Hyperspectral-based method for evaluating the degree of smoke of mural paintings

Abstract

The application discloses a kind of wall painting smoke degree evaluation methods based on hyperspectral, comprising: sample mural is identified to pigment, and sample mural is divided into several sample regions;Using elastic regression network, the hyperspectral reflectivity of multiple sample regions of the same pigment and smoke pollution degree index are processed, and the preferred wave band of each pigment and adaptive wave band weight are determined;The pigment of the wall painting to be measured is identified, and the wall painting to be measured is divided into several to be measured region;The smoke pollution degree value SRC of each to be measured region is calculated λ p , as follows: SRC λ p =w j p ·(R clean (λ)‑R smoke (λ)),Wherein, R smoke (λ) represents the hyperspectral reflectivity of to be measured region λ, R clean (λ) represents the hyperspectral reflectivity of un-contaminated region in the wall painting to be measured, w j p It represents the adaptive wave band weight of the preferred wave band of the pigment corresponding to to be measured region λ.The application can realize the accurate evaluation of wall painting smoke degree.

Description

Technical Field

[0001] This invention relates to the field of software, and more particularly to a method for assessing the degree of smoke damage to murals based on hyperspectral imaging. Background Technology

[0002] Ancient murals, as an important part of human cultural heritage, carry profound historical and cultural value. However, long-term exposure to environmental pollution and human activities makes their surfaces highly susceptible to smoke pollution. The main causes of this pollution are smoke from daily cooking and heating, as well as incense offerings. Smoke pollution primarily consists of carbon particles and tar-like substances produced during combustion, which are extremely harmful to murals. On the one hand, it darkens the colors of the murals, forming a black coating on the surface, greatly affecting people's visual perception of the mural's content; on the other hand, smoke particles may penetrate the pigment layer, altering its optical properties and damaging the original appearance of the mural. Even more challenging is the difficulty of cleaning smoke pollution. Traditional physical cleaning methods, based on the principles of "minimal intervention" and "not altering the original state of the artifact," are highly likely to damage the original pigment layer during operation, causing irreversible damage to the mural. Addressing the technical problem of the inability to automatically assess the degree of smoke pollution in murals using existing technologies, there is an urgent need to propose a method that can assess the degree of smoke pollution in murals without causing secondary damage. Summary of the Invention

[0003] One object of the present invention is to solve at least the above-mentioned problems and / or defects, and to provide at least the advantages described below.

[0004] One object of the present invention is to provide a method for assessing the degree of smoke damage to murals based on hyperspectral imaging, which can achieve an accurate assessment of the degree of smoke damage to murals without causing secondary damage.

[0005] To achieve these objectives and other advantages according to the present invention, a method for assessing the degree of smoke damage to murals based on hyperspectral imaging is provided, comprising:

[0006] Pigment identification is performed on the sample mural, and the sample mural is divided into several sample areas, wherein two adjacent sample areas correspond to different pigments, and the same pigment corresponds to multiple sample areas;

[0007] The hyperspectral reflectance and smoke pollution level indicators of each sample area were obtained. The hyperspectral data of each sample area included reflectance in multiple bands.

[0008] The hyperspectral reflectance and smoke pollution level indices of multiple sample regions corresponding to the same pigment were processed using an elastic regression network to determine the preferred bands for each pigment and the adaptive band weights of the preferred bands.

[0009] Pigment identification is performed on the mural to be tested, and the mural to be tested is divided into several test areas, wherein two adjacent test areas correspond to different pigments;

[0010] Obtain the hyperspectral reflectance of each region to be measured;

[0011] The smoke pollution level (SRC) of each test area was calculated using the formula for spectral reflectance variation. λ p The formula for the change in spectral reflectance is as follows:

[0012] SRC λ p =w j p ·(R clean (λ)-R smoke (λ))

[0013] Among them, R smoke (λ) represents the hyperspectral reflectance of the region λ to be measured, R clean (λ) represents the hyperspectral reflectance of the uncontaminated area in the mural to be tested, w j p The adaptive band weight represents the preferred band weight of the pigment corresponding to the test region λ, and p represents the number of bands of the pigment corresponding to the test region λ.

