time, and if the covariance coefficient is high, it indicates that the response fluctuations of the two bands are consistent or coupled, which is usually the result of physical interference or optical coupling. Mutual information analysis is a non-parametric information theory method used to measure the shared information between two random variables, which can discover potential associations even in the absence of significant linear relationships. In the software system, PCA provides main direction modeling, covariance provides fluctuation intensity judgment, and mutual information supplements the detection of nonlinear interaction, forming a multi-dimensional feature extraction system that makes interference modeling not limited to a certain mathematical dimension, but forms a structured, quantifiable, and comparable interference characteristic factor, providing high-reliability data support for subsequent classification, fusion, and control logic.

[0031] The extracted characteristic factors are integrated and constructed into a band interference modeling matrix according to statistical behavior, frequency coupling relationship, and response change pattern, which describes the joint behavior characteristics of the two spectral bands in terms of spectral response intensity, change trend, and interference coupling.

[0032] In order to convert the extracted multiple interference characteristic factors into a structured and computable waveband interference modeling matrix, the system needs to classify and organize the characteristic factors, standardize the processing and structure mapping at the software level. Specifically, the system logically divides the characteristic factors into three categories based on their sources: factors based on statistical analysis (such as principal component proportion, covariance value), factors based on frequency domain information (such as frequency resonance index, spectral overlap degree), and factors based on time response behavior (such as response delay time, mutation slope, etc.), and are classified into three dimensions of statistical behavior, frequency coupling relationship and response change mode. On this basis, the software defines the dimension, weight influence interval and numerical mapping position of each factor by constructing a three-dimensional matrix template, fills these factors into the corresponding positions of the modeling matrix according to the set order, and retains auxiliary attributes such as original value, normalized value and deviation coefficient, to form a structured input matrix for calculating the interference degree. The purpose of constructing such a matrix is to uniformly represent the multi-dimensional and multi-source waveband interference information, so that the subsequent decision logic can be based on a consistent data structure for fusion scoring, classification grouping or control feedback, while providing a standardized input form for the system to introduce a deep learning model or optimization strategy module, thereby realizing digital modeling and systematic management of interference characteristics.

[0033] "Statistical behavior" refers to the mathematical statistical relationship between two spectral wavebands in the original or normalized spectral values, mainly used to measure the overall fluctuation consistency and linear co-movement, such as the first principal component contribution, average covariance, variance ratio, etc. This type of factor usually represents the basic common fluctuation characteristics and is a stable basis in modeling. "Frequency coupling relationship" reflects whether there is energy coupling, resonance transmission or amplitude overlap between two wavebands in the frequency domain, which can be quantified by cross-power spectral density function, resonance frequency band energy ratio or frequency phase synchronization rate, etc. This type of factor is often used to reveal non-intuitive coupling mechanisms between cross-wavebands caused by material structure, interlayer interference, etc., and is the core feature of modeling to judge the interference nature. "Response change mode" focuses on describing the response difference, mutation behavior and trend deviation of two wavebands in time series, including but not limited to response fluctuation direction consistency, mutation point lag, change trend turning amplitude, etc., mainly used to identify short-time coupling, phase shift or dynamic mutation, etc. The synergistic integration of the three types of factors makes the modeling matrix not only reflect the interference strength, but also present the multi-dimensional characteristics of interference behavior, providing the system with more rich and real judgment basis.

[0034] To further clarify the data source, structure and filling logic of the waveband interference modeling matrix, the following detailed supplementary explanation is made: The waveband interference modeling matrix is established for each pair of spectral wavebands calibrated as a cross-waveband combination, based on the preprocessed spectral data collected within a preset time window. The data source is a multi-channel spectral sensor array arranged at each detection position of the reflective film production line, which collects the reflectance spectral intensity of the film layer at the specified waveband in real time, and obtains the structured spectral data sequence after normalization, denoising, time synchronization and other preprocessing. Taking each pair of cross-spectral wavebands as a unit, a two-dimensional time sequence matrix is constructed, with the rows representing the sampling time points and the columns representing the response values of the two wavebands. Based on the time sequence matrix, principal component analysis (PCA), covariance calculation and mutual information analysis are sequentially performed to extract characteristic factors that can reflect the interference characteristics between the waveband pairs, including but not limited to the first principal component contribution rate, covariance coefficient, mutual information entropy, frequency energy concentration degree, trend deviation angle, etc. After classification and integration of the above characteristic factors, they are mapped into a unified structure template according to three characteristic dimensions of "statistical behavior", "frequency coupling" and "response change pattern", and filled into the waveband interference modeling matrix of fixed dimension, such as the 9-dimensional structure containing principal component proportion, frequency energy focusing value, redundancy factor, trend deviation angle, nonlinear change coefficient, rate fluctuation value, etc. Each index is calculated based on the original data and has a clear statistical physical meaning. Each cell in the matrix is a standardized numerical value or model output index, with repeatable calculation logic and unified data generation rules, ensuring the objectivity and consistency of interference modeling. The matrix not only reflects the coupling degree of two wavebands in multiple dimensions, but also serves as the input basis for subsequent interference strength evaluation and fusion strategy judgment.

