Multi-spectral color data calibration method and system based on multi-source perception data fusion

Abstract

The invention relates to the field of electrical digital data processing, and relates to a multispectral color data calibration method and system based on multisource perception data fusion, and the method comprises the steps: obtaining original color data and environment data of a target object measured by a multispectral sensor; judging whether the original color data meets a calibration condition or not; if the original color data meets the calibration condition, calling main reference color data corresponding to the multispectral sensor; comparing the original color data with the main reference color data, and calculating to obtain a first compensation parameter for correcting an inherent difference value between the equipment; the environment data are input into a preset environment compensation model, a second compensation parameter is output through the environment compensation model, and the second compensation parameter is used for correcting the environment interference value; and performing fusion calibration on the original color data by using the first compensation parameter and the second compensation parameter to generate calibrated color data. Data of different devices and in different environments are normalized to objective standards.

Description

Technical Field

[0001] This application relates to the field of electronic digital data processing technology, and in particular to a multispectral color data calibration method and system based on multi-source sensing data fusion. Background Technology

[0002] In fields such as industrial inspection, printing and textiles, and artwork digitization, where extremely high accuracy in color measurement is required, multispectral sensors are gradually becoming core tools for the objectification and digital representation of color data due to their ability to acquire continuous spectral information of objects. However, transforming high-precision laboratory equipment into reliable tools that can adapt to complex industrial environments and ensure long-term stability and data interoperability still faces a series of severe technical challenges. Currently, the mainstream technical approaches in the industry to solve the measurement stability problem mainly focus on hardware improvement and periodic calibration. For example, reducing hardware drift by selecting more stable optical components and improving heat dissipation design, or requiring users to periodically return the equipment to the laboratory for recalibration using higher-level standards to correct errors. While these methods have some effect, they are costly and cannot solve the problems of real-time performance and data collaboration in the field.

[0003] Specifically, most existing technical solutions focus on a single source of error: either compensating for changes in the device's own state by measuring a built-in standard whiteboard, or simply adjusting the gain of the measurement results using ambient light sensor readings. These methods treat "device performance drift" and "external environmental interference" as independent and linearly superimposed factors, employing fixed, separate compensation strategies. However, in complex real-world operating conditions, these two factors often couple and change dynamically, causing simple compensation models to quickly fail. For example, changes in device temperature can cause performance drift in internal components and may also alter the characteristics of ambient light; existing separate processing methods struggle to accurately attribute and compensate for these factors.

[0004] Secondly, existing technologies are particularly inadequate in modeling and processing specific error sources that are more subtle and complex. On the one hand, the understanding of differences between devices often remains at the factory calibration data, lacking the ability to finely model and learn online the "individuality" of the equipment (such as manufacturing tolerances) and its "aging" during its life cycle (such as long-term drift and sudden changes). Existing methods struggle to distinguish between inherent, stable systematic biases and random, transient state disturbances in measurement differences, leading to compensation models that are either too rigid to adapt to changes or lose stability due to overfitting noise. On the other hand, the complexity of ambient light far exceeds conventional assumptions. Most environmental compensation models are designed based on the assumption that "ambient light is a broadband continuous light source." When high-intensity, narrow-band monochromatic light interference exists at the measurement site, this interference light couples with the equipment's own light source within the sensor, forming an atypical "false signal" with fixed spectral characteristics. Existing methods cannot effectively identify this special interference, and their pre-set broadband interference compensation models cannot properly handle it, ultimately leading to untraceable and systematic biases in the calibration results. Summary of the Invention

[0005] In order to solve one or more problems in the prior art, the main objective of this application is to provide a multispectral color data calibration method and system based on multi-source sensing data fusion.

[0006] To achieve the aforementioned objectives, this application proposes a multispectral color data calibration method based on multi-source sensing data fusion, the method comprising: Acquire raw color data and environmental data of the target object measured by the multispectral sensor; Determine whether the original color data meets the calibration conditions; If the original color data meets the calibration conditions, the master reference color data corresponding to the multispectral sensor is called. The original color data is compared with the main reference color data to calculate the first compensation parameter used to correct the inherent differences between devices; Environmental data is input into a preset environmental compensation model, and a second compensation parameter is output through the environmental compensation model. The second compensation parameter is used to correct environmental disturbance values. The original color data is fused and calibrated using the first compensation parameter and the second compensation parameter to generate calibrated color data.

[0007] This application also provides a multispectral color data calibration system based on multi-source sensing data fusion, including: The acquisition module is used to acquire the raw color data and environmental data measured by the multispectral sensor on the target object; The judgment module is used to determine whether the original color data meets the calibration conditions; The calling module is used to call the master reference color data corresponding to the multispectral sensor when the original color data meets the calibration conditions; The calculation module is used to compare the original color data with the main reference color data and calculate a first compensation parameter for correcting the inherent differences between devices. The input module is used to input environmental data into a preset environmental compensation model, and output a second compensation parameter through the environmental compensation model. The second compensation parameter is used to correct the environmental disturbance value. The generation module is used to perform fusion calibration on the original color data using the first compensation parameter and the second compensation parameter to generate calibrated color data.

[0008] This application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of any of the methods described above.

[0009] This application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the methods described above.

[0010] The multispectral color data calibration method and system based on multi-source sensing data fusion in this application establishes a joint input for error diagnosis by simultaneously acquiring raw color data and environmental data, enabling the calibration process to have environmental context awareness. Secondly, it introduces intelligent judgment of calibration conditions, achieving adaptive triggering of the calibration process while balancing high precision and system efficiency. Two targeted compensation parameters are generated in parallel: the first compensation parameter accurately corrects inherent system deviations caused by manufacturing tolerances and long-term use of each device by comparing with personalized master reference data, ensuring the long-term stability of individual devices; the second compensation parameter, through a pre-trained environmental compensation model, offsets on-site interference introduced by stray light, temperature changes, etc., in real time, ensuring the on-site robustness of measurements. Finally, by fusing the two compensation parameters to perform integrated correction on the raw data, data measured by different devices in different environments are normalized to an objective standard, fundamentally achieving high-confidence consistency and comparability of color data across links and devices in the supply chain. Attached Figure Description

[0011] Figure 1 This is a flowchart illustrating a multispectral color data calibration method based on multi-source sensing data fusion according to an embodiment of this application. Figure 2This is a flowchart illustrating a multispectral color data calibration method based on multi-source sensing data fusion according to an embodiment of this application. Figure 3 This is a schematic block diagram of a multispectral color data calibration system based on multi-source sensing data fusion according to an embodiment of this application; Figure 4 This is a schematic block diagram of the structure of a computer device according to an embodiment of this application.

[0012] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0013] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0014] Reference Figure 1 This application provides a multispectral color data calibration method based on multi-source sensing data fusion, the method comprising: S1. Acquire the raw color data and environmental data of the target object measured by the multispectral sensor; S2. Determine whether the original color data meets the calibration conditions; S3. If the original color data meets the calibration conditions, call the master reference color data corresponding to the multispectral sensor; S4. Compare the original color data with the main reference color data to calculate the first compensation parameter used to correct the inherent differences between devices; S5. Input the environmental data into the preset environmental compensation model, and output the second compensation parameter through the environmental compensation model. The second compensation parameter is used to correct the environmental interference value. S6. Using the first compensation parameter and the second compensation parameter, the original color data is fused and calibrated to generate calibrated color data.

