New material antiquing trace detection method combined with fluorescence spectrum identification

Abstract

The invention discloses a new material antiquing trace detection method combined with fluorescence spectrum identification, and relates to the technical field of material detection. The method comprises the steps of performing intelligent analysis on a surface image of a to-be-detected material, identifying a suspicious distressing area, predicting distressing type probability distribution, and evaluating a first distressing detection complexity index and a second distressing detection complexity index; meanwhile, a natural aging standard sample library is retrieved based on material attributes, and a third distressing detection complex index is evaluated; carrying out weighted fusion on the three indexes to obtain overall distressing detection complexity; and adaptively configuring an optimal fluorescence spectrum identification scheme according to the overall distressing detection complexity, and executing detection. According to the method, objective quantification and adaptive optimization of the detection process are realized, the defects that a traditional method depends on subjective experience, is low in efficiency and is difficult to deal with complex distressing scenes are effectively overcome, and the identification precision, efficiency and automation level are improved.

Description

Technical Field

[0001] This invention relates to the field of materials testing technology, specifically to a novel method for detecting antique-style aging traces in materials by combining fluorescence spectroscopy identification. Background Technology

[0002] In the fields of cultural relic preservation and art authentication, the detection of signs of aging and counterfeiting is a crucial technology. With the widespread application of new materials, the practice of using traditional methods such as chemical dyeing, acid etching, and baking to create artificial antiquities is increasing, posing a challenge to the identification of genuine and fake items.

[0003] Current technologies for detecting antiquing traces on materials primarily rely on visual observation, microscopic analysis, or single spectroscopic methods, which have significant limitations: visual observation and microscopic analysis are highly dependent on the operator's subjective experience, have poor repeatability, and cannot be quantitatively assessed; while single fluorescence or infrared spectroscopy techniques can provide information on material composition, they usually require pre-setting detection parameters and lack adaptability to the complexity of the object being detected. This results in insufficient accuracy and reliability of detection schemes when facing unknown or mixed antiquing traces, leading to wasted resources or missed detections and misjudgments. Furthermore, existing methods generally lack a framework for the collaborative analysis of macroscopic image features and microscopic spectral data, failing to effectively integrate multi-source information to achieve an intelligent detection process from rapid screening to in-depth diagnosis. Summary of the Invention

[0004] This invention addresses the technical problems of existing technologies, such as reliance on subjective experience, low detection efficiency, lack of adaptability, and difficulty in accurately identifying complex and diverse antique-style traces. It provides a new method for detecting antique-style traces in materials by combining fluorescence spectroscopy identification.

[0005] The technical solution of the present invention to solve the above-mentioned technical problems is as follows: This invention provides a novel method for detecting antique-style aging traces in materials by combining fluorescence spectroscopy identification, including: The surface image of the material to be tested is used to identify signs of aging, obtain the distribution of suspicious aging areas and the probability distribution of predicted aging types, and evaluate and determine the first and second aging detection complexity indices. Obtain the standard data abundance corresponding to the material to be tested in the pre-constructed natural aging standard sample library, and determine the third aging detection complexity index based on the standard data abundance. The overall aging detection complexity is obtained by weighting and fusing the first aging detection complexity index, the second aging detection complexity index, and the third aging detection complexity index. Based on the overall complexity of the antiquing detection, the optimal fluorescence spectroscopy identification scheme is configured to detect traces of antiquing on the material to be tested.

[0006] The beneficial effects of this invention are: Compared to existing technologies, this invention first achieves objective quantitative analysis of antiquing traces through intelligent image recognition and multi-dimensional complexity index evaluation, effectively reducing reliance on subjective experience. Secondly, by integrating image features, predicted types, and information from a standard sample database, it constructs a comprehensive index reflecting the complexity of the detected object, providing a scientific basis for subsequent detection strategy formulation. Thirdly, based on the overall detection complexity, it adaptively matches different levels of fluorescence spectral identification schemes, achieving precise resource allocation from rapid screening to in-depth analysis, improving detection efficiency and accuracy. Finally, it forms a closed-loop intelligent detection process from macroscopic image analysis to microscopic spectral verification, enhancing the ability to identify complex antiquing techniques on new materials and improving system adaptability. Attached Figure Description

[0007] Figure 1 A flowchart illustrating the novel method for detecting antique-style aging traces in materials, which combines fluorescence spectroscopy identification, provided by this invention. Figure 2 This is a schematic diagram illustrating the construction process of the mapping table for the old detection complexity-fluorescence spectroscopy identification scheme provided by the present invention. Detailed Implementation

[0008] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0009] In the description of this invention, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0010] In the description of this invention, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this invention is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed herein.

[0011] Examples, such as Figure 1 As shown, embodiments of the present invention provide a novel method for detecting antique-style aging traces in materials by combining fluorescence spectroscopy identification, including: S10: Identify signs of aging on the surface image of the material to be tested, obtain the distribution of suspicious aging areas and the probability distribution of predicted aging types, and evaluate and determine the first and second aging detection complexity indices. Specifically, the surface image of the material to be tested is used to identify signs of aging, obtain the distribution of suspected aging areas and the probability distribution of predicted aging types, including: Multi-angle surface image acquisition is performed on the material to be tested to obtain surface images of the material; A material aging trace detector was built based on a convolutional neural network. The material surface image is input into the material aging trace identifier, which outputs the distribution of suspicious aging areas and the probability distribution of predicted aging types.

[0012] First, multi-angle surface image acquisition is performed on the material to be tested. Specifically, this step involves acquiring digital images of the material surface from multiple perspectives to form a complete set of material surface images. The obtained material surface images can comprehensively cover all areas of the material to be tested, providing basic visual data for subsequent automatic identification and analysis.

[0013] Secondly, a material aging trace detector based on a convolutional neural network is constructed. This material aging trace detector is a deep machine learning model trained through supervised learning. Its internal network structure maps a complex nonlinear relationship between the input material surface image and the output aging region and aging type probability. This material aging trace detector can be used to automatically analyze and identify trace areas in images formed by various artificial aging techniques such as chemical staining, acid etching, fire baking, and applying antique paint, and quantitatively assess the probability of each aging type.