[0014] Preferably, in the hyperspectral-based method for assessing the degree of smoke damage to murals, the optimization objective function of the elastic regression network is as follows:

[0015]

[0016] Where, n p λ1 represents the number of sample regions corresponding to a certain pigment. p , λ2 p These are hyperparameters set for the corresponding pigments; y i This is an indicator of the degree of smoke pollution in the i-th sample region corresponding to the pigment; x ij β is the reflectance of the i-th sample region corresponding to the pigment in the j-th band; j These are the regression coefficients of the pigment on the j-th band, β=[β1,β2,...,β p ] represents the regression coefficients for all bands; w j p It is the adaptive band weight of the corresponding pigment in the j-th band. ε is a small value.

[0017] Preferably, in the hyperspectral-based method for assessing the degree of smoke damage to murals, the elastic regression network is fitted using the Lasso regression formula to obtain the regression coefficient β for each band. j The Lasso regression formula is as follows: n is the number of sample regions for a particular pigment, p is the number of wavelength bands, β = [β1, β2, ..., βp] is the regression coefficient for all wavelength bands, β j Let X be the regression coefficient for the j-th band. i Let y be the hyperspectral reflectance of the i-th sample. i Let λ be the smoke pollution level index for the i-th sample, and λ be the regularization intensity.

[0018] Preferably, in the hyperspectral-based method for assessing the degree of smoke on murals, a random forest algorithm is used to identify pigments in the sample murals and the murals to be tested.

[0019] Preferably, in the hyperspectral-based method for assessing the degree of smoke in murals, each test area is classified into pollution levels based on a pollution level formula according to the degree of smoke pollution in each test area. The pollution level formula is as follows:

[0020]

[0021] Among them, Pollution Level is the indicator of the degree of smoke pollution in the corresponding area to be tested, and τ1, τ2, and τ3 are threshold values.

[0022] Preferably, in the hyperspectral-based method for assessing the degree of smoke in murals, τ1 = 5%, τ2 = 10%, and τ3 = 25%.

[0023] Preferably, in the hyperspectral-based method for assessing the degree of smoke damage to murals, the hyperspectral reflectance of each sample area and each test area is obtained by processing the original hyperspectral reflectance of each sample area and each test area using the following method: Radiometric correction is performed on the original hyperspectral reflectance of each sample area and each test area, using the following correction formula:

[0024]

[0025] R represents the corrected reflectance; Rraw represents the original hyperspectral data; Rwhite represents the data obtained from the standard reflector in the field; Rdark represents the dark current noise data obtained with the light source off and the lens covering the image. The reflectance of the standard reflector is 99%.

[0026] Preferably, in the hyperspectral-based method for assessing the degree of smoke in murals, a minimum noise separation transform is used to reduce the dimensionality of the corrected reflectance.

[0027] The present invention has at least the following beneficial effects:

[0028] 1) Non-destructive and efficient identification: This technical solution utilizes hyperspectral imaging technology to achieve pixel-level classification and accurate pollution detection, which can meet the identification needs of smoke-stained pigments and avoid secondary damage to murals.

[0029] 2) Accuracy and robustness: The improved elastic regression network can adapt to the spectral characteristics of different pigments, determine the preferred wavelengths for different pigments, and determine the weights of the preferred wavelengths, thereby improving the accuracy and robustness of smoke pollution detection.

[0030] 3) Based on the assessment of the degree of smoke pollution, a comprehensive assessment of smoke pollution can be made, providing support for subsequent digital cleaning of smoked murals and assisting in mural restoration decisions.

[0031] Other advantages, objectives and features of the present invention will become apparent in part from the following description, and in part from those skilled in the art through study and practice of the invention. Attached Figure Description

[0032] Figure 1 This is a flowchart of a method for identifying smoked mural pigments based on hyperspectral imaging in an embodiment of the present invention. Detailed Implementation

[0033] The present invention will now be described in further detail with reference to the accompanying drawings, so that those skilled in the art can implement it based on the description.