[0035] Based on the established waveband interference modeling matrix, the cross-interference strength of each cross-waveband combination is determined, and each cross-waveband combination is classified. In this embodiment, based on the established waveband interference modeling matrix, the cross-interference strength of each cross-waveband combination is determined, and each cross-waveband combination is classified, which specifically includes the following steps: The cross-interference characteristic information of each cross-waveband combination is extracted from the established waveband interference modeling matrix, and preprocessed after extraction. In the software implementation process, the system first traverses the established band interference modeling matrix, which is a multidimensional numerical matrix based on each pair of cross-spectral band combination, containing multidimensional interference characteristic original data such as statistical behavior, frequency coupling, response mode, etc. In order to extract the cross-interference characteristic information, the system extracts all numerical field sets corresponding to the combination according to the logic of “one cross-band combination corresponds to one sub-matrix” based on the structure definition of the matrix, through the matrix field label or index structure. The specific extraction method can use field pointer mapping and data aggregation operation to combine and encapsulate the characteristic dimensions in the interference modeling matrix of each pair of combination, such as principal component proportion, mutual information density, frequency domain amplitude index, response gradient curvature, trend offset angle, etc., as a multidimensional feature vector set, which is the standard representation of cross-interference characteristic information. In the software structure, the feature set is usually output in a structured data format (such as vector list or tensor matrix) for subsequent preprocessing and index calculation module calling. This process can be automatically completed through logical indexing, dimension slicing and attribute grouping, thereby ensuring the unified extraction and management efficiency of the system for large-scale band combination characteristic information.

[0036] The preprocessing of the extracted cross-interference characteristic information is a key step to ensure the accuracy of subsequent index calculation, model training and classification judgment. Since the original characteristic information may have problems such as inconsistent dimension scale, extreme value deviation, uneven distribution, noise disturbance, etc., if not preprocessed, it may lead to magnified calculation error, distorted model response or unstable parameter convergence. The preprocessing process usually includes the following operations: first, the system normalizes all feature vectors, such as mapping the feature values to the [0, 1] or [−1, 1] interval, to avoid a feature being dominated in the weighted model output due to large value; second, perform outlier detection and clipping, use statistical outlier detection (such as Z-score or IQR) method to remove extreme value interference in the feature; third, implement feature redundancy removal operation, such as calculating the correlation matrix between features and removing highly collinear items to reduce the dimension complexity of the subsequent model; finally, for feature items with periodic or nonlinear fluctuations, the system can perform nonlinear mapping or curve fitting smoothing processing to improve the stability and model interpretation ability of the overall feature. The whole preprocessing flow is composed of feature standardization module, noise filtering module and dimension management module in the software to realize automatic and unified standard preprocessing strategy.

[0037] From the cross-interference characteristic information of each cross-band combination after preprocessing, the energy distribution information and response disturbance information are extracted and analyzed to generate the coupling concentration coefficient and response disturbance index of each cross-band combination, respectively; After the preprocessing of the cross-interference characteristic information is completed, the system performs information dimension extraction operations on the feature vector set of each cross-band combination based on the feature dimension label and attribute weight strategy. Specifically, the system first classifies all energy distribution-related feature items (such as principal component contribution rate, frequency energy proportion, and information compression factor) into the energy distribution information extraction module according to the type label of the feature index in the modeling matrix; then, all time-domain disturbance behavior-related feature items (such as response trend offset angle, nonlinear fitting residual, and derivative fluctuation amplitude) are classified into the response disturbance information extraction module. The software maps each field in the original feature vector to the two types of information dimensions by setting a feature mapping template, and integrates multiple numerical items within the same type of dimension through an information fusion function, such as taking the weighted average value, information entropy combination value, or constructing a vector subspace, to generate two standard information vectors: the energy distribution information vector and the response disturbance information vector. The entire extraction process is completed on the matrix structure through algorithms such as logical grouping, index pairing, and numerical aggregation, forming a unified, standardized, and callable structured information object, which serves as the input basis for subsequent parameter calculation and interference judgment, ensuring that the analysis logic is rigorous, the data path is clear, and the system processing is automatically controllable.

[0038] A weighted summation model is established for the generated coupling concentration coefficients and response disturbance indices of each cross-band combination, and a cross-interference index is generated for each cross-band combination through weighted summation; A pre-set cross-interference index threshold interval is determined, and after determination, it is compared with the generated cross-interference indices of each cross-band combination. According to the comparison result, the cross-interference strength of each cross-band combination is evaluated, and according to the evaluation result, each cross-band combination is divided into an effective fusion group, a limited fusion group, and a rejected fusion group.

[0039] To reasonably set the threshold interval of the cross interference index, the system needs to combine historical fusion data and actual film layer quality detection results at the software level to determine the threshold interval through data modeling and distribution analysis. Specifically, the system first calls the cross interference index values corresponding to each cross waveband combination in multiple production cycles in the historical record database, and matches and labels them with the film layer quality grade tags (such as qualified, edge, and abnormal) reflected in actual detection. Then, the system statistically models these interference index values, uses density estimation methods (such as Gaussian kernel density estimation) to construct the probability distribution curve of the index values, and combines label distribution clustering analysis to divide the interference index values into intervals. In the division process, the system identifies the distribution inflection point, density variation point, or optimal segmentation threshold point of discrimination accuracy, automatically generates a recommended threshold interval, such as low interference, medium interference, and high interference. In addition, to improve the adaptability of the model, the system can also dynamically update the threshold according to new data, continuously optimize the discrimination boundary through sliding window strategy and incremental learning mechanism, so as to realize the intelligent and adaptive setting of the interference classification threshold. Finally, the determined threshold interval will be solidified as a callable configuration parameter to participate in the subsequent interference strength comparison and classification decision logic, ensuring the sustainable evolution ability and robust judgment effect of the system.