[0015] As described in steps S1-S3 above, the multi-source sensing data includes raw color data (from the sensor's measurement of the target) and environmental data (from the monitoring of environmental sensors). These represent the target information and external interference information, respectively. Step 1 activates the multispectral sensor's built-in light source to illuminate the target and receives the reflected light. This light is then converted into intensity values ​​of discrete spectral channels by a spectrometer array, forming the "raw color data." Simultaneously, auxiliary sensors integrated into the device, such as ambient light sensors and temperature sensors, synchronously collect "environmental data" from the measurement site, including but not limited to ambient light intensity, coarse spectral composition, and the device's temperature. This synchronous acquisition mechanism ensures that each color data sample carries the environmental context label of its generation time. This establishes a joint input of "measurement data - environmental state" for subsequent intelligent compensation, enabling calibration to respond to dynamically changing site conditions. Step 2 introduces intelligent scheduling logic for the calibration process. Judgment conditions are based on a preset, configurable rule set, which comprehensively considers the needs of system efficiency and accuracy assurance. For example, the rules might include: the cumulative number of measurements since the last successful calibration, the continuous operating time of the device since the last calibration, the fluctuation range of environmental data relative to the baseline state during the current measurement, or the signal-to-noise ratio of the original color data itself. By evaluating these indicators in real time, the system decides whether to initiate the complete calibration calculation process or reuse historical calibration parameters to improve response speed. This achieves adaptive triggering of calibration behavior, avoiding redundant calculations in scenarios where data is stable or high-frequency calibration is not required, and optimizing system resource utilization and real-time response. Simultaneously, when a significant change that may affect accuracy is detected (such as a sudden temperature change), calibration can be initiated promptly, ensuring the continuous reliability of measurement results. Step 3 provides an individualized reference baseline for calibration. The "master reference color data" is the reference data saved by each device under controlled conditions when measuring a standard color chart at the factory or during the last high-standard calibration. It is stored in encrypted form in the device's local secure storage area or securely retrieved from the cloud based on the device's unique identifier. This data carries the "spectral fingerprint" of the specific device under ideal conditions and is the fundamental basis for measuring changes in its current state. By invoking personalized benchmarks, rather than generic theoretical values, the method fundamentally acknowledges and prepares to correct for the unique inherent characteristics of each device. This is the logical starting point for achieving the goal of "cross-device consistency," ensuring the targeted and accurate nature of the compensation.

[0016] As described in steps S4-S6 above, step 4 aims to quantify the systematic bias of the device itself. A difference mapping relationship is calculated by mathematically comparing the currently measured raw data with individual baseline data channel-by-channel or feature-based. This relationship is encapsulated as the "first compensation parameter," which is essentially a mathematical model or a set of coefficients used to describe the degree of deviation of the current device state from its ideal baseline state, and aims to correct this deviation in reverse. This directly addresses the long-standing pain point of "device state drift." By calculating the device-specific first compensation parameter, it is possible to effectively compensate for the inherent systematic errors of the device that occur slowly due to light source aging, changes in detector sensitivity, and performance degradation of optical components, thereby maintaining the long-term measurement accuracy of the device. Step 5 quantifies the impact of environmental interference. The "environmental compensation model" is a mathematical or statistical model pre-trained with a large amount of experimental data, establishing a mapping relationship from the input of "environmental data" to the output of "interference amount on spectral measurement." The model can understand how factors such as ambient light of different intensities and spectral compositions, as well as temperature, systematically affect the shape and intensity of the measured spectrum. Inputting real-time environmental data, the model can output the corresponding "second compensation parameter." This proactively combats the problem of "environmental interference." This step expands calibration beyond laboratory environments, enabling intelligent identification and compensation for errors introduced by complex factors such as stray light and temperature fluctuations in the field, thus enhancing the method's applicability in uncontrolled environments like industrial sites. Step 6 is the comprehensive execution and data regeneration stage of compensation. Two compensation parameters, one for inherent equipment bias and the other for environmental interference, are applied to the original color data according to a predetermined fusion strategy (such as sequential application or weighted synthesis). This process performs a mathematical transformation on the original data, aiming to simultaneously eliminate both systematic equipment errors and instantaneous environmental errors, thereby outputting "standard color data" that is closer to that obtained under ideal equipment and ideal conditions. This achieves integrated correction of dual error sources. The final calibrated data output possesses both long-term equipment stability and instantaneous environmental interference immunity. It solves the problem of cross-device data comparability, because data measured by different devices and under different environments are normalized to a more objective and consistent standard after this process.

[0017] As mentioned above, by synchronously acquiring raw color data and environmental data, a joint input for error diagnosis was established, enabling the calibration process to possess environmental context awareness. Secondly, intelligent judgment of calibration conditions was introduced, achieving adaptive triggering of the calibration process, balancing high precision and system efficiency. Two targeted compensation parameters were generated in parallel: the first compensation parameter, by comparing with personalized master reference data, accurately corrected the inherent system deviations of each device caused by manufacturing tolerances and long-term use, ensuring the long-term stability of individual devices; the second compensation parameter, through a pre-trained environmental compensation model, offset field interference introduced by stray light, temperature changes, etc., in real time, ensuring the field robustness of measurements. Finally, by fusing the two compensation parameters to perform integrated correction on the raw data, data measured by different devices under different environments were normalized to an objective standard, fundamentally achieving high-confidence consistency and comparability of color data across links and devices in the supply chain.

[0018] Reference Figure 2 In one embodiment, the step of comparing the original color data with the master reference color data to calculate a first compensation parameter for correcting inherent differences between devices includes: S41. Based on the device identifier of the multispectral sensor, obtain the corresponding device characteristic vector, wherein the device characteristic vector characterizes the spectral response tolerance of the multispectral sensor in the manufacturing process; S42. Calculate the original spectral difference vector based on the original color data and the main reference color data; S43. Perform coupling analysis between the original spectral difference vector and the device characteristic vector; S44. Based on the results of the coupling analysis, the stable difference component contributed by the inherent system deviation of the equipment and the random difference component contributed by the current instantaneous state are separated. The random difference component is used for accumulation processing and is used to drive the update of the feature vector of the equipment when a preset condition is met. S45. Generate the first compensation parameter based on the stable difference component.