[0034] like Figure 1 As shown, this invention provides a method for assessing the degree of smoke damage to murals based on hyperspectral imaging, comprising:

[0035] Pigment identification is performed on the sample murals, dividing them into several sample regions. Adjacent sample regions correspond to different pigments, and the same pigment can correspond to multiple sample regions. Mural pigment classification methods can employ traditional supervised classification, unsupervised classification, other machine learning classification methods, and deep learning classification methods.

[0036] The hyperspectral reflectance and smoke pollution level indices for each sample region are obtained, wherein the hyperspectral data for each sample region includes reflectance across multiple bands. The smoke pollution level value for each sample region can be calculated using a formula that does not consider the variation in spectral reflectance across preferred bands (see Formula 8 in the embodiment).

[0037] The hyperspectral reflectance and soot pollution levels of multiple sample regions corresponding to the same pigment were processed using an elastic regression network to determine the preferred bands for each pigment and the adaptive band weights of the preferred bands.

[0038] Pigment identification is performed on the mural to be tested, and the mural to be tested is divided into several test areas, wherein two adjacent test areas correspond to different pigments.

[0039] Obtain the hyperspectral reflectance of each region to be tested.

[0040] The smoke pollution level (SRC) of each test area was calculated using the formula for spectral reflectance variation. λ p The formula for the change in spectral reflectance is as follows:

[0041] SRC λ p =w j p ·(R clean (λ)-R smoke (λ))

[0042] Among them, R smoke (λ) represents the hyperspectral reflectance of the region λ to be measured, R clean (λ) represents the hyperspectral reflectance of the uncontaminated area in the mural to be tested, w j p The adaptive band weight represents the preferred band weight of the pigment corresponding to the test region λ, and p represents the number of bands of the pigment corresponding to the test region λ.

[0043] Hyperspectral imaging is a non-contact, non-destructive testing method. Compared to traditional methods (such as chemical testing and XRF), hyperspectral imaging is a non-contact detection technology, avoiding secondary damage to murals and is suitable for analyzing large-area smoke pollution. Traditional visible light imaging can only distinguish color changes, but cannot determine whether the darkening of the color is caused by smoke or the original pigment itself being darker. Hyperspectral analysis can distinguish between smoke layers and pigment layers based on spectral characteristics. Hyperspectral imaging technology can detect pollution layers that are difficult to see with the naked eye, accurately identifying the impact of smoke on different pigment layers. After a mural is smoked, its surface condition is complex, and some pollution layers are difficult to observe with the naked eye. Hyperspectral technology, with its high-resolution spectral detection capabilities, can keenly capture these subtle changes, revealing pollution conditions invisible to the naked eye. Through the analysis of the spectral characteristics of different pigment layers, it can also clearly distinguish between pollution and original pigment, thus achieving precise separation of the smoke layer.

[0044] In the hyperspectral analysis of smoked murals, different pigments exhibit different spectral characteristics, and certain wavelengths may be more sensitive to smoke contamination. Therefore, directly using an elastic regression network may overlook the specificity between pigments, leading to inaccurate band selection and affecting the identification and assessment of smoke contamination. In analyzing sample murals, pigment zoning involves first classifying the pigments and then training an elastic regression network separately for each pigment region. This identifies the preferred bands for each pigment to be more sensitive to smoke contamination, determines the weights of these preferred bands, and uses the weights of the preferred bands corresponding to the pigments in the tested area when assessing the mural. This improves the accuracy of band selection and the robustness of contamination assessment.

[0045] This method can accurately measure the spectral characteristics of different pigment areas in murals after smoking, extract the sensitive bands to smoke pollution, quantify the polluted areas, and intuitively show the degree of pollution. It provides a precise basis for digital cleaning, reduces human error, and helps to restore ancient murals efficiently and accurately.

[0046] In a preferred embodiment, the optimization objective function of the elastic regression network in the hyperspectral-based method for assessing the degree of smoke damage to murals is as follows:

[0047]

[0048] Where, n p λ1 represents the number of sample regions corresponding to a certain pigment. p , λ2 p These are hyperparameters set for the corresponding pigments; y i This is an indicator of the degree of smoke pollution in the i-th sample region corresponding to the pigment; x ij β is the reflectance of the i-th sample region corresponding to the pigment in the j-th band; j These are the regression coefficients of the pigment on the j-th band, β=[β1,β2,...,β p ] represents the regression coefficients for all bands; w j p It is the adaptive band weight of the corresponding pigment in the j-th band. ε is a small value.