[0014] Specifically, a material aging trace detector is constructed based on a convolutional neural network, including: Using the attribute information of the material to be tested as a constraint, a set of sample material surface images is collected. The distribution set of sample aging areas is obtained by manually annotating the aging areas of different sample material surface images. The probability distribution set of sample aging types is obtained by manually annotating the aging types and aging confidence levels of different sample material surface images. The aging types include one or more of the following: chemical dyeing, acid etching, alkali etching, fire baking, smoke smoking, application of antique paint, and adhesive aging. Using the sample material surface image set as input data, and the sample aging region distribution set and sample aging type probability distribution set as supervision labels, a convolutional neural network is trained until convergence to generate a material aging trace recognizer.

[0015] First, the specific properties of the materials to be tested need to serve as constraints for screening and data collection. Specifically, the properties of the materials to be tested can include material categories, such as wood, lacquerware, ceramics, metal, or calligraphy and painting. Under this constraint, a large number of sample materials of the corresponding categories are systematically collected, and their surface images from multiple angles are acquired using high-definition image acquisition equipment, forming a sample material surface image set covering various typical antiquing conditions.

[0016] Secondly, each image of the sample material surface in the image set is manually annotated. Specifically, the annotation work consists of two parts: the first part involves pixel-level or region-level annotation of areas in the sample material surface images that may show signs of artificial aging, clearly delineating the boundaries and locations of suspected aging areas, thus forming a set of sample aging area distributions that correspond one-to-one with the sample material surface image set; the second part involves expert evaluation of the aging techniques used in the identified aging areas, assigning a confidence score to the evaluation results. Specifically, aging techniques include common types such as chemical dyeing, acid etching, alkali etching, fire baking, smoking, applying antique-style coatings, and adhesive aging. Each annotated area will obtain one or more combinations of aging types and their corresponding confidence scores, thereby generating a sample aging type probability distribution set.

[0017] The "aging confidence score" is a quantitative value assigned by a domain expert during the annotation process, representing the expert's degree of certainty that a specific area in the image belongs to the labeled aging type. The sample aging type probability distribution set is a structured dataset that records all possible aging types associated with each labeled area in each sample image and their corresponding confidence scores, with a one-to-one correspondence between aging types and aging confidence scores. This sample aging type probability distribution set accurately reflects the uncertainty and multiple possibilities in identifying aging techniques during the manual authentication process.

[0018] Finally, using the aforementioned set of sample material surface images as input data for the convolutional neural network, and simultaneously using the sample aging region distribution set and sample aging type probability distribution set as target labels for supervised learning, the convolutional neural network is trained until convergence, generating a material aging trace detector. The trained material aging trace detector has the function of automatically outputting the distribution of suspicious aging regions and the probability distribution of aging types for new material surface images to be detected.

[0019] For example, since there is a highly nonlinear and complex correlation between the image features of the material surface and the distribution of the aged area and the probability distribution of the aged type, and since convolutional neural networks have significant advantages in image feature abstraction and spatial pattern recognition, a convolutional neural network is chosen to construct a material aging trace detector.

[0020] Specifically, the material aging trace detector can employ an encoder-decoder structure. The encoder consists of alternating convolutional and pooling layers, used to progressively extract and abstract multi-level visual features from the input image. The decoder, through upsampling operations and skip connections, progressively restores the spatial resolution of the feature map, ultimately outputting a distribution map of suspicious aging areas matching the input image size, as well as a probability distribution vector of aging type at each spatial location. The network introduces non-linearity by employing the ReLU activation function after the convolutional layers and uses Dropout technology during training, with a dropout rate set between 0.3 and 0.5, to suppress overfitting of the model to training samples and enhance its generalization performance.

[0021] During training, key hyperparameters included an initial learning rate of 0.0001, 200 training epochs, and a batch size of 16. The learning rate was set to balance image data complexity with training stability; the number of training epochs ensured the model fully learned the visual patterns of various aging marks; and the batch size balanced memory capacity with gradient update stability. Supervised learning was employed. A set of sample material surface images was used as the input sample set, and the corresponding distribution sets of sample aging regions and probability distribution sets of sample aging types were used as the supervision label set, forming training data pairs. All data was randomly divided into training, validation, and test sets in a 6:2:2 ratio.

[0022] Next, the sample material surface images from the training set are input into a convolutional neural network for forward propagation. The network outputs a predicted region distribution map and type probability distribution. The prediction results are compared with the true supervision labels, and a composite loss function is constructed to measure the bias. This loss function consists of two weighted parts: one part is the Dice loss for region segmentation, which focuses on optimizing the matching degree of the aging region boundary and location; the other part is the cross-entropy loss for multi-label classification, which focuses on optimizing the accuracy of the probability prediction of each aging type. During training, the Adam optimizer is used to iteratively update the network weight parameters based on the backpropagation algorithm. The training process is monitored through a validation set. For example, when the composite loss function value on the validation set no longer decreases significantly for 20 consecutive training epochs, and the IoU index of region segmentation and the average accuracy of classification both reach predetermined thresholds, such as exceeding 90% and 85% respectively, the training process is terminated, and a converged material aging trace detector is obtained. This material aging trace detector can effectively capture subtle aging trace features in material surface images, and simultaneously accurately locate suspicious areas and predict the probability of their aging type, providing a reliable automated analysis basis for subsequent complexity assessment.

[0023] Finally, the surface image of the material to be detected obtained from the above steps is input into the material aging trace recognition device that has been trained, and two key results are calculated and output simultaneously: one is the distribution of suspicious aging areas identified in the material surface image; the other is the probability distribution of the identified areas belonging to various aging types, i.e., the probability distribution of aging types, where each aging type is associated with a specific confidence value.

[0024] Furthermore, after obtaining the distribution of suspicious aging areas and the probability distribution of predicted aging types through the material aging trace identifier, since it is difficult to quantitatively assess the actual difficulty and resource requirements of subsequent fluorescence spectroscopy detection based solely on the direct output of the material aging trace identifier, it is necessary to conduct a special analysis based on the above output to evaluate and determine the first aging detection complexity index and the second aging detection complexity index.

[0025] Specifically, the First Aging Detection Complexity Index is a quantitative indicator calculated based on the distribution of suspected aging areas. It represents the combined influence of the size of the area with aging traces on the surface of the material being tested and the degree of spatial dispersion. The higher the value of this First Aging Detection Complexity Index, the larger the proportion of the suspected area or the more dispersed its distribution. This means that the subsequent point-to-point spectral detection needs to cover a wider range and be more complex in terms of location, resulting in a higher initial complexity of the overall detection.

