Mural smudging degree evaluation method based on hyperspectrum

Abstract

The invention discloses a hyperspectrum-based mural smudging degree evaluation method, which comprises the following steps of: performing pigment identification on a sample mural, and dividing the sample mural into a plurality of sample areas; processing the hyperspectral reflectivity and smoking pollution degree indexes of a plurality of sample areas of the same pigment by using an elastic regression network, and determining an optimal wave band and a self-adaptive wave band weight of each pigment; performing pigment identification on the to-be-detected mural, and dividing the to-be-detected mural into a plurality of to-be-detected areas; the smoking pollution degree value SRC lambda p of each to-be-detected area is calculated according to the following formula: SRC lambda p = wjp. (Rclear (lambda)-Rsmoke (lambda)), Rsmoke (lambda) represents the hyperspectral reflectivity of the to-be-detected area lambda, Rclear (lambda) represents the hyperspectral reflectivity of an unpolluted area in the to-be-detected mural, and wjp represents the self-adaptive waveband weight of the preferred waveband of the pigment corresponding to the to-be-detected area lambda. According to the method, the mural smudging degree can be accurately evaluated.

Description

Technical Field

[0001] The invention relates to the field of software, and in particular to a method for evaluating the smoke degree of a mural based on a hyperspectral spectrum. Background Art

[0002] As an important part of human cultural heritage, ancient murals carry profound historical and cultural values. However, due to long-term exposure to adverse factors such as environmental pollution and human activities, the surface of murals is extremely vulnerable to smoke pollution. Among them, the main causes of pollution are the smoke from daily cooking and heating, as well as the incense offered. Smoke pollution is mainly composed of carbon particles and tar substances produced during the combustion process, which is extremely harmful to murals. On the one hand, it will cause the color of the murals to darken and form a black covering layer on the surface of the murals, which greatly affects people's visual perception of the content of the murals; on the other hand, smoke particles may penetrate into the pigment layer, thereby changing the optical properties of the pigment layer and damaging the original appearance of the murals. What is more difficult is that the cleaning of smoke pollution is extremely difficult. The traditional physical cleaning method is based on the principles of "minimum intervention" and "no change to the original state of the cultural relics", but it is very likely to damage the original pigment layer during the operation, causing irreversible damage to the murals. In view of the technical problem that the smoke degree assessment of murals cannot be automatically realized in the existing technology, it is urgent to propose a method that can achieve the smoke degree assessment of murals without causing secondary damage. Summary of the invention

[0003] An object of the present invention is to solve at least the above problems and / or disadvantages and to provide at least the advantages which will be described hereinafter.

[0004] An object of the present invention is to provide a method for evaluating the smoke degree of a mural based on a hyperspectral spectrum, which can achieve an accurate evaluation of the smoke degree of a mural without causing secondary damage.

[0005] In order to achieve these purposes and other advantages according to the present invention, a method for evaluating the smoke degree of a mural based on a hyperspectral method is provided, comprising:

[0006] Perform pigment identification on the sample mural, and divide the sample mural into a plurality of sample areas, wherein two adjacent sample areas correspond to different pigments, and the same pigment corresponds to multiple sample areas;

[0007] Obtaining the hyperspectral reflectance and smoke pollution index of each sample area, wherein the hyperspectral data of each sample area includes the reflectance of multiple bands;

[0008] The elastic regression network is used to process the hyperspectral reflectance and smoke pollution index of multiple sample areas corresponding to the same pigment, and the preferred band of each pigment and the adaptive band weight of the preferred band are determined;

[0009] Perform pigment identification on the mural to be measured, and divide the mural to be measured into several regions to be measured, where two adjacent regions to be measured correspond to different pigments;

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

[0011] Calculate the degree value of soot pollution SRC of each region to be measured through the spectral reflectance change formula λ p , where the spectral reflectance change formula is as follows:

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

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

[0014] Preferably, in the method for evaluating the soot degree of a mural based on hyperspectral, the optimization objective function of the elastic regression network is as follows:

[0015]

[0016] where n p represents the number of sample regions corresponding to a certain pigment; λ 1 p 、λ 2 p are hyperparameters set for the corresponding pigment; y i is the soot pollution degree index of the i-th sample region corresponding to the corresponding pigment; x ij is the reflectance of the i-th sample region corresponding to the corresponding pigment in the j-th band; β j is the regression coefficient of the pigment corresponding to the j-th band, β = [β 1 , β 2 ,..., β p , is the regression coefficient of all bands; w j p 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 mural sootiness assessment method, the elastic regression network uses the Lasso regression formula for fitting to obtain the regression coefficient β of each band. j , where the Lasso regression formula is as follows: n is the number of sample regions for collecting a certain pigment, p is the number of bands, β = [β1, β2,..., βp], which is the regression coefficient of all bands, and β j is the regression coefficient of the j-th band, and X i is the hyperspectral reflectance of the i-th sample, and y i is the soot pollution level index of the i-th sample, and λ is the regularization strength.