[0049] In a preferred embodiment, in the hyperspectral-based method for assessing the degree of smoke damage to murals, the elastic regression network is fitted using the Lasso regression formula to obtain the regression coefficient β for each band. j The Lasso regression formula is as follows: n is the number of sample regions for a particular pigment, p is the number of wavelength bands, β = [β1, β2, ..., βp] is the regression coefficient for all wavelength bands, β j Let X be the regression coefficient for the j-th band. iLet y be the hyperspectral reflectance of the i-th sample. i Let λ be the smoke pollution level index for the i-th sample, and λ be the regularization intensity. β = [β1, β2, ..., βp] represents the regression coefficients for all bands. j The regression coefficient for the j-th band provides the basis for constructing adaptive weights. Non-zero band coefficients are extracted to accurately identify the preferred bands for pigments. The core function of the LASSO regression formula is to achieve preliminary screening of hyperspectral bands by minimizing prediction error and L1 regularization penalty, outputting the most discriminative key bands, thus providing a foundation for subsequent elastic regression networks and pollution quantification and classification.

[0050] In a preferred embodiment, the method for assessing the degree of smoke on murals based on hyperspectral imaging uses a random forest algorithm to identify pigments in the sample murals and the murals to be tested.

[0051] In a preferred embodiment, the method for assessing the degree of smoke in murals based on hyperspectral imaging involves classifying the pollution level of each test area according to a pollution level formula, wherein the pollution level formula is as follows:

[0052]

[0053] Among them, Pollution Level is the indicator of the degree of smoke pollution in the corresponding area to be tested, and τ1, τ2, and τ3 are threshold values.

[0054] In a preferred embodiment, in the hyperspectral-based method for assessing the degree of smoke in murals, τ1 = 5%, τ2 = 10%, and τ3 = 25%.

[0055] In a preferred embodiment, the hyperspectral-based method for assessing the degree of smoke damage to murals is used to obtain the hyperspectral reflectance of each sample area and each test area by processing it as follows: Radiometric correction is applied to the original hyperspectral reflectance of each sample area and each test area using the following correction formula:

[0056]

[0057] R represents the corrected reflectance; Rraw represents the original hyperspectral data; Rwhite represents the data obtained from the standard reflector in the field; Rdark represents the dark current noise data obtained with the light source off and the lens covering the image. The reflectance of the standard reflector is 99%.

[0058] In a preferred embodiment, the method for assessing the degree of smoke in murals based on hyperspectral imaging employs a minimum noise separation transform to reduce the dimensionality of the corrected reflectance.

[0059] In the conservation of ancient murals, quantifying the severity of smoke pollution is crucial for digital cleaning. Because smoke damages the complex surface conditions of murals, traditional subjective judgments are prone to significant errors. Therefore, this invention constructs a smoke pollution assessment method suitable for hyperspectral and image analysis. Utilizing forest classifiers and hyperspectral imaging techniques, different pigment regions are objectively distinguished, achieving a more accurate pigment classification and smoke pollution detection scheme. Hyperspectral analysis and improved elastic network regression are employed to obtain the sensitive wavelengths of different pigments to smoke pollution. Finally, weighted calculations based on changes in spectral reflectance are used to differentiate the degree of smoke pollution, thereby achieving a precise assessment of the smoke pollution level in murals.

[0060] The following is a specific embodiment to further illustrate the hyperspectral-based method for assessing the degree of smoke in murals provided by the present invention.

[0061] 1 Overall Flowchart

[0062] Flowchart as follows Figure 1 As shown, mural samples were first created, and spectral data were collected using a hyperspectral imager. A fumigation device was then used to simulate fumigation damage, and further fumigation spectral data was collected from the murals. The collected data underwent preprocessing. Based on the selection of fumigation-sensitive wavelengths for different pigments, weighted sensitive wavelengths were incorporated into changes in spectral reflectance. Weighted calculations were performed on these changes in spectral reflectance to assess the degree of fumigation; a greater change in reflectance indicated more severe fumigation pollution. According to traditional classifications of fumigation pollution levels, the degree of fumigation was categorized as unpolluted, lightly polluted, moderately polluted, and heavily polluted.