[0026] The second aging detection complexity index is a quantitative indicator calculated based on the probability distribution of predicted aging types. It represents the combined impact of the number of aging techniques that the material to be tested may involve, the certainty of type identification, and the inherent difficulty coefficient of each technique. The higher the value of this second aging detection complexity index, the more likely the material may involve multiple aging types simultaneously, have a lower confidence level in type identification, or involve several aging techniques that are more difficult to identify. This means that more complex and diverse spectral analysis strategies are needed for confirmation and differentiation, resulting in a higher level of complexity in the overall in-depth analysis of the detection.

[0027] Specifically, the assessment determines the first level of difficulty in aging detection, including: The area percentage of the suspected aging areas is calculated based on the distribution of the suspected aging areas. Calculate the regional distribution dispersion of the suspected aging areas; After dimensionless processing of the area proportion and regional distribution dispersion of the suspected aging area, a first aging detection complexity index is evaluated and determined, wherein the first aging detection complexity index is positively correlated with the area proportion and regional distribution dispersion of the suspected aging area.

[0028] First, based on the distribution of suspicious aging areas output by the material aging trace detector, the area ratio of suspicious aging areas is calculated. This area ratio is obtained by statistically analyzing the ratio of the total number of pixels identified as suspicious areas in the suspicious aging area distribution to the total number of pixels in the entire material surface image. The calculated area ratio of suspicious aging areas reflects the relative coverage of aging traces on the surface of the material to be inspected.

[0029] Secondly, the regional distribution dispersion of the suspected aged areas is calculated. This regional distribution dispersion is used to quantify the degree of dispersion of the suspected areas on the material surface image.

[0030] Preferably, quantitative analysis can be performed based on the geometric centers, or centroids, of the suspected areas. First, through image connectivity analysis, the centroid coordinates of each independent connected region are identified and extracted from the distribution of suspected aging areas. Assuming N suspected areas are identified, N centroid coordinates are obtained. Second, the Euclidean distances between each pair of the N centroids are calculated. Specifically, a distance set D containing all possible point-to-point distances can be generated, containing C(N,2) distance values. The statistical characteristic value of this distance set D is calculated, and this characteristic value is defined as the specific numerical value of the regional distribution dispersion.

[0031] For example, the average value of the distance set D can be directly used as the dispersion of the regional distribution. The larger the average value, the greater the average spatial interval between the centroids of the suspected regions, that is, the more dispersed the regional distribution. In addition, the standard deviation of the distance set D can be used as the dispersion of the regional distribution. The standard deviation is used to measure the fluctuation of the distance between each centroid relative to the average value. A higher standard deviation indicates that some regions are clustered while others are far apart, and the uneven distribution is significant.

[0032] Therefore, the dispersion of regional distribution has a quantifiable index with a clear numerical value. The higher the dispersion value of regional distribution, the more dispersed and less concentrated the suspicious areas are.

[0033] Furthermore, the calculated area proportion and regional distribution dispersion of suspected aging areas are dimensionlessly processed. This dimensionless processing aims to eliminate the influence of dimensions caused by different material sizes, image resolutions, and calculation units, transforming the two indicators into standardized values. Preferably, this processing can be performed using the min-max normalization method, or by employing Z-score standardization combined with scaling in statistics. After dimensionless processing, both the area proportion and regional distribution dispersion of suspected aging areas are transformed into standardized values ​​between 0 and 1. These standardized values ​​maintain monotonic consistency with their original physical meaning; that is, the larger the value, the higher the area proportion or the more dispersed the distribution.

[0034] Finally, based on the dimensionless proportion of the suspected aging area and the dispersion of the regional distribution, the first aging detection complexity index is evaluated and determined. Preferably, it can be calculated using a linear weighted fusion method. Specifically, the first aging detection complexity index = α × the dimensionless proportion of the suspected aging area + β × the dimensionless dispersion of the regional distribution. Here, α and β are the corresponding weighting coefficients, set according to the relative importance of each indicator in the overall detection complexity assessment, and α + β = 1. For example, if the impact of the suspected area coverage on detection complexity is emphasized more, α = 0.6 and β = 0.4 can be set; if the impact of the degree of regional distribution dispersion is emphasized more, α = 0.4 and β = 0.6 can be set.

[0035] Specifically, the calculated first aging detection complexity index is positively correlated with the area ratio of suspected aging areas and the dispersion of regional distribution. That is, the larger the area ratio or the more dispersed the distribution, the higher the calculated first aging detection complexity index value, indicating that the spatial range and positioning complexity that need to be processed in the preliminary detection stage are greater.

[0036] Furthermore, the assessment determined a second complexity index for aging detection, including: The predicted aging type probability distribution is subjected to feature analysis to calculate and obtain the probability information entropy and probability dominance index. The predicted aging type probability distribution is filtered according to the preset aging type probability threshold to obtain the high probability distribution of aging types, and the number of high probability aging types is counted. Based on the inherent identification difficulty coefficient of the aging type identifier, the high probability distribution of the aging type is weighted and fused to assess and determine the aging risk coefficient; After dimensionless processing of the probability information entropy, probability dominance index, number of high-probability aging types, and aging risk coefficient, a second aging detection complexity index is evaluated and determined, wherein the second aging detection complexity index is positively correlated with the probability information entropy, probability dominance index, number of high-probability aging types, and aging risk coefficient.

[0037] First, feature analysis is performed on the probability distribution of predicted aging types output by the material aging trace detector. Specifically, two key indicators are calculated: probability information entropy and probability dominance index. The probability information entropy quantifies the uncertainty or uniformity of the overall predicted aging type probability distribution. A higher value indicates a more ambiguous judgment of the aging type by the model, with multiple types having roughly equal probabilities, thus increasing the difficulty of judgment. The probability dominance index is obtained by calculating the difference between the highest and second-highest probability values. A larger difference indicates that the prediction of a certain aging type is absolutely dominant, and the judgment situation is relatively simple; a smaller difference indicates that two or more types have roughly equal probabilities, and the judgment situation tends to be more complex.