[0019] As described above, in step 1, the minute performance fluctuations and assembly differences of the internal optical components (such as light sources, filters, and detectors) of each multispectral sensor during manufacturing will cause an unavoidable and unique systematic shift in the overall spectral response curve, i.e., manufacturing tolerance. This step uses the device's unique identifier to call a pre-calibrated and stored "device characteristic vector." This vector, in the form of a set of mathematical coefficients or a response function, precisely quantifies the inherent deviation pattern of the device relative to the standard ideal response model at the time of manufacture. It digitizes the individual differences of devices in the physical world into a specific mathematical model that can be directly called by the algorithm. This lays a calculable foundation for subsequent high-precision, personalized difference analysis, upgrading calibration from compensating for the general errors of "a class of devices" to a dedicated correction for the unique fingerprint of "this one device," fundamentally improving the potential for cross-device consistency. Step 2 aims to quantify the total deviation between the current measurement state and the device's ideal reference state. By performing element-wise mathematical comparisons on multiple spectral channels using the real-time measured raw color data and the master reference color data representing the gold standard of the device's ideal performance, a difference sequence is calculated. This sequence constitutes the original spectral difference vector, which macroscopically and indiscriminately includes the sum of all factors causing the current measurement to deviate from the ideal value. It generates primary data characterizing the current overall performance deviation of the equipment. This transforms the abstract concept of "inaccuracy" into a concrete, computable mathematical object, providing the most basic input material for in-depth analysis and separation of various error components. Step 3, coupling analysis, is an advanced data processing procedure whose core is exploring the intrinsic relationship and interaction patterns between two vectors. Here, the algorithm correlates the equipment characteristic vector characterizing the inherent tolerances of the equipment with the original spectral difference vector characterizing the current total deviation. Through pattern matching, correlation analysis, or more complex mathematical modeling, the algorithm attempts to determine how much of the total deviation can be explained or predicted by the known inherent characteristics of the equipment. This introduces crucial prior knowledge for difference analysis. This allows the system to move beyond simple difference calculations and begin intelligent attribution analysis. Through coupling analysis, the system initially determines the extent to which the current deviation conforms to the "inherent" error pattern of the equipment, providing logical guidance for accurately separating long-term inherent deviations and short-term random disturbances. Step 4, based on the mode correlation information provided by coupling analysis, employs algorithms from the field of signal processing, such as blind source separation or state-space estimation, to decompose the original spectral difference vector into two sub-components with different statistical properties. The stable difference component is strongly correlated with the device characteristic vector, changing slowly and reflecting the systematic and trend-based drift of device hardware performance over time and use. The random difference component, on the other hand, is weakly correlated or uncorrelated with the device characteristic vector, changing rapidly and without a fixed pattern, mainly originating from random interference such as electronic noise, instantaneous temperature fluctuations, and accidental changes in ambient light during a single measurement. This achieves accurate source tracing and physical separation of the mixed error signal.This is the core breakthrough in the method's intelligence. It enables the system to clearly distinguish between "what kind of slow changes are occurring within the equipment itself" and "what kind of temporary interference has occurred during this measurement," thus creating conditions for adopting the most appropriate handling strategies for errors of different natures. Although the random difference components in step 5 appear as random noise in a single instance, the evolution trend of their long-term statistical distributions (such as mean, variance, and higher-order moments) implicitly contains subtle information about the equipment's health status. The system continuously collects random difference components and performs long-term tracking and statistical analysis. When statistical indicators (e.g., within a sufficiently long observation window, the mean of the random components undergoes a sustained and significant shift, or its variance shows a systematic increasing trend) indicate that the instantaneous noise base of the equipment has undergone a non-negligible systematic change, the equipment feature vector update mechanism is triggered. This adjusts the vector parameters with a small learning step size to reflect the latest long-term statistical characteristics of the equipment. This endows the equipment identity model with the intelligence of "progressive learning" and "adaptive evolution." By monitoring the long-term statistical patterns of seemingly random noise, the system can keenly detect the slow performance degradation of equipment caused by material aging, component fatigue, etc., and automatically and cautiously update its digital model. This effectively addresses the slow drift problem throughout the equipment's lifecycle, ensuring the calibration model remains up-to-date and effective over the long term, while avoiding the risk of erroneous model updates due to a single drastic disturbance. Step 6, the stable difference component, directly and purely characterizes the systematic deviation of the current equipment relative to its ideal state. Based on this component, a first compensation parameter is generated by designing a corresponding inverse transform function or fitting a set of correction coefficients. This parameter is essentially a mathematical operator that, when applied to the original color data, aims to precisely and inversely cancel out the spectral distortion introduced by the current systematic deviation of the equipment. Using the separated, highly pure systematic deviation information to generate the compensation parameter ensures that the correction action of the first compensation parameter is highly accurate and reliable. It directly and effectively neutralizes the inherent, slowly changing systematic errors of the equipment, providing a fundamental guarantee for the accuracy of each measurement result and is also the core correction step for achieving data comparability between different devices.

[0020] In this embodiment, traditional solutions typically treat device differences as a general, static, or simply time-varying offset, using fixed lookup tables or linear coefficients for global compensation. This approach fails to address the complexity of the intertwined stable and random components in device errors, and is ill-suited to adapting to the performance evolution of devices over long-term use. This embodiment introduces a "device characteristic vector" as the initial digital identity of the device and employs "coupling analysis" and "signal separation" techniques to successfully separate the "stable difference component" representing long-term trends and the "random difference component" representing instantaneous disturbances from the mixed total measurement differences. This physical mechanism-based separation is a prerequisite for accurate compensation, forming an intelligent closed loop of "measurement-analysis-learning."

[0021] In one embodiment, the step of calculating the original spectral difference vector based on the original color data and the master reference color data includes: The original color data and the primary reference color data are respectively converted into a first spectral vector and a second spectral vector with the same dimension; Call the spectral vector space, and calculate the channel-by-channel relative deviation vector between the first spectral vector and the second spectral vector in the spectral vector space; Based on the calculation results, the channel-by-channel relative deviation vector is normalized to obtain a standardized spectral difference vector. The spectral difference vector is multiplied by a pre-stored device spectral sensitivity feature vector to obtain a weighted spectral difference vector, which is then used as the original spectral difference vector.

[0022] As mentioned above, the original color data and the primary reference color data in Step 1 may have different data organization methods or lengths in their initial state due to their source or storage format. This step ensures that both are represented as spectral vectors with identical dimensions through data reconstruction and alignment algorithms, such as interpolation, truncation, or specific channel mapping. Each vector dimension corresponds to a specific spectral channel, and its value represents the response intensity of that channel. This essentially places the color data in a unified mathematical space for expression. By forcing the data to align in dimensions, comparison errors caused by inconsistent data formats or channel definitions are eliminated, ensuring that subsequent difference calculations are performed one-to-one on each independent and comparable spectral channel, guaranteeing the purity and accuracy of the difference analysis. Step 2 performs element-wise mathematical operations on the first spectral vector (current measurement) and the second spectral vector (ideal reference) in the unified spectral vector space. This step does not simply calculate the absolute difference, but rather the relative deviation, for example, using the ratio of the difference to the reference value, logarithmic difference, or other normalized difference forms. This operation generates a new vector, where each element represents the proportion or degree of deviation of the current measurement value from the ideal reference value on the corresponding spectral channel. A difference spectrum reflecting the relative error distribution was obtained. The relative deviation calculation mitigates the impact of different absolute signal intensities across channels, making the deviations observed in weak and strong signal channels comparable. This vector visually reveals the distribution of equipment performance deviations in different spectral regions (such as blue, green, red, and near-infrared bands), providing detailed spectral clues for diagnosing equipment problems. Step 3 performs scale normalization on the relative deviation vector obtained in the previous step, for example, mapping the numerical range of all its elements to a fixed interval, such as [0,1] or [-1,1], or making it conform to a standard normal distribution with a mean of 0 and a variance of 1. This process is completed through a specific mathematical transformation. This eliminates the influence of overall amplitude fluctuations in the difference vector, focusing its core information on the "pattern of difference" rather than the "absolute magnitude of difference." Normalization ensures that subsequent analysis, especially its combination with weight vectors and the comparison of difference vectors between different measurements, is no longer affected by fluctuations in the overall brightness or signal level of a single measurement, improving the stability of the analysis and its adaptability to different measurement conditions. Step 4: The device spectral sensitivity feature vector is a weighted vector obtained in advance through statistical analysis of historical data of the device (or a device of the same model). The weight of each dimension of this vector quantifies the sensitivity or importance of the corresponding spectral channel to the overall state drift of the device. By performing a dot product operation between the standardized spectral difference vector and this feature vector, a weighted summation is essentially performed: the deviation value of each channel in the difference vector is multiplied by the sensitivity weight of that channel, and then summed, finally outputting a scalar value or a new vector after weight adjustment as the "original spectral difference vector".This represents a leap from "treating all channel differences equally" to "intelligently focusing on key differences." This step embeds the device's historical experience (which bands are most prone to drift and which have the greatest impact on the final color result) into real-time difference calculations. For historically stable or unimportant channels, even large deviations are weighted down; while for sensitive and important channels, even minor deviations are amplified and given greater attention. This makes the calculated "raw spectral difference vector" no longer a simple observation, but a "diagnostic index" weighted by device knowledge, better reflecting the device's true health status and the severity of problems, greatly improving the quality and relevance of input signals in subsequent coupling analysis steps.

[0023] In this embodiment, the "calculation of the original spectral difference vector" is transformed into an intelligent sensing process that integrates data standardization and prior knowledge weighting. Traditional methods typically treat all spectral channels equally when calculating differences, using equal weights for absolute or relative differences, ignoring the fundamentally different importance of different spectral channels in device error mechanisms and color perception. This solution introduces the key element of the "device spectral sensitivity feature vector," transforming historical performance statistics of the device into guiding weights for real-time analysis. The entire process is meticulously designed: first, comparability is ensured through format standardization and relativization; then, amplitude interference is removed through normalization to focus on the difference pattern; finally, dot product operations are used to organically combine empirical weights with real-time differences. These steps work together to ensure that the final output "original spectral difference vector" is no longer a coarse raw difference value containing a large amount of redundant and misleading information, but a refined signal highlighting the core deviation characteristics of the device.