[0040] In this embodiment, the acquisition logic of the coupling concentration coefficient of each cross waveband combination is as follows: The energy distribution information is extracted from the cross interference feature information of each preprocessed cross waveband combination, specifically including the first principal component contribution rate, frequency energy concentration, and information redundancy factor of each cross waveband combination in the established waveband interference modeling matrix, and are respectively labeled as , and , represents the first principal component contribution rate of the th cross waveband combination in the established waveband interference modeling matrix, represents the frequency energy concentration of the th cross waveband combination in the established waveband interference modeling matrix, represents the information redundancy factor of the th cross waveband combination in the established waveband interference modeling matrix, , is a positive integer; In the established waveband interference modeling matrix, each cross-waveband combination corresponds to a structured matrix containing original characteristic data of its interference behavior in multiple dimensions such as statistics, frequency domain, and response variation. In order to extract the first principal component contribution rate of the combination, the system can perform principal component analysis (PCA) operation on the interference modeling matrix corresponding to the combination. Specifically, after the software centralizes the matrix, it calculates the covariance matrix and obtains the eigenvalues of the principal component direction through eigenvalue decomposition. The first principal component contribution rate, which is the proportion of the maximum value in the eigenvalues, indicates whether the interference characteristics are concentrated in a single dominant dimension in the modeling matrix. This operation is completely based on the statistical feature structure inside the modeling matrix, does not require the introduction of external data, and can automatically extract the ratio through the software algorithm library to reflect the compression and dominant trend degree of the interference in the structure distribution, which is a key basis for coupling concentration evaluation.

[0041] In the established waveband interference modeling matrix, the frequency response characteristics of each cross-waveband combination are generally stored in the frequency domain feature dimension in the form of frequency density distribution vector or spectral energy coefficient. Based on the frequency domain component, the system can analyze the energy concentration of the interference spectrum structure of each combination. In the specific operation, the software performs local integration and normalization calculation on the amplitude response vector of the combination in the frequency domain dimension, extracts the energy sum of the main peak frequency band, and performs ratio operation with the total energy of the full frequency band, to obtain the frequency energy concentration. The higher the ratio, the more the interference signal is concentrated in a few frequency regions, indicating that it has a spectral focusing characteristic. This process completely depends on the frequency domain structure inside the modeling matrix and does not involve original spectral signal processing, and can be directly operated on the matrix through frequency window sliding clustering, peak integration, etc. algorithm, which is suitable for efficiently identifying combinations with dominant frequency interference trend.

[0042] In the established waveband interference modeling matrix, each cross-waveband combination contains its information entropy characteristics and joint feature synergy items in different feature dimensions, and the system can extract the information redundancy factor of the combination based on this structure. The specific way is: the system obtains the information entropy estimates of the combination in each independent dimension (such as principal component dimension entropy, response mode entropy, etc.) and the fusion information entropy in the joint dimension from the modeling matrix, and then calculates the information redundancy degree using mutual information analysis logic. The information redundancy factor can be defined as the ratio of mutual information to joint entropy, that is, the proportion of information repetition in the combination. Since all information entropy indicators come from the archived statistical structure and information distribution layer in the modeling matrix, this calculation process does not need to backtrack the original data, but can be realized by performing information theory function mapping on the entropy structure field in the modeling matrix, which is a key quantitative indicator for measuring the repetition of waveband coupling information.

[0043] Calculate the coupling concentration coefficient of each cross-waveband combination The specific calculation logic is: multiply the first principal component contribution rate of each cross-band combination by the frequency energy concentration and square it, add one to the result, take the natural logarithm, and then subtract the square root of the corresponding information redundancy factor. The resulting value is the coupling concentration coefficient of each cross-band combination. The specific calculation formula is as follows: Where, For the The coupling concentration coefficient of the cross-band combination.

[0044] This formula is used to calculate the coupling concentration coefficient of each cross-band combination. Its core purpose is to comprehensively measure whether the interference between a pair of spectral bands presents a characteristic aggregation phenomenon with strong concentration and high discriminability. First, the formula uses The product structure is to convert the principal component contribution rate ( ) and frequency energy concentration ( ) These two factors, which represent the "interference dominance" from the spatial feature dimension and the frequency domain feature dimension respectively, are fused. If both are high at the same time, it means that the interference performance of the band combination is highly focused. Secondly, the square operation of the product is performed to enhance the discrimination of the high-value part, widen the numerical difference between the strong interference combination and the ordinary combination, and at the same time suppress the low-value fluctuation and improve the sensitivity of the system. Then, the The logarithmic function structure is to solve the problem of high value amplification while preventing numerical divergence, smoothing and compressing abnormally high interference values, and ensuring that the overall coefficient distribution has good numerical stability and comparability. Finally, subtract the end of the formula , that is, the corresponding information redundancy factor ( ) is used to introduce an "information validity penalty" into the coefficients. This prevents combinations that appear to be concentrated interference but actually contain a large amount of redundant or ineffective interference features from receiving overly high scores, thereby improving the authenticity and reliability of the coefficients. In summary, this formula, through the triple mechanism of "energy focusing enhancement + logarithmic stabilization compression + redundancy penalty adjustment," achieves a comprehensive and highly accurate modeling of the concentrated characteristics of cross-band interference. This makes it suitable for accurately measuring and screening the dominance of interference behavior during multispectral fusion.