[0018] Preferably, in the hyperspectral-based mural sootiness assessment method, the random forest algorithm is used to identify pigments for the sample mural and the mural to be measured.

[0019] Preferably, in the hyperspectral-based mural sootiness assessment method, based on the pollution level formula, each region to be measured is classified according to the soot pollution degree value of each region to be measured, where the pollution level formula is as follows:

[0020]

[0021] where Pollution Level is the soot pollution degree index of the corresponding region to be measured, and τ 1 , τ 2 , τ 3 are thresholds.

[0022] Preferably, in the hyperspectral-based mural sootiness assessment method, τ 1 = 5%, τ 2 = 10%, τ 3 = 25%.

[0023] Preferably, in the hyperspectral-based mural sootiness assessment method, the hyperspectral reflectance of each sample region and each region to be measured is obtained through the following method: radiometric correction is performed on the original hyperspectral reflectance of each sample region and each region to be measured, and the correction formula is as follows:

[0024]

[0025] R is the corrected reflectance, Rraw is the original hyperspectral data; Rwhite is the standard reflectance plate data obtained on-site; Rdark is the dark current noise data obtained after the light source is turned off and the lens is covered, and the reflectance of the standard reflectance plate is 99%.

[0026] Preferably, in the method for evaluating the degree of soot pollution of murals based on hyperspectral, minimum noise fraction transform is used to reduce the dimension of the corrected reflectivity.

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

[0028] 1) Non-destructive and efficient identification: This technical solution uses hyperspectral imaging technology, which can achieve pixel-level classification and accurate pollution detection, and can meet the identification requirements of soot-polluted pigments, avoiding secondary damage to the murals.

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

[0030] 3) Based on the evaluation of the degree of soot pollution, it can comprehensively evaluate the soot pollution, provide support for the subsequent digital cleaning of soot-polluted murals, and assist in the decision-making of mural restoration.

[0031] Other advantages, objectives and features of the present invention will be partially reflected by the following description, and partially will be understood by those skilled in the art through the research and practice of the present invention. Description of the Drawings

[0032] Figure 1 It is a flow chart of the method for identifying soot-polluted mural pigments based on hyperspectral in the embodiment of the present invention. Detailed Embodiments

[0033] The following further detailed description of the present invention is provided in conjunction with the drawings, so that those skilled in the art can implement it with reference to the text of the specification.

[0034] As Figure 1 shown, the present invention provides a method for evaluating the degree of soot pollution of murals based on hyperspectral, including:

[0035] Identify the pigments of the sample mural, and divide the sample mural into several sample areas, where two adjacent sample areas correspond to different pigments, and the same pigment corresponds to multiple sample areas. The mural pigment classification method can adopt traditional supervised classification, unsupervised classification, other machine learning classification methods and deep learning classification methods.

[0036] Obtain the hyperspectral reflectivity and the index of the degree of soot pollution of each sample area, where the hyperspectral data of each sample area includes the reflectivity of multiple bands. Here, the value of the degree of soot pollution of each sample area can be calculated using the spectral reflectivity change formula without considering the preferred bands (see formula 8 in the embodiment).

[0037] Process the hyperspectral reflectance and the degree index of soot pollution of multiple sample areas corresponding to the same pigment using an elastic regression network to determine the preferred wavelength band of each pigment and the adaptive band weights of the preferred wavelength band.

[0038] Perform pigment identification on the mural to be tested, and divide the mural to be tested into several areas to be tested, where two adjacent areas to be tested correspond to different pigments.

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

[0040] Calculate the degree value of soot pollution SRC of each area to be tested through the spectral reflectance change formula λ p , where the spectral reflectance change formula is as follows:

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

[0042] where R smoke (λ) represents the hyperspectral reflectance of the area to be tested at wavelength λ, R clean (λ) represents the hyperspectral reflectance of the non-polluted area in the mural to be tested, w j p represents the adaptive band weight of the preferred wavelength band of the pigment corresponding to the area to be tested at wavelength λ, and p represents the number of wavelength bands of the pigment corresponding to the area to be tested at wavelength λ.