[0024] In one embodiment, the step of inputting environmental data into a preset environmental compensation model and outputting a second compensation parameter through the environmental compensation model includes: Multiple environmental feature parameters are extracted from the environmental data, including ambient light intensity, ambient light spectral distribution characteristics, and equipment operating temperature. The environmental feature parameters of the multiple dimensions are input into the environmental compensation model, which includes an environmental sensitivity matrix corresponding to the multispectral sensor. Based on the environmental sensitivity matrix, the environmental compensation model calculates the interference weight of each dimension's environmental feature parameter on each spectral channel, and then generates a multi-dimensional environmental interference vector. Based on the output results, a mapping analysis is performed on the multidimensional environmental interference vector to determine the second compensation parameter.

[0025] As mentioned above, the raw environmental data acquired in Step 1 is typically raw voltage or digital signals, which do not possess direct physical meaning. This step is a feature engineering process that transforms these raw signals into quantitative features with clear physical definitions through specific algorithms and calibration relationships. Ambient light intensity is obtained by illuminance calibration of the ambient light sensor signal; the ambient light spectral distribution characteristics are estimated by analyzing the response of multi-channel or spectral ambient light sensors to determine the dominant wavelength, color temperature, or similarity index with a standard light source; the equipment operating temperature is directly derived from the calibration reading of the temperature sensor. These parameters together constitute a standardized and structured description of the environmental conditions at the measurement site. The core of the environmental compensation model in Step 2 is a pre-calibrated "environmental sensitivity matrix." Each row of this matrix corresponds to an environmental characteristic parameter dimension, and each column corresponds to a spectral channel of a multispectral sensor. Each element value in the matrix represents the systematic impact on the measured value of a specific spectral channel when a unit change in a specific environmental parameter occurs. This matrix is ​​obtained by systematically changing various environmental parameters and recording the sensor response changes in a controlled laboratory, and then accurately solving it through regression analysis or system identification methods. After inputting environmental characteristic parameters, the model prepares to infer based on the causal laws implied in this matrix. This provides a stable and computable mathematical core for the quantitative compensation of environmental interference. The environmental sensitivity matrix encodes the complex response patterns of the device to the external environment into a structured mathematical model. This transforms the compensation process from coarse adjustments based on experience to predictable mathematical calculations based on precise physical calibration. In step 3, the model treats the currently input environmental characteristic parameter values ​​as "perturbations" to the baseline environmental state. Using the environmental sensitivity matrix, the independent interference amount generated by each environmental parameter perturbation on each spectral channel is calculated. For example, the portion of the current ambient light intensity exceeding the baseline value is calculated and multiplied by the sensitivity coefficient of "light intensity - channel n" in the matrix to obtain the interference contribution of that light intensity to channel n. After completing this calculation for all parameters and all channels, the system performs vector synthesis of the interference contributions from all environmental parameters on each spectral channel, forming a "multidimensional environmental interference vector". Each component of this vector represents the predicted total interference amount corresponding to that spectral channel under the current comprehensive environment. This achieves refined, channel-level prediction and deconstruction of environmental interference. This step not only provides the overall magnitude of the interference, but more importantly, clearly reveals the specific distribution of these interferences across different spectral bands. This generates a high-fidelity "interference spectrum," providing a precise blueprint for subsequent targeted and differentiated spectral compensation. The environmental interference vector generated in step 4 directly describes the predicted "quantity" and "shape" of the interference. The mapping analysis step aims to transform this interference description into correction instructions that can be practically applied to the original color data, namely, the second compensation parameter.This mapping could be a direct mathematical inverse operation, such as calculating a correction vector opposite to the interference vector; or it could be a more complex functional relationship, such as mapping the interference vector to the optimal set of correction coefficients through a lookup table or nonlinear function, taking into account the nonlinearity of the detector response. The ultimate goal is to maximize the cancellation of predicted environmental interference when this second compensation parameter is applied. This completes the crucial transition from "interference diagnosis" to "correction prescription." This step ensures that the intelligently calculated interference prediction can be effectively and executablely translated into specific actions driving data calibration. The final output second compensation parameter is an operational instruction that can be directly embedded into the calibration process, enabling closed-loop completion of environmental compensation.

[0026] In one embodiment, the step of driving an update to the device feature vector when a preset condition is met includes: Based on the environmental data during this measurement, the environmental context categories of the environmental data are analyzed. The environmental context categories include the baseline environmental state and the non-baseline disturbance state. Based on the analyzed environmental context category, an updated weighting factor is assigned to the random difference component, including: when the environmental context category is the baseline environmental state, a first weighting factor is assigned; when the environmental context category is a non-baseline perturbation state, a second weighting factor smaller than the first weighting factor is assigned. The random difference components mentioned in this study are weighted according to the assigned update weighting factor and then accumulated into the historical perturbation sequence. When the statistical characteristics of the historical disturbance sequence meet the preset update triggering conditions, an adjustment amount is generated to update the device characteristic vector.

[0027] As mentioned above, Step 1 introduces a crucial decision-making basis for equipment feature updates: environmental risk assessment. Based on preset classification rules, pattern recognition and state assessment are performed on real-time environmental data. For example, by determining whether the equipment's operating temperature is within the stable range specified by the manufacturer, and whether the intensity and spectral fluctuations of ambient light are below a certain quiet threshold, the current environment is classified into a "baseline environmental state" or a "non-baseline disturbance state." The baseline environmental state refers to operating conditions that are close to the equipment's factory calibration conditions, with stable and predictable environmental factors; the non-baseline disturbance state covers complex operating conditions such as rapid temperature changes, strong stray light, or mechanical vibrations that may induce reversible or transient performance changes in the equipment. This endows the update mechanism with "contextual awareness." It enables the system to recognize whether the current measurement environment is sufficiently "clean" and "reliable," thus providing a prerequisite for deciding whether to trust the measurement data and allow it to be used to update the equipment's core identity model (equipment characteristic vector). This effectively prevents the collection of unreliable data during periods of severe environmental fluctuation. Step 2, after completing the environmental classification, executes a differentiated trust assignment strategy. When the environment is determined to be in a "baseline environmental state," the likelihood of the current measurement being affected by external anomalies is considered low. In this case, the "random difference component" separated from the data is more likely to accurately reflect the device's instantaneous noise characteristics or minor disturbances, thus it is assigned a higher "first weight factor," allowing it to participate in subsequent accumulation with a larger proportion. Conversely, if the environment is determined to be in a "non-baseline disturbance state," the random difference component is considered to likely contain a large number of disturbance signals induced by the external environment and not inherent to the device. In this case, it is assigned a significantly reduced "second weight factor," aiming to greatly suppress the data influence that may "contaminate" the device's feature vector. This implements data filtering based on environmental risk. By dynamically adjusting the weights, the system purifies the information flow used to learn device characteristics at the data source. In a favorable environment, it actively learns; in a harsh environment, it cautiously observes or even temporarily blocks information. Essentially, this installs a "smart valve" controlled by the environmental state in the device feature vector update process, fundamentally avoiding the risk of mistakenly treating temporary changes as inherent characteristics of the device when it experiences reversible temporary drift due to drastic environmental changes. In step 3, the random variance components obtained from each measurement, weighted according to the environmental context, are not immediately used for updates. Instead, they are sequentially added to a continuously growing historical data sequence. This "historical disturbance sequence" constitutes a time window for observing the long-term statistical trends of the device's instantaneous behavior. The accumulation process is an act of integrating the weighted signal over time, designed to smooth out the randomness of individual measurements and highlight long-term, trend-based patterns of change. This bases update decisions on long-term statistical evidence rather than single events.The accumulation mechanism prevents the system from overreacting to single abnormal fluctuations, requiring any adjustments to the core characteristics of the equipment to be based on statistical patterns consistently observed over a period of time. This enhances the robustness and noise resistance of update decisions, ensuring that only those trends that truly reflect the slow, stable evolution of the equipment are ultimately adopted. Step 4 continuously monitors the statistical characteristics of historical disturbance sequences, such as long-term drift of the sequence mean, systematic increasing trends in variance, or changes in higher-order moments. These statistical characteristics are used to determine whether the noise floor or transient response characteristics of the equipment have undergone real and stable changes. The preset update trigger conditions are a set of strict statistical test thresholds. Only when the statistical characteristics of the sequence exceed these thresholds, indicating that the change is sufficiently significant and persistent, is it determined that the equipment characteristics have indeed evolved, and a small, cautious "adjustment amount" is calculated accordingly. This adjustment amount is ultimately used to fine-tune the equipment characteristic vector to keep it up-to-date. This achieves a prudent model self-evolution based on rigorous statistical inference. This step elevates the updating of the equipment characteristic vector from a potentially noise-driven passive response to an evidence-based proactive decision. It ensures that updates are objective, cautious, and highly reliable, triggering only the minimum necessary adjustments when accumulated statistical evidence strongly indicates a non-negligible drift in device characteristics. This prevents a conflict between the perceived "stability" and the reality of being misled by noise or temporary disturbances.