[0045] No. The coupling concentration coefficient of the cross-band combination It reflects the degree of focus of the interference characteristics of the combination in the principal component dimension and frequency dimension. Therefore, its value is positively correlated with the interference dominance and discrimination significance of the combination in the multispectral data fusion process. Specifically, when When the value is high, it indicates that the main characteristic dimension (such as the principal component direction and the frequency response region) of the combination in the interference modeling matrix has strong consistency, and the interference signals are concentrated in a small number of characteristic channels, indicating that the interference characteristics of the combination are significant and the structure is clear, which is easy to cause single-point dominant influence on the fusion result, and therefore is regarded as a high interference risk combination. Conversely, when When the value is low, it indicates that the distribution of the combination in the energy characteristics is relatively dispersed, the interference does not have directionality, and there is a certain degree of information redundancy and inefficient characteristics, indicating that its interference behavior is relatively dispersed or non-dominant, and the influence on the fusion model is limited, and therefore it is usually evaluated as a combination with low interference strength. Therefore, as one of the composition parameters of the cross-interference index, the larger the coupling concentration degree coefficient, the more likely it is to trigger interference influence determination in the fusion model, and the weight and sensitivity in the final interference strength evaluation are higher, which has good discrimination ability and interference screening value.

[0046] In this embodiment, the response disturbance index of each cross-band combination is obtained as follows: The response disturbance information is extracted from the cross-interference characteristic information of each preprocessed cross-band combination, specifically including the trend deviation angle, the nonlinear change coefficient and the response rate fluctuation factor of each cross-band combination in the established band interference modeling matrix, and are respectively marked as , and , represents the trend deviation angle of the th cross-band combination in the established band interference modeling matrix, represents the nonlinear change coefficient of the th cross-band combination in the established band interference modeling matrix, represents the response rate fluctuation factor of the th cross-band combination in the established band interference modeling matrix, , is a positive integer; In the established band interference modeling matrix, each cross-band combination includes the spectral response change sequence in the time dimension or the trend vector characteristics calculated therefrom. The system can calculate the angle between the trend direction and the standard or reference response trend based on the main trend change direction vector (such as the first derivative direction) extracted from the spectral response curve of the combination, thereby obtaining the trend deviation angle. Specifically, the software system performs vector cosine similarity conversion or slope vector projection comparison on the “trend evolution sub-matrix” stored in the modeling matrix to obtain the angle difference of the trend deviation direction, and the angle range is usually between the arcs. This calculation process can be completed directly on the modeling matrix structure through vector comparison, gradient direction extraction, and other algorithm modules without calling the original response curve. The extracted trend offset angle is used to quantify the jump degree of the interference behavior in the directionality layer and is the basic factor for identifying the "trend disturbance intensity".

[0047] In the waveband interference modeling matrix, the system usually performs polynomial fitting, spline regression, or orthogonal basis projection on the time series response characteristics of each cross-waveband combination and records the corresponding fitting residual statistical characteristics and curve fitting type characteristics. The non-linear change coefficient can be extracted by analyzing the "fitting residual vector", "local curvature information", and "high-order derivative fluctuation coefficient" fields stored in the modeling matrix. When implemented in software, the system can locate the residual mean, residual variance, and fitting curve curvature change rate associated with each combination in the modeling matrix and combine them into a non-linear change coefficient according to the set weight rule (or PCA compression). The larger this index, the more the interference behavior of the waveband combination deviates from the linear mode and the stronger the non-linear fluctuation. The entire extraction process can be completed on the modeling matrix through residual statistical analysis functions and curvature modeling functions, which is an important measurement parameter for evaluating interference irregularity.

[0048] The response rate fluctuation factor is used to measure whether the spectral response of a cross-waveband combination changes dramatically in unit time and whether there is high-frequency fluctuation behavior. In the waveband interference modeling matrix, this factor can be obtained by extracting the first-order derivative vector or the difference response rate sequence corresponding to each pair of combinations and calculating the variance, standard deviation, and other fluctuation statistics. Specifically, the system can locate the "response rate sequence" field of the combination in the modeling matrix, perform second-order difference, sliding window variance analysis, and other operations on the sequence, and then obtain the fluctuation intensity estimate . At the software level, the system can process the existing derivative level data fields in the matrix through signal differentiation functions, standard deviation calculation, and window analysis modules without reanalyzing the original response curve, thereby achieving efficient extraction of the fluctuation intensity. The larger this index, the more dramatic the response change in the interference process and the greater the impact on the stability of the fusion judgment system, which is an important supporting data for determining the disturbance amplitude.

[0049] The response disturbance index of each cross-waveband combination is calculated . The specific calculation logic is as follows: take the absolute value of the sine value of the trend offset angle of each cross-waveband combination, add one to the product of the corresponding exponential value of the non-linear change coefficient and the response rate fluctuation factor, and take the natural logarithm, and then sum the two to obtain the response disturbance index of each cross-waveband combination. The specific calculation formula is as follows: Where, For the The response disturbance index of the cross-band combination.