[0038] Secondly, the predicted aging type probability distribution is filtered based on a preset aging type probability threshold. This preset aging type probability threshold is a value set between 0 and 1, representing the boundary between the high-confidence and low-confidence regions of the predicted probability values. This preset aging type probability threshold is typically set based on statistical analysis of historical testing data, expert experience, or the reliability requirements of the identification results; for example, it can be set to 0.7. All aging types with predicted aging type probability values ​​greater than or equal to this preset aging type probability threshold are classified as high-probability types, thus forming a high-probability aging type distribution. The number of high-probability aging types included in this high-probability aging type distribution is then counted. A higher number of high-probability aging types indicates that multiple aging processes may exist simultaneously on the material surface, and the signals from different processes may interfere with each other, making detection and identification more complex, and correspondingly increasing the overall complexity of the detection.

[0039] Secondly, an assessment is conducted based on the inherent difficulty coefficient of identification for each type of aging process. Specifically, for each possible type of aging, such as chemical staining, acid etching, fire treatment, smoking, application of antique paint, and adhesive aging, an inherent difficulty coefficient di is pre-defined based on the experience of experts in the field. This coefficient is typically set to an integer value from 1 to 5, where a value of 1 represents the lowest difficulty of identification and a value of 5 represents the highest difficulty of identification, used to quantify the inherent difficulty of identifying this type of aging process.

[0040] Based on this, a weighted fusion calculation is performed on the aforementioned high-probability distributions of distressed types to assess and determine the distressing risk coefficient. Specifically, Where R represents the aging risk coefficient; K represents the total number of high-probability aging types; pi represents the probability value of the i-th high-probability aging type in the predicted probability distribution, which has been filtered by the preset aging type probability threshold; and di represents the inherent identification difficulty coefficient corresponding to the i-th high-probability aging type.

[0041] The risk factor for this aging process comprehensively reflects the difficulty in identifying high-probability aging techniques. Even if the probability distribution is relatively concentrated, if the concentrated types belong to processes with high identification difficulty, the complexity of the detection should be increased accordingly.

[0042] Furthermore, the calculated probability information entropy, probability dominance index, number of high-probability aging types, and aging risk coefficient are all dimensionless to eliminate the influence of differences in the dimensions and numerical ranges of each indicator, converting them into standardized values. Finally, the second aging detection complexity index is evaluated and determined. Preferably, it can be calculated using a linear combination model, weighting and fusing the four dimensionless indicators mentioned above. The second aging detection complexity index = γ1 × standardized probability information entropy + γ2 × standardized negative probability dominance index + γ3 × standardized number of high-probability aging types + γ4 × standardized aging risk coefficient.

[0043] Here, the standardized probabilistic information entropy represents the dimensionless probabilistic information entropy; the standardized negative probability dominance index is obtained by subtracting the dimensionless probabilistic dominance index from 1, and is used to characterize the degree of lack of dominance; the standardized number of high-probability aging types represents the dimensionless number of high-probability aging types; and the standardized aging risk coefficient represents the dimensionless aging risk coefficient. γ1, γ2, γ3, and γ4 are corresponding weight coefficients, set through expert evaluation, analytic hierarchy process, or machine learning optimization based on historical detection results, and satisfying γ1+γ2+γ3+γ4=1. For example, if the impact of probability distribution uncertainty on complexity is emphasized, γ1=0.4, γ2=0.2, γ3=0.2, and γ4=0.2 can be set; if the risk impact of high-difficulty processes is emphasized, γ1=0.2, γ2=0.2, γ3=0.2, and γ4=0.4 can be set.

[0044] This fusion model ensures that the second aging detection complexity index is positively correlated with probabilistic information entropy, the number of high-probability aging types, and the aging risk coefficient, while being negatively correlated with the probability dominance index. That is, the more uniform and fuzzy the probability distribution, the greater the number of high-probability types, and the higher the inherent identification difficulty of the involved processes, the larger the obtained second aging detection complexity index value, indicating a higher degree of complexity in type identification and analysis.

[0045] S20: Obtain the standard data abundance corresponding to the material to be tested in the pre-constructed natural aging standard sample library, and determine the third aging detection complexity index based on the standard data abundance. Specifically, the abundance of standard data corresponding to the material to be tested in a pre-constructed natural aging standard sample library is obtained, and a third aging detection complexity index is determined based on the abundance of standard data, including: Using the property information of the material to be tested as a constraint, information retrieval is performed in a pre-constructed natural aging standard sample library to obtain a similar standard dataset; The aforementioned standard datasets were evaluated to obtain data size abundance, data quality abundance, data diversity abundance, and counterexample completeness abundance. Standard data abundance is obtained based on the aforementioned data size abundance, data quality abundance, data diversity abundance, and counterexample completeness abundance assessment. The ratio of the preset data abundance scalar to the standard data abundance is used as the third aging detection complexity index.

[0046] First, information is retrieved from a pre-constructed natural aging standard sample library, using the property information of the material to be tested as a constraint. This pre-constructed natural aging standard sample library is a systematically collected, labeled, and structured database containing identified natural aging sample data for common materials used in various cultural relics and artworks. The natural aging sample data includes, but is not limited to, multispectral image data, chemical composition spectrum data, microscopic morphology data, and associated complete metadata information. By precisely matching the property information of the material to be tested, such as material type, craftsmanship, and historical period, standard sample data that are highly similar or of the same category and characteristics as the material to be tested can be selected from this pre-constructed natural aging standard sample library, thereby obtaining a similar standard dataset.

[0047] Secondly, a comprehensive evaluation of the obtained standard datasets of the same type was conducted. Specifically, quantitative indicators were obtained from four dimensions: data size abundance, data quality abundance, data diversity abundance, and counterexample completeness abundance.

[0048] Data abundance measures the sufficiency of the number of samples in a standardized dataset of similar types. It is calculated by counting the total number of valid samples N in the standardized dataset of similar types and comparing it with a pre-set empirical optimal sample size N. optimal A comparison is made. The specific calculation formula is: Data abundance = min(N / N) optimal If the total number of samples in the same standard dataset reaches or exceeds the optimal number, the data abundance value is 1.0, indicating that the data size is sufficient; otherwise, the data abundance value is taken proportionally, and the lower the value, the more insufficient the data size.