[0043] Hyperspectral imaging is a non-contact and non-destructive testing method. Compared with traditional methods (such as chemical testing, XRF), hyperspectral imaging is a non-contact detection technology that avoids secondary damage to the mural and is applicable to large-area soot pollution analysis. Traditional visible light images can only distinguish color changes and cannot determine whether the color darkening is caused by soot or the original pigment itself is darker. Hyperspectral analysis can distinguish the soot layer and the pigment layer based on spectral characteristics. Hyperspectral imaging technology can detect pollution layers that are difficult to detect with the naked eye and accurately identify the impact of soot on different pigment layers. After the mural is smoked, its surface condition is complex, and some pollution layers are difficult to observe with the naked eye. With its high-resolution spectral detection ability, hyperspectral technology can keenly capture these subtle changes and reveal pollution situations that are invisible to the naked eye. By analyzing the spectral characteristics of different pigment layers, it can also clearly distinguish pollution from the original pigment, and then achieve precise separation of the soot layer.

[0044] In the hyperspectral analysis of smoked murals, the spectral characteristics of different pigments are different, and certain wavelengths may be more sensitive to smoked pollution. Therefore, if the elastic regression network is directly used, the specificities between pigments may be ignored, resulting in inaccurate band selection and affecting the identification and evaluation of smoked pollution. When analyzing the sample murals, by classifying the pigments first and then training the elastic regression network separately in each pigment area, the preferred bands that are more sensitive to smoked pollution for each pigment can be identified, the weights of the preferred bands can be determined, and the weights of the preferred bands corresponding to the pigments in the area to be measured are used when evaluating the mural to be measured, thereby improving the accuracy of band selection and the robustness of pollution evaluation.

[0045] This method can accurately measure the spectral characteristics of different pigment areas of the mural after smoking, extract the sensitive bands to smoked pollution, quantify the polluted areas, visually display the degree of pollution, provide accurate basis for digital cleaning, reduce human errors, and assist in the efficient and accurate restoration of ancient murals.

[0046] In a preferred embodiment, in the method for evaluating the degree of smoking of murals based on hyperspectral, the optimization objective function of the elastic regression network is as follows:

[0047]

[0048] where n p represents the number of sample areas corresponding to a certain pigment; λ 1 p 、λ 2 p is the hyperparameter set for the corresponding pigment; y i is the index of the degree of smoked pollution of the i-th sample area corresponding to the corresponding pigment; x ij is the reflectance of the i-th sample area corresponding to the corresponding pigment at the j-th band; β j is the regression coefficient of the pigment corresponding to the j-th band, β = [β 1 ,β 2 ,...,β p , which is the regression coefficient of all bands; w j p is the adaptive band weight of the corresponding pigment at the j-th band, ε is a small value.

[0049] In a preferred embodiment, in the method for evaluating the degree of smoking of murals based on hyperspectral, the elastic regression network uses the Lasso regression formula for fitting to obtain the regression coefficient β j of each band, where the Lasso regression formula is as follows: n is the number of sample regions for collecting a certain pigment, p is the number of bands, β = [β1, β2,..., βp] is the regression coefficient for all bands, and β j is the regression coefficient for the j-th band, and X i is the hyperspectral reflectance of the i-th sample, and y i is the index of the soot pollution level of the i-th sample, and λ is the regularization strength. β = [β1, β2,..., βp] is the regression coefficient for all bands, and β j is the regression coefficient for the j-th band, which provides a basis for constructing an adaptive weight subsequently. By extracting non-zero band coefficients, the preferred bands of the pigment can be accurately identified. The core role of the LASSO regression formula is to perform a preliminary screening of the hyperspectral bands by minimizing the prediction error and the L1 regularization penalty, and output the most discriminative key bands, thereby providing a basis for subsequent elastic regression networks and pollution quantification and grading.

[0050] In a preferred embodiment, in the method for evaluating the soot degree of murals based on hyperspectral, a random forest algorithm is used to identify pigments for the sample murals and the murals to be measured.

[0051] In a preferred embodiment, in the method for evaluating the soot degree of murals based on hyperspectral, each area to be measured is graded for pollution according to the soot pollution degree value of each area to be measured based on the pollution level formula, where the pollution level formula is as follows:

[0052]

[0053] where Pollution Level is the index of the soot pollution degree of the corresponding area to be measured, and τ 1 、τ 2 、τ 3 are thresholds.

[0054] In a preferred embodiment, in the method for evaluating the soot degree of murals based on hyperspectral, τ 1 = 5%, τ 2 = 10%, τ 3 = 25%.