[0028] In this embodiment, traditional solutions either ignore the long-term evolution of device characteristics or use simple time averaging or fixed-period updates, making it difficult to address the "confusion in device identity recognition caused by drastic environmental changes" mentioned in the background art. This means that reversible temporary drifts caused by devices in extreme environments are incorrectly "learned" and solidified, contaminating their long-term identity model. This embodiment solves this problem by introducing a dual gating mechanism of "environmental context awareness" and "statistical evidence-driven." The entire process forms a decision chain of "environmental risk assessment - data trust weighting - long-term evidence accumulation - rigorous statistical triggering." It not only allows device feature vectors to adaptively learn their true, slow aging, but more importantly, it establishes a robust immune system: through environmental context classification and weight allocation, it isolates "bad data" under high environmental risks at the data entry point; through long-term accumulation and statistical testing, it ensures the rigor and reliability of updates at the decision exit point.

[0029] In one specific embodiment, multispectral sensors often experience harsh environments in industrial settings. For example, a device used for online color detection in an automotive paint shop may move between a room-temperature workshop and a high-temperature baking area; or a device used for outdoor building material testing may experience drastic temperature differences between day and night and between seasons. Under such extreme high and low temperature cycling scenarios, the optical filters inside the sensor will experience a temporary shift of several nanometers in their center wavelength due to thermal expansion and contraction, and the detector response will also change with the junction temperature. These physical changes directly cause rapid, significant, but partially reversible drift in the spectral measurement data. If a traditional update mechanism without environmental awareness is used, when the device moves from a low-temperature environment to a high-temperature environment and immediately performs measurements, the aforementioned thermally induced drift will be detected by the algorithm and is likely to be classified as a "stable" change in device performance. If the device's feature vector (i.e., its digital identity) is updated accordingly, then this identity model, which includes the thermally induced temporary drift, is contaminated. When the device returns to normal temperature, the optical components are restored to their original state, but the contaminated feature vectors remain. This causes the system to compensate based on the wrong identity, which in turn introduces continuous errors, resulting in long-term instability and cross-device data chaos.

[0030] To mitigate the aforementioned scenarios, this solution addresses the drastic changes in temperature sensor data when equipment enters a high-temperature workshop. Analyzing this data immediately categorizes the current environmental context as a "non-baseline disturbance state." This risk identification step signifies that the current measurement is in a high-uncertainty environment. Based on this determination, a very low second weighting factor (e.g., close to 0) is assigned to the "random difference component" isolated from this measurement. This means that although the measurement data may contain significant thermal drift signals, the system has marked it as "low confidence" and almost completely prohibits its use in subsequent updates to the equipment feature vector. This is akin to adding a filter to the learning process. Only when the equipment remains in a stable "baseline environmental state" (e.g., a normal-temperature workshop) for an extended period will the measurement data be assigned a higher first weighting factor, allowing it to accumulate in the historical sequence with normal weight. The statistical characteristics (e.g., slow mean drift) of these disturbance data accumulated in favorable environments are observed over a long window (e.g., several days or weeks). An update is only triggered when the trend is sufficiently significant and stable. This ensures that the evolution of the device's feature vectors only responds to real, slow hardware aging, such as the permanent decay of LED light sources, rather than temporary fluctuations caused by the environment.

[0031] It's worth noting that in high-end manufacturing or scientific research settings, the operating environment for multispectral sensors can be quite unique. For example, near intelligent production lines using lasers for positioning or cutting, sensors detecting workpiece color might be exposed to intense laser reflections or scattered light; or in optical laboratories, equipment used for spectral analysis of artifacts might be affected by high-purity monochromatic light leaking from nearby monochromators. In such scenarios, ambient light is not the common sunlight or broadband continuous light emitted by lamps, but rather high-intensity, extremely narrow-spectral-width monochromatic light. This narrow-band interference light physically superimposes with the sensor's own light source in the detector. Since the sensor is not an interferometer and cannot distinguish the source of photons, its output signal will be a mixture of the broadband reflected signal from the device's light source and the ambient monochromatic light signal. This leads to a fundamental problem: all subsequent algorithms are based on a flawed assumption—that the measurement signal originates solely from the device's own light source. Traditional environmental compensation models based on broadband ambient light calibration struggle to understand and correct this particular type of interference because it exceeds their modeling scope. More seriously, such strong interference signals with fixed spectral positions can be misjudged by subsequent analysis algorithms and may be mistaken for a specific, stable change in the spectral response of the device, leading to incorrect calibration and model contamination, and causing the entire system to fail.

[0032] This solution proposes an embodiment in which, prior to the step of extracting multi-dimensional environmental feature parameters from the environmental data, the method further includes: Obtain the ambient light spectral information from the environmental data; Based on the ambient light spectral information in the environmental data, determine whether narrowband interference light exists in the ambient light spectral information; If the ambient light spectral information does not contain narrowband interference light, then continue to extract environmental feature parameters from the environmental data in multiple dimensions; If the ambient light spectral information contains narrowband interference light, an additional spectral channel filtering parameter is generated and applied based on the characteristics of the narrowband interference light to preprocess the interfered spectral channel before the original color data enters the calibration.