[0050] This formula is used to calculate the The core goal of the response disturbance index of a cross-band combination is to comprehensively evaluate the directional change intensity, nonlinear disturbance degree and fluctuation severity of the band combination in the time domain response. Represents the change in the disturbance direction caused by the trend deviation angle of the combination. The larger the angle, the closer the sine value is to ±1, indicating that the change trend of the band response curve is more obvious and the mutation is more severe. This part reflects the directional sensitivity of the disturbance. By nonlinear variation coefficient Response rate fluctuation factor The product of represents the coupling strength of the combination in terms of fluctuation slope, fluctuation frequency, and curvature. It exponentially amplifies the amplitude of the mutation and is highly responsive to strong disturbances. The logarithm is then taken to compress the extreme value fluctuation range and maintain overall numerical stability. The absolute value operation in the outer layer of the entire formula is used to uniformly handle the positive and negative differences in the numerical calculation of rising and falling disturbances, making the exponent comparable and standardized. Therefore, this calculation method can accurately capture the directional jump behavior and dynamic nonlinearity of the cross-band combination in the actual spectral response, and is a key parameter for evaluating disturbance intensity and interference trends.

[0051] No. The response disturbance index of the cross-band combination It is a comprehensive indicator to measure the disturbance direction change amplitude and dynamic fluctuation intensity of the combination in the spectral response sequence. Its value is positively correlated with the cross-interference intensity of the combination. When the value is large, it means that the band combination shows significant trend jump, nonlinear response and high-frequency fluctuation behavior in the modeling process, indicating that the interference generated by the combination is not only severe but also unstable, and poses a high risk of disturbance to the fusion judgment model, so it should be judged as a high interference intensity combination. On the contrary, when A lower value indicates that the response curve of the combination is relatively stable, with strong directional consistency and a stable change pattern, indicating that its perturbation behavior is well controlled and has little impact on the fusion effect. Therefore, as an important component parameter of the cross-interference index, the response perturbation index is more severe. The larger its value, the more severe the interference behavior and the greater the potential for negative impact on reflective film quality monitoring. It is an important criterion for identifying unstable interference combinations.

[0052] In this embodiment, the coupling concentration coefficient of each cross-band combination generated is and response disturbance index Establish a weighted summation model and generate the cross-interference index of each cross-band combination through weighted summation The specific calculation formula is as follows: In the formula, is the cross interference index of the mth cross-band combination, and are the non-zero weight coefficients of the coupling concentration coefficient and the response disturbance index of the mth cross-band combination, and .

[0053] In order to realize the comprehensive quantitative evaluation of the cross interference intensity of each cross-band combination, the system fuses the coupling concentration coefficient and the response disturbance index of the combination by constructing a weighted summation model to generate the cross interference index under the unified quantitative standard. In the specific implementation process, the system first generates the coupling concentration coefficient and the response disturbance index of each cross-band combination, which respectively represent the interference performance of the combination in the interference concentration and dynamic fluctuation dimensions. Subsequently, the system assigns non-zero weight coefficients and to the two indexes and constructs the following fusion model: . Wherein, and are adjustment factors for balancing the influence of the two indexes on the final interference index, and satisfy . The two weight coefficients can be set according to different monitoring strategies and model preferences through manual configuration, historical data fitting or cross-validation, etc. For example, when the system is more sensitive to the interference dominance, the value of can be appropriately increased; if more attention is paid to the instability of interference response, the proportion of can be enhanced. In software implementation, the weights can be preset in the model configuration file as system parameters, or obtained through machine learning algorithm adaptive optimization to ensure that the fusion model has controllability and flexibility. The weighted fusion method makes the final cross interference index have the dual judgment ability of structure focusing and behavior disturbance, providing a high reliability basis for subsequent interference intensity classification and control strategy execution.

[0054] In this embodiment, the pre-set cross interference index threshold interval is determined, and the generated cross interference index of each cross-band combination is compared with the pre-set cross interference index threshold interval, the cross interference intensity of each cross-band combination is evaluated according to the comparison result, and each cross-band combination is divided into an effective fusion group, a limited fusion group and an excluded fusion group according to the evaluation result. The specific comparison and analysis and division are as follows: If​​ The cross interference intensity of the cross-band combination is low intensity, and the cross-band combination is divided into an effective fusion group; When the cross interference index of a certain cross-band combination is less than the minimum value of the preset threshold interval, it indicates that the combination exhibits a highly controllable interference behavior in the spectral fusion process. On the one hand, the coupling concentration coefficient is low, indicating that the combination does not form a dominant interference trend in the structural dimension; on the other hand, the response disturbance index is also at a low level, indicating that it is stable in dynamic fluctuation and gentle in trend change. Such combinations usually do not cause significant deviation in overall reflectivity or optical property evaluation in the multispectral fusion model, but may bring stable feature contribution, and therefore can be classified as an effective fusion group, which is preferentially retained and given a normal weight in subsequent fusion calculation. The retention of this group helps to improve the accuracy and robustness of the fusion model.