[0049] Data quality abundance measures the reliability and completeness of the data itself. Its assessment comprehensively considers multiple sub-indicators, including the data completeness of each sample, the average signal-to-noise ratio of the spectral data, and the authority and confidence of the label source. The data quality abundance is obtained by assigning a quality score to each sample and calculating the average quality score for the entire dataset. This data quality abundance reflects the usability of the data.

[0050] Data diversity abundance is used to assess the breadth of coverage across multiple dimensions of a standard dataset. These dimensions include at least time span, geographic origin, preservation status, and aging characteristics. For each dimension, the dispersion of its distribution is calculated, or the number of distinct categories is counted. For example, for time span, the standard deviation of the time range can be calculated; for geographic origin, the number of different origins can be counted. The score for each dimension is normalized to a range of 0-1. The final data diversity abundance is the weighted average of the scores for all dimensions, with weights set according to the importance of each dimension for identification.

[0051] The completeness abundance of counterexamples is used to assess the sufficiency of coverage of artificially aged samples in a similar standard dataset. First, the number M of samples explicitly labeled as antique-style is counted in the similar standard dataset. Second, the types of aging techniques covered by the samples are checked to see if they include common categories such as chemical staining, acid etching, and fire roasting, and the number C of covered categories is counted. The calculation of the completeness abundance of counterexamples consists of two parts: Sample quantity sufficiency score = min(M / M optimal ,1.0), where M optimal The minimum number of negative examples required is preset; Category coverage score = C / C total C total This represents the total number of categories of common anti-aging techniques. The final counterexample completeness abundance score is a weighted sum of these two scores, with the weights adjustable according to actual needs. A higher counterexample completeness abundance score indicates a more comprehensive understanding of artificial anti-aging phenomena in the database.

[0052] Furthermore, based on the abundance indices of the above four dimensions, a comprehensive standard data abundance value is calculated through weighted averaging or other fusion methods to comprehensively characterize the database's support for and understanding of the current material category to be detected. Preferably, the standard data abundance value = ω1 × data size abundance + ω2 × data quality abundance + ω3 × data diversity abundance + ω4 × counterexample completeness abundance.

[0053] Here, ω1, ω2, ω3, and ω4 are the corresponding weighting coefficients, set according to the relative importance of the abundance of each dimension to the overall data support capability, and satisfying ω1+ω2+ω3+ω4=1. For example, if the fundamental role of data scale and quality is emphasized, ω1=0.3, ω2=0.3, ω3=0.2, and ω4=0.2 can be set; if the ability to identify abnormal aging situations is particularly important, ω1=0.2, ω2=0.2, ω3=0.2, and ω4=0.4 can be set.

[0054] Finally, the ratio of the preset data abundance scalar to the calculated standard data abundance is used as the third aging detection complexity index. The preset data abundance scalar is a pre-defined initial reference value, typically set as the average data abundance level based on historical statistics, considered sufficient to support reliable detection. Therefore, the third aging detection complexity index directly reflects the relative strength of the support capability of the natural aging standard sample library: the lower the standard data abundance, i.e., the weaker the cognitive foundation of the corresponding material in the natural aging standard sample library, the higher the ratio, i.e., the third aging detection complexity index, indicating greater complexity and uncertainty in data-driven analysis.

[0055] S30: The first aging detection complexity index, the second aging detection complexity index, and the third aging detection complexity index are weighted and fused to obtain the overall aging detection complexity; Specifically, this step aims to integrate the first, second, and third aging detection complexity indices, which respectively characterize spatial distribution complexity, process type identification complexity, and data support insufficiency complexity, into a unified overall evaluation index. Preferably, the first aging detection complexity index I1, the second aging detection complexity index I2, and the third aging detection complexity index I3 are linearly weighted and summed using preset weighting coefficients. Overall aging detection complexity = λ1×I1 + λ2×I2 + λ3×I3.

[0056] Here, λ1, λ2, and λ3 are the corresponding weighting coefficients, reflecting the contribution proportions of the three dimensions—space, process, and data—to the total complexity of the final detection operation, respectively. The weighting coefficients can be set based on the actual configuration of detection resources, the degree of influence of each dimension on the final identification result, or historical experience, and must satisfy λ1 + λ2 + λ3 = 1. For example, if detection resources are more geared towards addressing process identification challenges, λ2 can be set higher; if the database is generally weak, making data support the main bottleneck, the weight of λ3 can be increased accordingly.

[0057] Through fusion calculation, the overall complexity of aging detection is represented by a continuous quantitative value, which comprehensively characterizes the overall difficulty faced in the three aspects of regional positioning, process analysis and data reference when performing fluorescence spectroscopy identification on the material to be tested. This provides a decision-making basis for subsequently adaptively configuring the best detection scheme.

[0058] S40: Configure the optimal fluorescence spectroscopy identification scheme based on the overall complexity of the antiquing detection, and perform antiquing trace detection on the material to be tested.

[0059] Specifically, based on the pre-constructed mapping table of aging detection complexity and fluorescence spectroscopy identification scheme, the optimal fluorescence spectroscopy identification scheme is obtained according to the overall aging detection complexity.

[0060] Specifically, such as Figure 2 As shown, the process of constructing the mapping table for the aging detection complexity-fluorescence spectroscopy identification scheme includes: Configure several aging detection complexities, and randomly select the first aging detection complexity; Obtain the spectral identification parameter space of the fluorescence spectral identification scheme, wherein the spectral identification parameters include spectral technology type, number of detection points, spectral acquisition parameters, and signal processing level; Based on the spectral identification parameter space, the spectral identification scheme is optimized for the first aging detection complexity, and the first optimal spectral identification scheme is output. Establish a first mapping relationship between the first aging detection complexity and the first optimal spectral identification scheme, and sequentially analyze and obtain several mapping relationships of the several aging detection complexities to construct an aging detection complexity-fluorescence spectral identification scheme mapping table.

[0061] Specifically, the configuration of the optimal fluorescence spectroscopy identification scheme is based on a pre-constructed mapping table of aging detection complexity and fluorescence spectroscopy identification scheme. In particular, based on the overall aging detection complexity calculated above, the corresponding pre-optimized optimal fluorescence spectroscopy identification scheme can be obtained by querying this mapping table.

[0062] The process of constructing the mapping table for the complexity of aging detection and fluorescence spectroscopy identification scheme includes the following steps: First, configure several aging detection complexities, and randomly select the first aging detection complexity.