[0033] As mentioned above, Step 1 is a deepening and upgrade of "acquiring environmental data." It requires that the spectral characteristics of the ambient light itself be specifically collected before or simultaneously with the measurement of the target object. This can be achieved in two ways: first, by using a multispectral sensor itself, with its built-in light source off, to directly perform a one-time spectral scan of the ambient light; second, by using an integrated ambient light spectral sensor with multi-channel or simple spectroscopic functions for continuous monitoring. The goal is to obtain the energy distribution curve of the ambient light at various wavelengths, not just a total intensity value. This achieves a leap from "brightness perception" to "spectral fingerprint recognition" of ambient light. Acquiring spectral information allows the system to understand the essential composition of ambient light, such as whether it is continuous-spectrum sunlight, broadband LED light, or laser or monochromatic LED light with energy concentrated in an extremely narrow band. This is the data foundation for subsequent advanced interference identification, which traditional methods relying solely on illuminance meters cannot provide. Step 2 is an intelligent diagnostic process. The acquired ambient light spectral data is analyzed, using algorithms such as peak detection, full width at half maximum (FWHM) calculation, and signal-to-noise ratio (SNR) evaluation to find the presence of one or more spectral peaks. The judgment criteria are based on preset physical thresholds: First, the peak intensity must be significantly higher than the average level of spectral background noise; second, its spectral width must be less than a set narrowband threshold (e.g., less than 20 nanometers). Spectral peaks that simultaneously meet the characteristics of high intensity and narrow bandwidth are judged as "narrowband interference light." This light typically originates from laser pointers, monochrome LED indicators, signal light sources of specific wavelengths, etc. This gives the system the special ability to identify "unconventional environmental interference." It can proactively detect and warn of special threats that traditional environmental compensation models (usually designed for broadband continuous light) cannot effectively handle. Step 3: When the ambient light spectrum is judged to be normal (without unwanted narrowband spikes), the current environment is determined to be within the range of conventional interference. At this time, trusting and enabling a mature environmental compensation model designed for conventional broadband interference is the optimal choice. Therefore, the process jumps back to extracting conventional characteristic parameters such as light intensity and color temperature from the environmental data and inputting them into the environmental compensation model for calculation. This ensures the system's high efficiency and optimal resource utilization in most conventional environments. It avoids activating additional complex processing procedures when there is no specific threat, allowing valuable computing resources to be concentrated on high-precision routine compensation calculations, maintaining a balance between overall system performance and efficiency. Step 4: Once the presence of narrowband interference light is confirmed, the emergency response path is immediately activated. First, based on the center wavelength and bandwidth of the narrowband light, the detectors of which spectral channels in the multispectral sensor will receive the interference signal are precisely identified. Then, a "spectral channel filtering parameter" is generated, which is essentially a targeted set of repair instructions. The purpose is to clean or reconstruct the data of the contaminated channels before the original color data officially enters the calibration pipeline, for example, by marking it as invalid or replacing it with interpolated data from adjacent channels. "Isolation and cleanup" are implemented at the data source to prevent specific interference from contaminating the entire system.This is key to resolving the defects. Through targeted preprocessing, deceptive narrowband interference signals are stripped or neutralized from the raw data, ensuring that all subsequent steps based on difference analysis and model calculations, such as calculating the first compensation parameter and updating the device eigenvectors, are clean. The received input data is "clean." This avoids narrowband interference being incorrectly decomposed into "stable difference components" or "environmental interference vectors," thus preventing systematic miscompensation and model contamination.

[0034] In this embodiment, designed to be immune to the aforementioned special light pollution scenarios, the system instructs the sensor (with its own light source turned off) to perform a rapid "spectral scan" of the ambient light before formal measurement. By analyzing this spectral data, the system can immediately identify the presence of abnormal, sharp narrow-band spectral peaks. For example, the discovery of a peak with extremely high intensity and a width of less than 5 nm at a wavelength of 650 nm indicates the presence of interference from red laser indicator light. This step is a crucial pre-diagnosis, clarifying that the current threat is unconventional. Different strategies are implemented based on the diagnostic results. If no narrow-band interference light is detected, the system determines the environment is normal and continues using the efficient and accurate conventional environmental compensation model process—this is the optimal path. Once narrow-band interference light, such as the aforementioned 650 nm laser, is confirmed, the system switches to anti-interference mode. In this mode, the conventional environmental compensation model is bypassed because it is ineffective against this type of interference. In anti-interference mode, the affected sensor spectral channels are accurately calculated based on the identified narrow-band light characteristics (center wavelength 650 nm, extremely narrow bandwidth). Subsequently, a "spectral channel filtering parameter" is generated. This parameter is essentially a set of instructions used to preprocess the "raw color data" that is about to enter the core calibration process. For example, raw measurement data with contamination in the channel near 650nm is marked as unreliable, and the data from the unaffected channels on both sides are used to calculate the appropriate, uncontaminated replacement data for that channel using an intelligent interpolation algorithm.

[0035] In one embodiment, generating and applying an additional spectral channel filtering parameter based on the characteristics of the narrowband interference light to preprocess the interfered spectral channel before the original color data enters calibration specifically includes: By analyzing the characteristics of the narrowband interference light, the center wavelength and bandwidth of the narrowband interference light are obtained; Based on the center wavelength and bandwidth of the narrowband interference light, analyze the state of each spectral channel of the multispectral sensor; Based on the analysis results, the states of each spectral channel of the multispectral sensor are distinguished into the target spectral channel that is disturbed and the adjacent spectral channels that are not disturbed. Based on the data of undisturbed adjacent spectral channels, the alternative data of the target spectral channel is calculated by an interpolation algorithm; The original measurement data for the corresponding target spectral channel in the original color data is replaced with the alternative data, and a data confidence mark is added to the replaced channel.

[0036] As described above, Step 1 involves precise quantitative analysis of the identified narrowband interference light. By performing local peak fitting or waveform analysis on the ambient light spectral data, the precise location of the anomalous spectral peak, i.e., the center wavelength (e.g., 635.2 nm), is calculated. Simultaneously, by analyzing the wavelength range corresponding to the peak intensity decreasing to half, its spectral bandwidth is calculated (e.g., a full width at half maximum (FWHM) of 3 nm). These two parameters are core characteristics describing the physical nature of narrowband light. The qualitative judgment of "the existence of narrowband light" is transformed into precise physical parameters usable for subsequent fine-tuning. The center wavelength determines the precise location of the focusing interference energy, and the bandwidth determines the spectral range of its influence. This provides irrefutable data for the next step of accurately locating the affected sensor channel. Step 2: Each detection channel of the multispectral sensor corresponds to a known spectral response function, which defines the channel's sensitivity to different wavelengths of light. This step convolves or calculates the overlap between the characteristic parameters of the narrowband interference light and the spectral response function of each channel. A mathematical model is used to evaluate the degree of matching between the response function of each channel and the narrowband interference spectrum, thereby quantifying the potential risk of interference to that channel. This achieves a scientific mapping from abstract physical parameters to specific hardware impacts. Through calculation, the system can precisely determine that for a laser with a center wavelength of 635.2nm and a bandwidth of 3nm, channels 15 and 16 of the sensor, assuming their center wavelengths are 634nm and 637nm respectively, will be strongly affected, while channels 14 and 17 may be slightly affected, and other channels will be largely unaffected. Step 3, based on the interference risk quantified in Step 2, applies a preset threshold for binary decision-making. Channels with interference risk values ​​exceeding the threshold are classified as "interfered target spectral channels," and their original measurement data are considered unreliable. Simultaneously, channels on either side of the target channel with interference risk values ​​below the threshold and adjacent spectral positions are selected and classified as "uninterrupted adjacent spectral channels." The data from these channels are considered relatively clean and reliable and can be used as a reference for repairing damaged channels. Tactical grouping of all sensor channels is completed, clearly distinguishing between "wounded personnel requiring repair" and "healthy units that can provide assistance." This distinction is a prerequisite for implementing minimal and precise processing, avoiding a "one-size-fits-all" approach to the entire spectrum, maximizing the protection of undisturbed original data, and ensuring the efficiency and low side effects of the repair operation. Step 4 is the core step of data repair. The system utilizes the physical characteristic of the object's reflectance spectrum having continuous and smooth variations. Using the measurements of undisturbed adjacent channels as known data points, an interpolation function is constructed regarding wavelength and signal intensity, such as linear interpolation, polynomial interpolation, or spline interpolation. Then, the center wavelength of the target channel to be repaired is substituted into this interpolation function to calculate a predicted alternative value that excludes narrowband light interference. This value represents the original spectral response of the object at that wavelength. This achieves intelligent reconstruction and purification of contaminated data.Compared to simply zeroing out or averaging the data in the affected channel, interpolation algorithms based on spectral continuity can estimate the true signal more scientifically and accurately. It effectively eliminates the anomalous energy added by narrowband interference light while preserving the spectral characteristics of the object itself as much as possible, minimizing information loss and providing high-quality input for subsequent calibration. Step 5 writes the cleaned replacement data calculated in Step 4 into the original color data array, overwriting the original measurements contaminated by narrowband light. Simultaneously, the system sets a special "data confidence flag" in the data metadata of this channel. This flag can be a flag indicating that the data is "interpolated repair data," or a confidence score reflecting the degree of uncertainty in this repair. This completes the data-level purification and information preservation. The replacement operation directly produces a "clean" spectral data set, allowing it to safely enter the core calibration process. Adding the flag provides crucial metadata information, informing all downstream algorithms that this channel data is not directly measured but is a repaired product. This allows downstream algorithms to adjust their processing strategies accordingly. For example, when calculating the weights for updating device feature vectors, the reliance on repair channel data can be appropriately reduced, thereby avoiding treating "soft data" based on interpolation estimation as the same as "hard data" based on actual measurements, thus improving the rigor of the entire system logic and the reliability of the final results.