[0063] 2. Sample Preparation and Data Acquisition

[0064] First, mural test blocks were prepared. Based on ancient mural production materials and techniques, the test blocks, measuring 30cm x 30cm x 1.5cm, consisted of a 2 / 3 coarse clay layer, a 1 / 3 fine clay layer, and a base color layer approximately 0.5mm thick. The coarse clay layer used a 2:1 ratio of clay to sand, with 3% (by weight) of wheat straw approximately 1cm long. The fine clay layer used a 2:1 ratio of clay to sand, with 3% (by weight) of hemp fibers. After the test blocks dried, a mixture of calcite powder and 5% gelatin was applied to the surface of the fine clay layer as a color layer. Mineral pigments were then used to draw the patterns on the base color layer. Spectral data was collected using a hyperspectral imager.

[0065] A smoke-casting experiment was then conducted. The smoke-casting apparatus was 73 cm high and 34 cm wide. Combustion material was placed at the bottom of the apparatus. A mesh iron disc was then placed in the upper middle part of the mural, and the sample was placed upright on the disc. To simulate a smoke-casting environment, sawdust, charcoal, candle granules, and incense were used as combustion materials to continuously smoke the samples. As the treatment time increased, the sample surface was visually covered with varying degrees of smoke. After the smoke-casting treatment was completed, spectral data were collected again from these smoke-cast mural samples.

[0066] 3 Data Preprocessing

[0067] Radiometric correction and dimensionality reduction of hyperspectral data. The raw data acquired by a hyperspectral imaging system is radiance, which varies depending on the quantization bit depth of different systems. Radiance refers to the radiant energy reflected by a target and received by a sensor per unit area, unit time, and unit solid angle in a certain spatial direction. Even for the same target point, it varies with the incident energy. However, the reflectivity of a material is usually unique and independent of external illumination, often used to study the natural properties of a target. Therefore, before image processing and analysis, radiometric correction is performed on the raw hyperspectral data. The correction formula is:

[0068]

[0069] Where R represents the reflectance-corrected data; Rraw represents the original hyperspectral data of the mural; Rwhite represents the data obtained on-site from the standard reflector; and Rdark represents the dark current noise data obtained with the light source off and the lens covering the image. The reflectance of the standard reflector is 99%.

[0070] Dimensionality reduction of hyperspectral data was achieved by using Minimum Noise Fraction Rotation (MNF) to reduce the dimensionality of the preprocessed hyperspectral image of the mural. MNF is a commonly used dimensionality reduction method in hyperspectral data processing. This transformation uses two principal component analyses to transform the noise covariance matrix and the noise-whitened data, preserving the principal components with a high signal-to-noise ratio, thus achieving dimensionality reduction of the hyperspectral data.

[0071] 4-band optimization

[0072] (1) Classification of mural pigments

[0073] To improve the accuracy of smoke pollution assessment, a random forest was first used to partition the pigments, and then band selection was performed separately within each pigment region to ensure that the optimal bands were used for smoke pollution analysis in different pigment regions. This avoids confusion caused by the spectral characteristics of different pigments and improves the accuracy of band selection.

[0074] Random Forest is an ensemble learning method that consists of multiple decision trees, each classifying the input data, and a final classification result is obtained through a voting mechanism. The goal of the RF classification model is:

[0075]

[0076] f(X) is the final predicted pigment category. T is the total number of decision trees. h t (X) is the classification result of the t-th decision tree, i.e., the pigment category.

[0077] For a single decision tree, its classification decision is:

[0078] h t (x)=c t ,c t ∈{1,2,...,c} (3)

[0079] h t (x) is the classification result of the t-th decision tree, i.e., the predicted pigment category; c t is the category predicted by the decision tree for sample X. c is the total number of pigment categories (e.g., cinnabar, loess, ultramarine, malachite).