[0055] If The cross interference intensity of the cross-band combination is medium intensity, and the cross-band combination is divided into a restricted fusion group; When the cross interference index of a certain cross-band combination falls into the middle section of the preset threshold interval, it indicates that the combination is in a state of ambiguous or controllable but fluctuating risk in interference performance. The coupling concentration and disturbance index are usually in the medium range, representing that such combinations may have disturbance effects on the fusion results under certain conditions, especially when the film layer optical properties are abnormal or the measurement equipment sensitivity changes. Therefore, such combinations are divided into a restricted fusion group, which means that the system will control the participation of the combination in the subsequent fusion process, for example, reducing its weighted weight, increasing the dynamic checking mechanism, or temporarily shielding it under certain working conditions. This classification strategy helps to suppress the cumulative effect of medium-intensity interference without losing potential useful information.

[0056] If The cross interference intensity of the cross-band combination is high intensity, and the cross-band combination is divided into a rejected fusion group.

[0057] When the cross interference index of a certain cross-band combination is higher than the maximum value of the threshold interval, it indicates that the combination is at a high risk level in both structural interference concentration and response fluctuation intensity, showing significant interference characteristics, violent disturbance behavior, and high data instability. Such combinations are likely to cause model abnormal judgment in multispectral data fusion, thereby affecting the overall evaluation of the optical performance of the reflective film, leading to reflectivity misjudgment or film layer misjudgment of unqualified products. Therefore, dividing it into a rejected fusion group means that the combination will be completely excluded from the subsequent fusion calculation, and its data will not be used to participate in the final index calculation, so as to fundamentally eliminate the misleading influence of high-intensity interference sources on the fusion model. This strategy is of key significance to ensure the stability and detection consistency of the monitoring system.

[0058] According to the classification results of each cross-band combination, corresponding processing operations are respectively performed on different types of cross-band combinations; In this embodiment, according to the classification results of each cross-band combination, corresponding processing operations are respectively performed on different types of cross-band combinations, specifically: For the cross-band combination classified into the effective fusion group, the processing operation performed is specifically: directly retaining its corresponding spectral data as a fusion input to participate in subsequent spectral fusion calculation; For the cross-band combination classified into the effective fusion group, the system will default to retain its corresponding spectral data as a regular fusion input to participate in the calculation in the data fusion module. The specific implementation is to set the data channel corresponding to this type of combination to the “full weight input” state in the fusion parameter configuration, that is, to give it the same fusion participation weight as other high-quality bands without any weakening or adjustment. At the same time, the system will automatically skip the interference detection or dynamic monitoring process of this group and directly enter the matrix merging and weight superposition process of the fusion formula in the channel screening logic before fusion. The reason for adopting the direct retention strategy is that this type of combination shows good stability and low disturbance in interference evaluation, and does not constitute a deviation risk to the fusion output. Retaining these data not only does not cause interference, but also can enhance the data signal-to-noise ratio and prediction reliability of the fusion process, and improve the accuracy and robustness of the overall monitoring algorithm.

[0059] For the cross-band combination classified into the limited fusion group, the processing operation performed is specifically: applying a fusion weight reduction operation to its spectral data in the fusion calculation, and increasing a dynamic verification mechanism to limit its influence on the fusion result; For the cross-band combination classified into the limited fusion group, the system will perform fusion weight reduction or fusion effectiveness dynamic verification operation on its spectral data in the fusion calculation. The specific way includes: in the fusion weight matrix, the data weight coefficient corresponding to this group is proportionally attenuated (for example, set to 50% of the original full weight or dynamically mapped according to the interference level); or configure a dynamic fusion verification mechanism for this combination, which automatically judges whether to enable or shield its data input according to the fluctuation of its current spectral response in each round of fusion process. These processing actions are realized through the weight regulation engine and verification trigger built in the fusion strategy module. The adoption of this “limit participation but not completely exclude” strategy is based on the fact that this group of combinations is in the interference critical zone in the evaluation, although it has potential disturbance risk, but it may still contain some useful feature information, appropriate limitation of its influence range can effectively reduce its interference probability on the fusion output, while avoiding the precision decline caused by information loss.

[0060] For the cross-band combinations classified into the rejection fusion group, the processing operation performed is: completely rejecting the corresponding spectral data, not included in the fusion model calculation range.

[0061] For the cross-band combinations classified into the rejection fusion group, the system will completely reject the corresponding spectral data channel, that is, it will be removed from the fusion input set in the preprocessing stage before data fusion. The implementation is to set the "masking mark" or "weight zero" configuration item for this type of combination in the fusion channel management module, so that its data is excluded before participating in the fusion matrix construction and fusion function execution, and does not enter the operation path of the fusion model. At the same time, this type of combination will also be labeled with an interference risk label and archived to an abnormal interference library for subsequent model calibration or quality warning triggering. The adoption of this complete rejection strategy is because the interference intensity of this type of combination has exceeded the acceptable range in the interference intensity evaluation, and if it continues to be used, it will greatly increase the fusion misjudgment probability and reduce the stability of the film layer optical index evaluation. Therefore, from the perspective of system robustness and production line monitoring consistency, the data channel must be completely shielded to ensure the reliability and controllability of the fusion result.

[0062] Based on the historical fusion processing data and film layer detection results, the determination method of cross interference intensity, the classification strategy of cross-band combination and the processing operation are optimized, and the classification results and processing records are visualized and historically archived.