[0037] In one specific embodiment, in industrial settings such as food cold chain inspection or metal heat treatment, multispectral sensors often need to be rapidly moved and immediately operational between environments with extreme temperature differences. For example, a sensor used to detect the color of frozen food packaging is taken from a -18°C cold storage and quickly moved to a 25°C ambient temperature sorting station for measurement. During this process, the temperature of the sensor housing and internal optical components does not change synchronously and uniformly: the semiconductor junction temperature of the LED light source and the temperature of the silicon-based detector rise rapidly at a rate different from that of the ambient temperature sensor. At this moment, if calibration is performed using a traditional static model, it will lead to a misaligned reference system: the "device characteristic vector" used for analysis is based on the static value calibrated by the device at a stable temperature. However, during the rapid transition from low to high temperatures, the actual performance of the device (such as LED luminous efficiency and detector responsivity) is constantly changing on a second-by-second basis. Using static characteristics to analyze the measurement signal generated during this dynamic change will result in a complete misalignment of the reference system. In the aforementioned thermal transients, the changes in the measurement signal mainly originate from the thermal drift of the device's own performance, rather than changes in the color of the measured object. Attempting to compare and separate these changes with the static reference vector will lead to a misalignment. Due to the error in the reference frame, this separation will inevitably fail. A large number of thermal transient signals will be misunderstood by the algorithm. Some may be treated as "environmental interference" (but the readings of the ambient temperature sensor may be lagging), and more fatally, another part will be mistakenly identified as a "stable" performance change of the device, thereby contaminating the "stable difference component" used to generate the first compensation parameter, and may even trigger an incorrect device identity update.

[0038] To address the dynamic reference frame mismatch problem, this embodiment, before the step of coupling the original spectral difference vector with the device characteristic vector for analysis, further includes dynamic state estimation of the device characteristic vector. The steps include: The real-time temperature sequence of the multispectral sensor during this measurement process is obtained, and thermal state parameters characterizing its thermal transient changes are extracted. The thermal state parameters and the equipment characteristic vector are combined and input into a preset equipment thermal response model. The dynamic equipment characteristic vector reflecting the spectral response characteristics of the equipment under the current instantaneous thermal state is calculated through the equipment thermal response model. The dynamic device characteristic vector is used to replace the original device characteristic vector, and a coupling analysis is performed between the original spectral difference vector and the device characteristic vector.

[0039] As mentioned above, in step 1, under rapid temperature change scenarios, the internal temperature distribution of the device is uneven and changes over time. A single instantaneous temperature value is insufficient to describe this transient state. This step involves collecting temperature sequences over a period before and after the measurement begins. If the sampling frequency is higher than the measurement cycle, "thermal state parameters" such as the rate of temperature change and the estimated thermal equilibrium difference within the sensor are extracted. This aims to quantify the dynamic process of "how the device is getting hotter or colder." The acquired "environmental data" (including temperature) is utilized, but a more in-depth time-series analysis and feature extraction are performed, transforming it into characteristic quantities describing the device's thermal inertia and transient state, providing input for dynamic modeling. Step 2, the device thermal response model, is a pre-calibrated mathematical model that describes how the spectral response characteristics of the specific device (i.e., what is represented by the "device characteristic vector") dynamically change with thermal state parameters (not just temperature values). For example, the model may include coefficients for the wavelength changes of different spectral channels with temperature, the relationship between detector response efficiency and junction temperature, etc. The model receives a static "device characteristic vector" as a baseline and real-time "thermal state parameters," and outputs a "dynamic device characteristic vector" that accurately predicts the actual spectral response characteristics of the device under the current thermal transient state. This solves the problem of "device characteristic vector instability during measurement." Through the thermal response model, a static, long-term identity label is corrected in real-time to a dynamic, instantaneous state snapshot. This ensures that the reference for subsequent analysis is no longer the "outdated" static vector, but a dynamic vector that matches the actual physical state of the device at the current measurement moment. Step 3 clarifies the instruction system, using the newly calculated "dynamic device characteristic vector" in all subsequent analyses involving the "device characteristic vector." This ensures that the mathematical reference frame for the entire difference separation process originates from the same physical moment and device state as the current measurement signal, thus restoring the effectiveness of the separation logic. Through replacement, most of the effects of the instantaneous hardware drift caused by drastic temperature changes are absorbed into the "dynamic device characteristic vector." Therefore, during coupling analysis, the difference between the "original spectral difference vector" and this dynamic vector will reveal more of the true long-term intrinsic drift not caused by temperature. This allows for a more accurate separation of rapid temperature-induced changes into the "random difference component," or a significant reduction in its accuracy during comparison. This effectively prevents temperature transients from being incorrectly classified into the "stable difference component," thus protecting the purity of the device's long-term identity model (device characteristic vector) and ensuring the accuracy of the first compensation parameter.

[0040] In this embodiment, when the sensor is moved to a normal temperature environment, it not only reads the current ambient temperature, such as 25°C, but also initiates a high-speed sampling process to continuously record the reading sequence of the internal temperature sensor within hundreds of milliseconds before and after the measurement begins. By analyzing this sequence, the system calculates the current "thermal state parameters" of the device, such as the instantaneous rate of temperature change, such as 20°C / minute, and the estimated internal thermal balance difference. These parameters describe the dynamic thermal state of the device, which is "rapidly heating up and not yet internally balanced." To predict the device performance "at this moment" in real time, the system inputs the aforementioned "thermal state parameters" along with the stored static "device characteristic vector" into a pre-calibrated device thermal response model. This model contains the precise mathematical relationship between the performance of the sensor's LED and detector and the temperature and the rate of temperature change. The model calculates and outputs a dynamic device characteristic vector in real time. This new vector is no longer a fixed value, but a precise prediction of "the actual spectral response characteristics of the device under the current rapid heating transient state at the instant the shutter of this measurement is opened." Switching to the correct dynamic reference frame: When performing subsequent coupling analysis and separating stable and random differential components, the outdated, static device characteristic vector is no longer used. Instead, the newly generated dynamic device characteristic vector is used as the basis for analysis. This means that the algorithm is using the "performance model that the device should have at this moment" to analyze the "signal measured at this moment".

[0041] Reference Figure 3 This application also provides a multispectral color data calibration system based on multi-source sensing data fusion, comprising: Acquisition module 1 is used to acquire the raw color data and environmental data measured by the multispectral sensor on the target object; Judgment module 2 is used to determine whether the original color data meets the calibration conditions; Module 3 is invoked to invoke the primary reference color data corresponding to the multispectral sensor if the original color data meets the calibration conditions. Calculation module 4 is used to compare the original color data with the main reference color data and calculate a first compensation parameter for correcting inherent differences between devices. Input module 5 is used to input environmental data into a preset environmental compensation model and output a second compensation parameter through the environmental compensation model. The second compensation parameter is used to correct environmental interference values. The generation module 6 is used to perform fusion calibration on the original color data using the first compensation parameter and the second compensation parameter to generate calibrated color data.

[0042] As described above, it is understood that each component of the multispectral color data calibration system based on multi-source sensing data fusion proposed in this application can realize the function of any of the multispectral color data calibration methods based on multi-source sensing data fusion described above, and the specific structure will not be repeated.

[0043] Reference Figure 4 This application also provides a computer device, which may be a server, and its internal structure may be as follows: Figure 4 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores monitoring data and other data. The network interface is used for communication with external terminals via a network connection. When executed by the processor, the computer program implements a multispectral color data calibration method based on multi-source sensing data fusion.