[0080] In Random Forest (RF), given a hyperspectral sample X, its final classification result is given by the following formula:

[0081]

[0082] Where 1(ht(X)=c) is the index function:

[0083]

[0084] The final classification results of the random forest The classification of pigments is determined by a vote across all decision trees, with the category receiving the most votes ultimately chosen. Combining hyperspectral data allows for accurate differentiation of different pigments in the murals, providing fundamental data support for smoke pollution assessment.

[0085] (2) Band selection

[0086] An improved pigment partitioning Elastic Net regression method was employed to screen the spectral bands in which different pigments are most sensitive to smoke pollution. Traditional Elastic Net is a regression method that combines LASSO (L1 regularization) and Ridge (L2 regularization). By simultaneously incorporating L1 and L2 regularization terms into the loss function, it addresses both feature selection and multicollinearity issues, demonstrating good adaptability to high-dimensional data (such as hyperspectral data analysis).

[0087] In traditional Elastic Nets, L1 and L2 regularization are uniform, and all features are subject to the same constraints. However, in smoke-related diseases, certain bands are more important to smoke pollution and should be retained, while regularization can be reduced for bands that are not closely related to the target variable, making them easier to remove.

[0088] In the hyperspectral analysis of smoked murals, different pigments exhibit different spectral characteristics. Therefore, directly using a standard Elastic Net may overlook the specificity between pigments, leading to inaccurate band selection and affecting the identification and assessment of smoke pollution. Pigment-Specific Elastic Nets, by first classifying the pigments and then training the Elastic Net separately for each pigment region, can improve the accuracy of band selection and the robustness of pollution assessment.

[0089] The optimization objective function for pigment partitioning Elastic Net is:

[0090]

[0091] n represents the number of samples for a specific pigment (such as cinnabar), rather than the total number of samples in the entire dataset. This way, Elastic Net analyzes only the bands of that pigment region, without being affected by other pigments; λ1 p , λ2 p These are hyperparameters set for different pigments (p), allowing for different L1 / L2 intensities to be set for different pigments; y i It refers to the degree of smoke pollution; x ij It is the value of each sample in the j-th band; β j These are regression coefficients, representing the contribution of each band to the target variable; w j p It is an adaptive band weighting for a specific pigment, which can adjust the regularization intensity according to the spectral characteristics of the pigment in different bands. ε is a small value to prevent division by zero errors.

[0092] Lasso regression was used to initially fit the data, and the regression coefficients for each band were obtained.

[0093]

[0094] 5. Smoke level assessment

[0095] Spectral Reflectance Change (SRC) is an important indicator for assessing smoke pollution, often used to quantify the difference in spectral reflectance between polluted and unpolluted areas. A higher SRC value indicates a more severe degree of pollution.

[0096] SRC λ =R clean (λ)-R smoke (λ) (8)

[0097] Although the traditional SRC calculation method is simple and intuitive, it has certain limitations. In order to reduce robustness and improve calculation accuracy, we can optimize and improve it.

[0098] Traditional SRC calculations assume all wavelengths are equally important, but in hyperspectral data, certain wavelengths may be more sensitive to smoke contamination. Therefore, a weighting factor w can be introduced into the band-optimization step. j p :

[0099] SRC λ p =w j p ·(R clean (λ)-R smoke (λ)) (9)

[0100] By combining pigment partitioning with Elastic Net, the accuracy and adaptability of SRC calculations are improved, and confusion between different pigments can be avoided, thereby improving the accuracy of smoke pollution assessment.

[0101] 6. Classification of Smoke Degree

[0102] First, pollution levels need to be classified based on the calculated SRC values. Pollution levels are defined as follows:

[0103]

[0104] τ1, τ2, and τ3 are threshold values, with pollution levels 0, 1, 2, 3, and 4 representing no pollution, light pollution, moderate pollution, and heavy pollution, respectively. A reflectance decrease of less than 5% is defined as no pollution; a reflectance decrease of 5%-10% with slight color changes is defined as light smoke; a reflectance decrease of 10%-25% with some pigment layer affected is defined as moderate smoke; and a reflectance decrease of >25% with almost the entire pigment layer covered is defined as heavy smoke. τ1 = 5%, τ2 = 10%, and τ3 = 25%.