[0063] Secondly, the adjustable parameter range of the fluorescence spectroscopy identification scheme should be clearly defined, i.e., its spectral identification parameter space should be obtained. This spectral identification parameter space includes a combination of several key parameters that determine detection performance and resource consumption: First, the type of spectral technique, such as steady-state fluorescence spectroscopy, time-resolved fluorescence spectroscopy, fluorescence microscopy, excitation-emission matrix spectroscopy, etc., the selection of which depends on whether it is necessary to separate mixed signals or obtain spatial distribution information. Second, the number and distribution of detection points, i.e., the number of detection points and their coordinates or grid density on the material surface, the setting of which is directly based on the distribution of suspected aging areas, and the higher the complexity, the denser the points. Third, spectral acquisition parameters, including excitation wavelength, emission wavelength range, spectral resolution, integration time and averaging times, etc., these parameters together determine the quality and information content of the acquired data. Fourth, the signal processing level, such as level 1 basic baseline correction and smoothing, level 2 standard feature extraction, level 3 derivative spectrum analysis and multivariate unmixing, the higher the level, the greater the processing depth.

[0064] Furthermore, based on the spectral identification parameter space, the spectral identification scheme is optimized to improve the complexity of the first aging detection.

[0065] Specifically, based on the spectral identification parameter space, the spectral identification scheme is optimized for the complexity of the first aging detection, and a first optimal spectral identification scheme is output, including: Constrained by the complexity of the first antiquing detection, historical spectral identification records of similar materials are retrieved, and several first sample material feature sets are collected; Within the spectral identification parameter space, parameters are randomly selected to generate a first spectral identification scheme, and the first identification time and first identification resource consumption of the first spectral identification scheme are obtained. The pre-trained identification confidence prediction plugin predicts the identification confidence of the first spectral identification scheme and the several first sample material feature sets respectively, and counts the proportion of the number of predicted identification confidence values ​​greater than the preset confidence threshold, which is used as the first identification quality coefficient. If the first identification quality coefficient is less than or equal to the preset standard identification quality coefficient, then the first spectral identification scheme is discarded. If the first identification quality coefficient is greater than the preset standard identification quality coefficient, then the first spectral identification scheme is considered a qualified spectral identification scheme and added to the candidate scheme set. The first scheme fitness of the first spectral identification scheme is determined based on the first identification quality coefficient, the first identification time, and the first identification resource consumption assessment. Based on the spectral identification parameter space, iterative selection and evaluation of schemes are performed until a preset number of convergences is reached. The spectral identification scheme corresponding to the maximum fitness of the candidate scheme set is then output as the first optimal spectral identification scheme.

[0066] First, using the first level of aging detection complexity as a constraint, detection records of similar materials with the same or similar overall complexity are retrieved, and several first sample material feature sets are collected. These first sample material feature sets include the probability distribution of aging types, the distribution of suspected aging areas, and their standard data abundance for the corresponding historical samples. These several first sample material feature sets represent diverse combinations of features that materials may exhibit at this complexity level.

[0067] Secondly, within the spectral identification parameter space, a specific first spectral identification scheme is generated by randomly selecting the spectral technology type, detection point parameters, acquisition parameters, and signal processing level. Simultaneously, through simulation calculations or querying historical data records, the first identification time and first identification resource consumption required to execute this first spectral identification scheme are obtained. The identification time refers to the total time consumed in completing the entire process from spectral acquisition and signal processing to the final output of the identification result, including instrument preparation time, multi-point scanning time, data transmission time, and data analysis and processing time. Resource consumption refers to the total cost incurred or consumed during the execution of this identification scheme, including quantitative indicators such as equipment operating energy consumption, dedicated computing server resource consumption, and potential costs of detection consumables.

[0068] Secondly, an evaluation is performed using a pre-trained identification confidence prediction plugin. This plugin takes the first spectral identification scheme and the aforementioned set of first sample material features as input, and can predict the identification confidence that the first spectral identification scheme can achieve for each set of sample material features.

[0069] Specifically, the pre-trained confidence prediction plugin includes: Based on historical spectral identification records of similar materials, multiple sample material feature sets and multiple sample spectral identification schemes were collected. The historical antiquing and aging identification accuracy of the sample material feature sets under the corresponding sample spectral identification schemes was obtained as the sample identification credibility, and multiple sample identification credibility were obtained. Using the multiple sample material feature sets and multiple sample spectral identification schemes as inputs, and the identification credibility of the multiple samples as supervision, a deep learning model is trained until convergence, generating an identification credibility prediction plugin.

[0070] First, a training dataset is constructed based on historical spectral identification records of similar materials. Specifically, multiple sample material feature sets are collected, each containing the probability distribution of aging types, the distribution of suspected aging areas, and the standard data abundance for the corresponding historical spectral identification record. Simultaneously, multiple sample spectral identification schemes actually used in the corresponding historical spectral identification records are collected, each containing its specific spectral technology type, detection point parameters, acquisition parameters, and signal processing level. For each pair of sample material feature sets and their corresponding sample spectral identification scheme, historical identification results are queried to obtain the historical antiquing / aging identification accuracy rate obtained for that specific material under that specific scheme. This historical antiquing / aging identification accuracy rate is used as the sample identification credibility of that data pair.

[0071] In this way, a labeled dataset can be obtained, consisting of multiple sample material feature sets, multiple sample spectral identification schemes, and their corresponding sample identification confidence levels.

[0072] Secondly, a deep learning model is trained using multiple sample material feature sets and multiple sample spectral identification schemes as combined input features, and multiple sample identification confidence levels as the supervised objective. The architecture of this deep learning model needs to effectively integrate two types of heterogeneous inputs: material features and scheme parameters, and learn the complex mapping relationship between them and the final identification confidence level. The deep learning model is trained until convergence, generating an identification confidence prediction plugin. The convergence condition is set based on the model's performance on an independent validation set. Specifically, the training process continuously monitors the model's root mean square error (RMSE) on the validation set. When this error value decreases by less than a preset small threshold (e.g., less than 1‰) over multiple consecutive training epochs (e.g., 20 consecutive epochs), the model is considered to have converged, and training terminates. Simultaneously, auxiliary convergence conditions can be set, such as requiring the validation set error itself to be below a certain absolute threshold (e.g., 0.05), to ensure the model achieves sufficient prediction accuracy.