[0063] In the software system, in order to improve the long-term adaptability of the cross interference identification and processing strategy, the system will regularly execute the feedback closed-loop optimization mechanism based on the historical fusion processing data and the actual film layer detection results. The specific implementation is: the system establishes a fusion log database and a film layer detection result comparison library, and associates the interference index, classification level, processing method of each cross-band combination in the past fusion process with the subsequent film layer quality evaluation data. Then, through statistical models (such as confusion matrix analysis, bias regression analysis) or machine learning algorithms (such as random forest feature importance evaluation, logic regression discriminant function retraining), the system optimizes or adjusts the weights of the original threshold interval, classification boundary and key parameters in the processing strategy used for interference intensity determination. This optimization process can be set to trigger regularly or automatically activate when the fusion error rate exceeds the set range, so as to realize the model self-adaptation evolution. The purpose of this design is to make the classification and processing strategy adjust continuously with the changes of material batches, spectral response characteristics or equipment performance, so as to ensure that the interference identification accuracy, fusion judgment rationality and production line adaptability always maintain a high level in the long-term operation.

[0064] To improve the operability of the system and the traceability capability, the system stores and displays the classification results and processing records generated in each round of fusion processing in a structured manner. The specific implementation is as follows: the classification level, interference index, processing measure, timestamp and other fields of all cross-band combinations are summarized into standardized structured data and stored in the fusion monitoring log library. The front end of the system displays the state changes, processing trajectories and interference fluctuation trends of each combination in different time periods to the operator in an interactive graphical form through a data visualization engine (such as chart controls, time series heat maps, interference trend line graphs, etc.). At the same time, the system also supports automatic generation of archived reports for statistical analysis of historical fusion effects and processing efficiency according to batches, time windows or film layer types. This visualization and archiving mechanism not only helps operators intuitively grasp the execution effect of the interference control strategy, but also provides authoritative data support for later technical audits, process tracing, product consistency analysis, etc., and is a key means to realize transparent operation and management of the entire life cycle of fusion quality.

[0065] The above formulas are dimensionless numerical calculations. The formulas are obtained by collecting a large amount of data to simulate the current real situation. The preset parameters in the formulas are set by a person skilled in the art according to the actual situation.

[0066] The above embodiments can be implemented wholly or partially by software, hardware, firmware or any other combination. When implemented by software, the above embodiments can be implemented wholly or partially in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions according to the embodiments of the present application are wholly or partially generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another, for example, the computer instructions can be transferred from one website, computer, server or data center to another by wire or wireless (such as infrared, wireless, microwave, etc.) mode. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. containing one or more available medium collections. The available medium can be a magnetic medium (such as a floppy disk, a hard disk, a magnetic tape), an optical medium (such as a DVD) or a semiconductor medium. The semiconductor medium can be a solid-state disk.

[0067] It should be understood that the size of the sequence number of each process in various embodiments of the present application does not mean the order of execution, and the execution order of each process should be determined according to its function and inherent logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0068] Those skilled in the art can understand that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0069] In several embodiments provided in the present application, it should be understood that the disclosed system and method can be implemented in other ways. For example, the above-described embodiments are merely illustrative, for example, the division of units is only a logical function division, and actual implementation can have another division manner, for example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some interfaces, devices or units, and can be electrical, mechanical or other forms.

[0070] The units described as separate components can or can not be physically separated, and the components shown as units can or can not be physical units, that is, they can be located in one place, or can be distributed on multiple network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the embodiment.

[0071] In addition, each functional unit in each embodiment of the present application can be integrated into a processing unit, or each unit can be physically present separately, or two or more units can be integrated into one unit.

[0072] The above is merely a specific implementation of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A multi-spectral data fusion intelligent monitoring method for the reflective film production process, characterized in that: The specific steps include: The multi-channel spectral sensor is used to collect the spectral data of the film layer to be inspected in the reflective film production process in different spectral bands in real time and pre-process it; Based on the preprocessed spectral data, all spectral band combinations with cross-interference between different spectral bands are screened out and marked as cross-band combinations; Establishing a band interference modeling matrix for each pair of spectral bands in the cross-band combination, wherein the band interference modeling matrix is ​​used to describe interference behavior characteristics between the spectral band pairs; Based on the established band interference modeling matrix, the cross-interference strength of each cross-band combination is determined, and each cross-band combination is classified; According to the classification results of each cross-band combination, corresponding processing operations are performed on different types of cross-band combinations; Based on historical fusion processing data and film layer detection results, the method for determining cross-interference intensity, the classification strategy of cross-band combinations, and the processing operations are optimized, and the classification results and processing records are visualized and historically archived.

2. The multispectral data fusion intelligent monitoring method in the reflective film production process according to claim 1, characterized in that: The method screens out all spectral band combinations with cross-interference between different spectral bands based on the preprocessed spectral data and calibrates them as cross-band combinations. Specifically, the method comprises the following steps: performing statistical analysis operations on the preprocessed spectral data corresponding to each pair of spectral band combinations, calculating a change trend correlation index and a signal interference coupling characteristic index between each pair of spectral bands, and identifying all spectral band combinations with cross-interference based on whether any one of the change trend correlation index and the signal interference coupling characteristic index exceeds a preset interference judgment threshold, and calibrating them as cross-band combinations.