[0105] In summary, the present invention has at least the following beneficial effects:

[0106] 1) Non-destructive and efficient identification: This technical solution utilizes hyperspectral imaging technology to achieve pixel-level classification and accurate pollution detection, which can meet the identification needs of smoke-stained pigments and avoid secondary damage to murals.

[0107] 2) Accuracy and robustness: The improved elastic regression network can adapt to the spectral characteristics of different pigments, determine the preferred wavelengths for different pigments, and determine the weights of the preferred wavelengths, thereby improving the accuracy and robustness of smoke pollution detection.

[0108] 3) Based on the assessment of the degree of smoke pollution, a comprehensive assessment of smoke pollution can be made, providing support for subsequent digital cleaning of smoked murals and assisting in mural restoration decisions.

[0109] Although embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. It can be applied to various fields suitable for the present invention. Other modifications can be readily made by those skilled in the art. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details and examples shown and described herein.

Claims

1. A method for assessing the degree of smoke damage to murals based on hyperspectral imaging, characterized in that, include: Pigment identification is performed on the sample mural, and the sample mural is divided into several sample areas, wherein two adjacent sample areas correspond to different pigments, and the same pigment corresponds to multiple sample areas; The hyperspectral reflectance and smoke pollution level indicators of each sample area were obtained. The hyperspectral data of each sample area included reflectance in multiple bands. The hyperspectral reflectance and smoke pollution level indices of multiple sample regions corresponding to the same pigment were processed using an elastic regression network to determine the preferred bands for each pigment and the adaptive band weights of the preferred bands. Pigment identification is performed on the mural to be tested, and the mural to be tested is divided into several test areas, wherein two adjacent test areas correspond to different pigments; Obtain the hyperspectral reflectance of each region to be measured; The degree of smoke pollution in each test area was calculated using the formula for changes in spectral reflectance. The formula for the change in spectral reflectance is as follows: in, Represents the area to be tested High spectral reflectance, The hyperspectral reflectance represents the uncontaminated area of ​​the mural being tested. Represents the area to be tested The adaptive band weight of the preferred band J of the corresponding pigment, where p represents the region to be measured. The number of wavelengths corresponding to the pigment.

2. The method for assessing the degree of smoke damage to murals based on hyperspectral imaging as described in claim 1, characterized in that, The optimization objective function of the elastic regression network is as follows: in, This represents the number of sample regions corresponding to a particular pigment. These are hyperparameters set for the specific pigment; It is the smoke pollution level index of the i-th sample area corresponding to the pigment; It is the reflectance of the i-th sample region corresponding to the pigment in the j-th band; These are the regression coefficients of the pigment on the j-th band, β=[β1,β2,...,β p ] represents the regression coefficients for all wavebands; It is the adaptive band weight of the corresponding pigment in the j-th band. , It is a small value.

3. The method for assessing the degree of smoke damage to murals based on hyperspectral imaging as described in claim 2, characterized in that, The elastic regression network was fitted using the Lasso regression formula to obtain the regression coefficients for each band. The Lasso regression formula is as follows: , Let be the hyperspectral reflectance of the i-th sample. This represents the regularization strength.

4. The method for assessing the degree of smoke damage to murals based on hyperspectral imaging as described in claim 1, characterized in that, The random forest algorithm was used to identify pigments in the sample murals and the murals to be tested.

5. The method for assessing the degree of smoke damage to murals based on hyperspectral imaging as described in claim 1, characterized in that, The hyperspectral reflectance of each sample region and each test region was obtained by processing the raw hyperspectral reflectance of each sample region and each test region using the following method: Radiometric correction was performed on the original hyperspectral reflectance of each sample region and each test region, and the correction formula is as follows: R is the corrected reflectance. raw This is the raw hyperspectral data; white Standard reflector data obtained on-site; 𝑅 dark The dark current noise data was obtained with the light source off and the lens covering the image. The reflectivity of the standard reflector is 99%.

6. The method for assessing the degree of smoke damage to murals based on hyperspectral imaging as described in claim 5, characterized in that, The dimensionality of the corrected reflectivity is reduced by using a minimum noise separation transform.

Images