[0044] The processor described above executes the multispectral color data calibration method based on multi-source sensing data fusion, comprising: acquiring raw color data and environmental data measured by a multispectral sensor on a target object; determining whether the raw color data meets the calibration conditions; if the raw color data meets the calibration conditions, calling the master reference color data corresponding to the multispectral sensor; comparing the raw color data with the master reference color data to calculate a first compensation parameter for correcting inherent differences between devices; inputting environmental data into a preset environmental compensation model, and outputting a second compensation parameter through the environmental compensation model, the second compensation parameter being used to correct environmental interference values; and using the first compensation parameter and the second compensation parameter to perform fusion calibration on the raw color data to generate calibrated color data.

[0045] One embodiment of this application also provides a computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements a multispectral color data calibration method based on multi-source sensing data fusion, including the following steps: acquiring raw color data and environmental data measured by a multispectral sensor on a target object; determining whether the raw color data meets the calibration conditions; if the raw color data meets the calibration conditions, calling the master reference color data corresponding to the multispectral sensor; comparing the raw color data with the master reference color data to calculate a first compensation parameter for correcting inherent differences between devices; inputting environmental data into a preset environmental compensation model, and outputting a second compensation parameter through the environmental compensation model, the second compensation parameter being used to correct environmental interference values; and using the first compensation parameter and the second compensation parameter to perform fusion calibration on the raw color data to generate calibrated color data.

[0046] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in this application and in the embodiments can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual-speed SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0047] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, apparatus, article, or method. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, apparatus, article, or method that includes that element.

[0048] The above description is only a preferred embodiment of this application and does not limit the patent scope of this application. Any equivalent structural or procedural changes made based on the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A multispectral color data calibration method based on multi-source sensing data fusion, characterized in that, The method includes: Acquire raw color data and environmental data of the target object measured by the multispectral sensor; Determine whether the original color data meets the calibration conditions; If the original color data meets the calibration conditions, the master reference color data corresponding to the multispectral sensor is called. The original color data is compared with the main reference color data to calculate the first compensation parameter used to correct the inherent differences between devices; Environmental data is input into a preset environmental compensation model, and a second compensation parameter is output through the environmental compensation model. The second compensation parameter is used to correct environmental disturbance values. The original color data is fused and calibrated using the first compensation parameter and the second compensation parameter to generate calibrated color data.

2. The multispectral color data calibration method based on multi-source sensing data fusion according to claim 1, characterized in that, The step of inputting environmental data into a preset environmental compensation model and outputting a second compensation parameter through the environmental compensation model includes: Multiple environmental feature parameters are extracted from the environmental data, including ambient light intensity, ambient light spectral distribution characteristics, and equipment operating temperature. The environmental feature parameters of the multiple dimensions are input into the environmental compensation model, which includes an environmental sensitivity matrix corresponding to the multispectral sensor. Based on the environmental sensitivity matrix, the environmental compensation model calculates the interference weight of each dimension's environmental characteristic parameter on each spectral channel, and then generates a multi-dimensional environmental interference vector. Based on the output results, a mapping analysis is performed on the multidimensional environmental interference vector to determine the second compensation parameter.

3. The multispectral color data calibration method based on multi-source sensing data fusion according to claim 2, characterized in that, Prior to the step of extracting multi-dimensional environmental feature parameters from the environmental data, the method further includes: Obtain the ambient light spectral information from the environmental data; Based on the ambient light spectral information in the environmental data, determine whether narrowband interference light exists in the ambient light spectral information; If the ambient light spectral information does not contain narrowband interference light, then continue to extract environmental feature parameters from the environmental data in multiple dimensions; If the ambient light spectral information contains narrowband interference light, an additional spectral channel filtering parameter is generated and applied based on the characteristics of the narrowband interference light to preprocess the interfered spectral channel before the original color data enters the calibration.

4. The multispectral color data calibration method based on multi-source sensing data fusion according to claim 1, characterized in that, The step of comparing the original color data with the master reference color data to calculate a first compensation parameter for correcting inherent differences between devices includes: Based on the device identifier of the multispectral sensor, the corresponding device characteristic vector is obtained, and the device characteristic vector characterizes the spectral response tolerance of the multispectral sensor in the manufacturing process; Based on the original color data and the main reference color data, calculate the original spectral difference vector; The original spectral difference vector is coupled with the device characteristic vector for analysis; Based on the results of the coupling analysis, a stable difference component contributed by the inherent system deviation of the equipment and a random difference component contributed by the current instantaneous state are separated. The random difference component is used for accumulation processing to drive the update of the equipment feature vector when a preset condition is met. The first compensation parameter is generated based on the stable difference component.

5. The multispectral color data calibration method based on multi-source sensing data fusion according to claim 4, characterized in that, The step of calculating the original spectral difference vector based on the original color data and the master reference color data includes: The original color data and the primary reference color data are respectively converted into a first spectral vector and a second spectral vector with the same dimension; Call the spectral vector space, and calculate the channel-by-channel relative deviation vector between the first spectral vector and the second spectral vector in the spectral vector space; Based on the calculation results, the channel-by-channel relative deviation vector is normalized to obtain a standardized spectral difference vector. The spectral difference vector is multiplied by a pre-stored device spectral sensitivity feature vector to obtain a weighted spectral difference vector, which is then used as the original spectral difference vector.

6. The multispectral color data calibration method based on multi-source sensing data fusion according to claim 4, characterized in that, The step of driving the update of the device feature vector when the preset conditions are met includes: Based on the environmental data during this measurement, the environmental context categories of the environmental data are analyzed. The environmental context categories include the baseline environmental state and the non-baseline disturbance state. Based on the analyzed environmental context category, an updated weighting factor is assigned to the random difference component, including: when the environmental context category is the baseline environmental state, a first weighting factor is assigned; when the environmental context category is a non-baseline perturbation state, a second weighting factor smaller than the first weighting factor is assigned. The random difference components mentioned in this study are weighted according to the assigned update weighting factor and then accumulated into the historical perturbation sequence. When the statistical characteristics of the historical disturbance sequence meet the preset update triggering conditions, an adjustment amount is generated to update the device characteristic vector.

7. The multispectral color data calibration method based on multi-source sensing data fusion according to claim 3, characterized in that, Based on the characteristics of the narrowband interference light, an additional spectral channel filtering parameter is generated and applied to preprocess the interfered spectral channels before the original color data enters calibration. Specifically, this includes: By analyzing the characteristics of the narrowband interference light, the center wavelength and bandwidth of the narrowband interference light are obtained; Based on the center wavelength and bandwidth of the narrowband interference light, the state of each spectral channel of the multispectral sensor is analyzed; Based on the analysis results, the states of each spectral channel of the multispectral sensor are distinguished into the target spectral channel that is disturbed and the adjacent spectral channels that are not disturbed. Based on the data of undisturbed adjacent spectral channels, the alternative data of the target spectral channel is calculated by an interpolation algorithm; The original measurement data for the corresponding target spectral channel in the original color data is replaced with the alternative data, and a data confidence mark is added to the replaced channel.

8. A multispectral color data calibration system based on multi-source sensing data fusion, characterized in that, include: The acquisition module is used to acquire the raw color data and environmental data measured by the multispectral sensor on the target object; The judgment module is used to determine whether the original color data meets the calibration conditions; The calling module is used to call the master reference color data corresponding to the multispectral sensor when the original color data meets the calibration conditions; The calculation module is used to compare the original color data with the main reference color data and calculate a first compensation parameter for correcting the inherent differences between devices. The input module is used to input environmental data into a preset environmental compensation model, and output a second compensation parameter through the environmental compensation model. The second compensation parameter is used to correct the environmental disturbance value. The generation module is used to perform fusion calibration on the original color data using the first compensation parameter and the second compensation parameter to generate calibrated color data.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images