3. The multispectral data fusion intelligent monitoring method in the reflective film production process according to claim 2, characterized in that: The step of establishing a band interference modeling matrix for each pair of spectral bands in the cross-band combination specifically includes the following steps: Based on the pre-processed spectral data corresponding to each pair of spectral bands in the calibrated cross-band combination within a preset time window, a time series matrix is ​​constructed to characterize the joint response characteristics of the pair of spectral bands; Performing principal component analysis, covariance analysis, and mutual information analysis on the time series matrix to extract a plurality of characteristic factors reflecting the interference characteristics between the pair of spectral bands; The extracted characteristic factors are integrated into a band interference modeling matrix according to statistical behavior, frequency coupling relationship and response change pattern, which is used to describe the joint behavioral characteristics between the pair of spectral bands in terms of spectral response intensity, change trend and interference coupling.

4. The multispectral data fusion intelligent monitoring method in the reflective film production process according to claim 3, characterized in that: Based on the established band interference modeling matrix, the cross-interference strength of each cross-band combination is determined, and each cross-band combination is classified, which specifically includes the following steps: Extracting cross-interference feature information of each cross-band combination from the established band interference modeling matrix and performing preprocessing after extraction; Extracting energy distribution information and response disturbance information from the pre-processed cross-interference characteristic information of each cross-band combination, and analyzing them after extraction to generate the coupling concentration coefficient and response disturbance index of each cross-band combination respectively; A weighted summation model is established for the coupling concentration coefficient and response disturbance index of each cross-band combination, and the cross-interference index of each cross-band combination is generated by weighted summation; Determine a pre-set cross-interference index threshold range, and compare it with the generated cross-interference index of each cross-band combination after determination. Evaluate the cross-interference intensity of each cross-band combination based on the comparison results, and divide each cross-band combination into a valid fusion group, a restricted fusion group, and a eliminated fusion group based on the evaluation results.

5. The multi-spectral data fusion intelligent monitoring method in the reflective film production process according to claim 4, characterized in that: The logic for obtaining the coupling concentration coefficient of each cross-band combination is as follows: The energy distribution information is extracted from the cross-interference feature information of each cross-band combination after preprocessing, specifically including the first principal component contribution rate, frequency energy concentration and information redundancy factor of each cross-band combination in the established band interference modeling matrix, and calibrated as 、 and , Indicates the first The first principal component contribution rate of the cross-band combination, Indicates the first The frequency energy concentration of the cross-band combination, Indicates the first The information redundancy factor of the cross-band combination, , is a positive integer; Calculate the coupling concentration coefficient for each cross-band combination The specific calculation logic is: multiply the first principal component contribution rate of each cross-band combination by the frequency energy concentration and square it, add one to the result and take the natural logarithm, then subtract the square root of the corresponding information redundancy factor. The resulting value is the coupling concentration coefficient of each cross-band combination.

6. The multi-spectral data fusion intelligent monitoring method in the reflective film production process according to claim 5, characterized in that: The logic for obtaining the response disturbance index of each cross-band combination is as follows: The response disturbance information is extracted from the cross-interference characteristic information of each cross-band combination after preprocessing, specifically including the trend offset angle, nonlinear variation coefficient and response rate fluctuation factor of each cross-band combination in the established band interference modeling matrix, and calibrated as 、 and , Indicates the first The trend deviation angle of the cross-band combination, Indicates the first The nonlinear variation coefficient of the cross-band combination, Indicates the first The response rate fluctuation factor of the cross-band combination, , is a positive integer; Calculate the response perturbation index for each cross-band combination The specific calculation logic is: after taking the absolute value of the sine value of the trend offset angle of each cross-band combination, add one to the exponential value corresponding to the product of its nonlinear change coefficient and the response rate fluctuation factor, and then take the natural logarithm. The sum of the two is the response disturbance index of each cross-band combination.

7. The multi-spectral data fusion intelligent monitoring method in the reflective film production process according to claim 6, characterized in that: The coupling concentration coefficient of each cross-band combination generated and response disturbance index Establish a weighted summation model and generate the cross-interference index of each cross-band combination through weighted summation .

8. The multi-spectral data fusion intelligent monitoring method in the reflective film production process according to claim 7, characterized in that: Determine the pre-set cross-interference index threshold range , and after determination, the cross-interference index of each cross-band combination generated A comparison is performed, and the cross-interference strength of each cross-band combination is evaluated based on the comparison results. Based on the evaluation results, each cross-band combination is divided into an effective fusion group, a restricted fusion group, and a eliminated fusion group. The specific comparison analysis and division are as follows: like , the cross-interference intensity of the cross-band combination is low, and the cross-band combination is divided into an effective fusion group; like , the cross-interference intensity of the cross-band combination is medium, and the cross-band combination is divided into a restricted fusion group; like , the cross-interference intensity of the cross-band combination is high, and the cross-band combination is divided into a elimination fusion group.

9. The multi-spectral data fusion intelligent monitoring method in the reflective film production process according to claim 8, characterized in that: According to the classification results of each cross-band combination, corresponding processing operations are performed on different types of cross-band combinations, specifically: For the cross-band combination that is divided into a valid fusion group, the specific processing operations performed are: directly retaining its corresponding spectral data as fusion input to participate in the subsequent spectral fusion calculation; For the cross-band combinations that are classified as restricted fusion groups, the following processing operations are performed: a fusion weight reduction operation is applied to their spectral data in the fusion calculation, and a dynamic verification mechanism is added to limit its impact on the fusion results; For the cross-band combination that is divided into the elimination fusion group, the specific processing operation performed is: completely eliminate the corresponding spectral data and do not include it in the calculation range of the fusion model.

Images