[0073] The final identification confidence prediction plugin has a forward prediction function. When a new material feature set and a spectral identification scheme to be evaluated are input, it can output a predicted identification confidence value, thereby pre-evaluating the expected identification reliability of the scheme for the material feature without actually performing the detection.

[0074] Furthermore, the trained identification confidence prediction plugin is invoked to predict the identification confidence of the first spectral identification scheme and several first sample material feature sets respectively, and the proportion of the predicted identification confidence greater than the preset confidence threshold is counted as the first identification quality coefficient.

[0075] The preset confidence threshold is a value between 0 and 1, representing the minimum confidence standard for determining whether a spectral identification scheme's identification result for a specific material feature set is reliable. This preset confidence threshold is set based on the minimum requirements for identification accuracy in practical applications, the statistical distribution of historical identification results, or the experience of domain experts. For example, it can be set to 0.85, meaning that when the identification confidence prediction plugin assesses that a scheme's identification confidence for a certain material exceeds 85%, the identification is considered basically reliable.

[0076] Then, a scheme screening process is performed. If the first identification quality coefficient is less than or equal to the preset standard identification quality coefficient, the average identification reliability of the scheme is considered to have failed to meet the basic requirements and is discarded. If the first identification quality coefficient is greater than the preset standard identification quality coefficient, the first spectral identification scheme is considered a qualified scheme and added to the candidate scheme set.

[0077] Secondly, a comprehensive evaluation is performed on the first spectral identification scheme that enters the candidate scheme set. Based on its first identification quality coefficient, first identification time, and first identification resource consumption, its fitness is calculated. Preferably, the fitness of the first scheme = κ. q ×Q+κ t ×(1 / T)+κ c ×(1 / C).

[0078] Where Q represents the first discrimination quality coefficient, T represents the first discrimination duration, and C represents the first discrimination resource consumption. q κ t With κ c The corresponding weighting coefficients are set based on the relative importance placed on identification quality, detection efficiency, and cost control in actual detection tasks, and satisfy κ. q +κ t +κ c =1. For example, if identifying quality is the primary objective, κ can be set. q =0.7, κ t =0.15, κ c =0.15; If efficiency must be prioritized while ensuring a certain level of quality, then κ can be set. q =0.5, κ t =0.3, κ c =0.2.

[0079] Finally, based on the spectral identification parameter space, the above steps are repeated: randomly generating new spectral identification schemes, evaluating their identification quality coefficients, screening, and calculating the fitness of qualified schemes. This iterative process continues until a preset number of convergences is reached, for example, 1000 iterations. After iteration, the spectral identification scheme with the highest fitness is selected from the final accumulated candidate scheme set as the first optimal spectral identification scheme corresponding to the first aging detection complexity. This first optimal spectral identification scheme is the detection parameter configuration that achieves the best balance between quality, efficiency, and cost under a given complexity constraint.

[0080] Furthermore, a correspondence is established between the first aging detection complexity and the first optimal spectral identification scheme, i.e., the first mapping relationship. For each of the remaining pre-configured aging detection complexity nodes, the above optimization process is repeated to obtain their respective optimal spectral identification schemes and mapping relationships in turn.

[0081] Finally, by systematically aggregating all discrete mapping relationships, a complete mapping table of aging detection complexity and fluorescence spectroscopy identification scheme is constructed. This mapping table enables rapid matching and retrieval of any overall aging detection complexity value to a specific, executable optimal detection scheme.

[0082] Finally, based on the optimal fluorescence spectral identification scheme obtained from the configuration, the material to be tested was subjected to antiquing and aging trace detection.

[0083] In summary, the embodiments of this application have at least the following technical effects: Compared to existing technologies, this invention achieves intelligent, multi-dimensional, and adaptive detection of traces of antique aging. By integrating image analysis, database evaluation, and fluorescence spectroscopy, a complete identification chain from macroscopic appearance to microscopic components is constructed, reducing reliance on expert subjective experience and improving the objectivity and repeatability of the detection. Secondly, this invention innovatively proposes an adaptive strategy matching mechanism based on overall detection complexity. By quantitatively evaluating the complexity from multiple sources such as spatial distribution, process type, and data support, and using this to drive the automatic optimization and configuration of the detection scheme, efficient allocation of detection resources is achieved. It can quickly screen in simple cases and automatically initiate in-depth analysis in complex cases, thereby significantly improving overall detection efficiency while ensuring identification accuracy.

[0084] Finally, the closed-loop optimization framework constructed in this invention possesses continuous learning and evolution capabilities. The authentication credibility prediction model trained on historical data and the scheme optimization process enable the system to continuously accumulate knowledge and optimize its parameter mapping table, thereby continuously improving its adaptability to new materials and new aging techniques, and enhancing its authentication accuracy. This provides a sustainable technical solution for solving the increasingly complex problem of authenticating cultural relics and artworks.

[0085] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0086] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0087] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and modifications fall within the scope of this application and its equivalents, this application intends to include such modifications and modifications.

Claims

1. A novel method for detecting traces of antique aging in materials using fluorescence spectroscopy, characterized in that, The methods include: The surface image of the material to be tested is used to identify signs of aging, obtain the distribution of suspicious aging areas and the probability distribution of predicted aging types, and evaluate and determine the first and second aging detection complexity indices. Obtain the standard data abundance corresponding to the material to be tested in the pre-constructed natural aging standard sample library, and determine the third aging detection complexity index based on the standard data abundance. The overall aging detection complexity is obtained by weighting and fusing the first aging detection complexity index, the second aging detection complexity index, and the third aging detection complexity index. Based on the overall complexity of the antiquing detection, the optimal fluorescence spectroscopy identification scheme is configured to detect traces of antiquing on the material to be tested.

2. The method for detecting antique aging traces in new materials by combining fluorescence spectroscopy identification according to claim 1, characterized in that, The surface image of the material to be tested is used to identify signs of aging, obtain the distribution of suspected aging areas and the probability distribution of predicted aging types, including: Multi-angle surface image acquisition is performed on the material to be tested to obtain surface images of the material; A material aging trace detector was built based on a convolutional neural network. The material surface image is input into the material aging trace identifier, which outputs the distribution of suspicious aging areas and the probability distribution of predicted aging types.

3. The method for detecting antique-style aging traces in new materials by combining fluorescence spectroscopy identification according to claim 2, characterized in that, A material aging trace detector based on convolutional neural networks includes: Using the attribute information of the material to be tested as a constraint, a set of sample material surface images is collected. The distribution set of sample aging areas is obtained by manually annotating the aging areas of different sample material surface images. The probability distribution set of sample aging types is obtained by manually classifying the aging type and aging confidence standard of different sample material surface images. The aging type includes one or more of the following: chemical dyeing, acid etching, alkali etching, fire baking, smoke smoking, application of antique paint and adhesive aging. Using the sample material surface image set as input data, and the sample aging region distribution set and sample aging type probability distribution set as supervision labels, a convolutional neural network is trained until convergence to generate a material aging trace recognizer.

4. The method for detecting antique aging traces in new materials by combining fluorescence spectroscopy identification according to claim 1, characterized in that, The assessment determined the first complexity index for aging detection, including: The area percentage of the suspected aging areas is calculated based on the distribution of the suspected aging areas. Calculate the regional distribution dispersion of the suspected aging areas; After dimensionless processing of the area proportion and regional distribution dispersion of the suspected aging area, a first aging detection complexity index is evaluated and determined, wherein the first aging detection complexity index is positively correlated with the area proportion and regional distribution dispersion of the suspected aging area.

5. The method for detecting antique aging traces in new materials by combining fluorescence spectroscopy identification according to claim 1, characterized in that, The assessment determined the second complexity index for aging detection, including: The predicted aging type probability distribution is subjected to feature analysis to calculate and obtain the probability information entropy and probability dominance index. The predicted aging type probability distribution is filtered according to the preset aging type probability threshold to obtain the high probability distribution of aging types, and the number of high probability aging types is counted. Based on the inherent identification difficulty coefficient of the aging type identifier, the high probability distribution of the aging type is weighted and fused to assess and determine the aging risk coefficient; After dimensionless processing of the probability information entropy, probability dominance index, number of high-probability aging types, and aging risk coefficient, a second aging detection complexity index is evaluated and determined, wherein the second aging detection complexity index is positively correlated with the probability information entropy, probability dominance index, number of high-probability aging types, and aging risk coefficient.

6. The method for detecting antique aging traces in new materials by combining fluorescence spectroscopy identification according to claim 1, characterized in that, Obtain the standard data abundance of the material to be tested in a pre-constructed natural aging standard sample library, and determine the third aging detection complexity index based on the standard data abundance, including: Using the property information of the material to be tested as a constraint, information retrieval is performed in a pre-constructed natural aging standard sample library to obtain a similar standard dataset; The aforementioned standard datasets were evaluated to obtain data size abundance, data quality abundance, data diversity abundance, and counterexample completeness abundance. Standard data abundance is obtained based on the aforementioned data size abundance, data quality abundance, data diversity abundance, and counterexample completeness abundance assessment. The ratio of the preset data abundance scalar to the standard data abundance is used as the third aging detection complexity index.

7. The method for detecting antique aging traces in new materials by combining fluorescence spectroscopy identification according to claim 1, characterized in that, Based on the pre-constructed mapping table of aging detection complexity and fluorescence spectroscopy identification scheme, the optimal fluorescence spectroscopy identification scheme is obtained according to the overall aging detection complexity.

8. The method for detecting antique aging traces in new materials by combining fluorescence spectroscopy identification according to claim 7, characterized in that, The process of constructing the mapping table for the complexity of anti-aging detection and fluorescence spectroscopy identification scheme includes: Configure several aging detection complexities, and randomly select the first aging detection complexity; Obtain the spectral identification parameter space of the fluorescence spectral identification scheme, wherein the spectral identification parameters include spectral technology type, number of detection points, spectral acquisition parameters, and signal processing level; Based on the spectral identification parameter space, the spectral identification scheme is optimized for the first aging detection complexity, and the first optimal spectral identification scheme is output. Establish a first mapping relationship between the first aging detection complexity and the first optimal spectral identification scheme, and sequentially analyze and obtain several mapping relationships of the several aging detection complexities to construct an aging detection complexity-fluorescence spectral identification scheme mapping table.

9. The method for detecting antique-style aging traces in new materials by combining fluorescence spectroscopy identification according to claim 8, characterized in that, Based on the spectral identification parameter space, the spectral identification scheme is optimized for the complexity of the first aging detection, and the first optimal spectral identification scheme is output, including: Constrained by the complexity of the first antiquing detection, historical spectral identification records of similar materials are retrieved, and several first sample material feature sets are collected; Within the spectral identification parameter space, parameters are randomly selected to generate a first spectral identification scheme, and the first identification time and first identification resource consumption of the first spectral identification scheme are obtained. The pre-trained identification confidence prediction plugin predicts the identification confidence of the first spectral identification scheme and the several first sample material feature sets respectively, and counts the proportion of the number of predicted identification confidence values ​​greater than the preset confidence threshold, which is used as the first identification quality coefficient. If the first identification quality coefficient is less than or equal to the preset standard identification quality coefficient, then the first spectral identification scheme is discarded. If the first identification quality coefficient is greater than the preset standard identification quality coefficient, then the first spectral identification scheme is considered a qualified spectral identification scheme and added to the candidate scheme set. The first scheme fitness of the first spectral identification scheme is determined based on the first identification quality coefficient, the first identification time, and the first identification resource consumption assessment. Based on the spectral identification parameter space, iterative selection and evaluation of schemes are performed until a preset number of convergences is reached. The spectral identification scheme corresponding to the maximum fitness of the candidate scheme set is then output as the first optimal spectral identification scheme.

10. The method for detecting antique-style aging traces in new materials by combining fluorescence spectroscopy identification according to claim 9, characterized in that, Pre-trained confidence prediction plugin, including: Based on historical spectral identification records of similar materials, multiple sample material feature sets and multiple sample spectral identification schemes were collected. The historical antiquing and aging identification accuracy of the sample material feature sets under the corresponding sample spectral identification schemes was obtained as the sample identification credibility, and multiple sample identification credibility were obtained. Using the multiple sample material feature sets and multiple sample spectral identification schemes as inputs, and the identification credibility of the multiple samples as supervision, a deep learning model is trained until convergence, generating an identification credibility prediction plugin.

Images