[0055] In a preferred embodiment, in the method for evaluating the soot degree of murals based on hyperspectral, the hyperspectral reflectance of each sample region and each area to be measured is obtained through the following method: performing radiometric correction on the original hyperspectral reflectance of each sample region and each area to be measured, and the correction formula is as follows:

[0056]

[0057] R is the corrected reflectance, Rraw is the original hyperspectral data; Rwhite is the standard reflectance panel data obtained on-site; Rdark is the dark current noise data obtained with the light source turned off and the lens covered. The reflectance of the standard reflectance panel is 99%.

[0058] In a preferred embodiment, in the method for evaluating the degree of soot on murals based on hyperspectral data, minimum noise separation transform is used to reduce the dimension of the corrected reflectance.

[0059] In the protection of ancient murals, quantifying the severity of soot pollution is of great significance for digital cleaning. Due to the complex surface conditions of murals caused by soot, there are large errors in traditional subjective judgments. Based on this, the present invention constructs a method for evaluating the degree of soot suitable for hyperspectral and image analysis. Using techniques such as forest classifiers and hyperspectral imaging, objectively distinguish different pigment areas, and implement a more accurate pigment classification and soot pollution detection scheme; with the help of hyperspectral analysis and improved elastic net regression, obtain the sensitive bands of different pigments to soot pollution; finally, calculate the weighted change in spectral reflectance to distinguish the degree of soot pollution, so as to achieve an accurate assessment of the degree of soot on murals.

[0060] The following provides a specific embodiment to further illustrate the method for evaluating the degree of soot on murals based on hyperspectral data provided by the embodiments of the present invention.

[0061] 1 Overall flowchart

[0062] The flowchart is as Figure 1 shown. First, make a mural sample, use a hyperspectral imager to collect spectral data, simulate the soot disease through a soot device, and then collect the soot spectral data of the mural; preprocess the collected data; based on the selection of sensitive bands of different pigments to soot, introduce the sensitive bands weighted into the change in spectral reflectance, calculate the weighted change in spectral reflectance, and evaluate the degree of soot. The greater the change in reflectance, the more serious the soot pollution. According to the traditional classification of the degree of soot pollution, the degree of soot is divided into unpolluted, lightly polluted, moderately polluted, and severely polluted.

[0063] 2 Sample production and data collection

[0064] First, make mural test blocks. According to the materials and techniques of ancient murals, the size of the mural test block is 30cm * 30cm * 1.5cm, which is composed of 2 / 3 thick mud layer, 1 / 3 fine mud layer and a bottom color layer about 0.5mm thick. The thick mud layer uses clay and sand with a mass ratio of 2:1, and adds wheat straw with a mass fraction of 3% and a length of about 1cm. The fine mud layer uses clay and sand with a mass ratio of 2:1, and adds hemp fiber with a mass ratio of 3%. After the test block is dried, mix calcite powder and 5% gelatin and apply it on the surface of the fine mud layer as the color layer, and draw patterns with mineral pigments on the bottom color layer. Use a hyperspectral imager to collect spectral data.

[0065] Then, a smoking experiment was carried out. The smoking device was 73 cm high and 34 cm wide. A combustible was placed at the bottom of the device. Then, a net iron disc was placed in the upper-middle part of the mural, and the samples were stood on the disc. To simulate the smoking environment, we selected wood chips, carbon, candle particles, and incense as combustibles and continuously carried out smoking treatment on the samples. As the treatment time increased, until the surfaces of the samples visually showed different degrees of smoking coverage. After the smoking treatment was completed, spectral data collection was carried out again for these smoked mural samples.

[0066] 3 Data preprocessing

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

[0068]

[0069] Among them, R is the data after reflectivity correction; Rraw is the raw hyperspectral data of the mural; Rwhite is the data of the standard reflectance panel obtained on site; Rdark is the dark current noise data obtained after the light source is turned off and the lens is covered. The reflectivity of the standard reflectance panel is 99%.

[0070] Minimum noise fraction rotation (MNF) is used for dimensionality reduction of the preprocessed hyperspectral image of the mural. The MNF transform is a commonly used dimensionality reduction method in hyperspectral data processing. This transform performs two principal component analyses on the noise covariance matrix of the data and the noise-whitened data, and retains the principal components with a higher signal-to-noise ratio, thereby realizing the dimensionality reduction of hyperspectral data.

[0071] 4 Band selection

[0072] (1) Mural pigment classification

[0073] To improve the accuracy of the assessment of smoking pollution, random forest is first used for pigment zoning, and then band selection is carried out separately within each pigment area to ensure that the optimal bands are used for smoking pollution analysis in different pigment areas. This can avoid the spectral feature confusion of different pigments and improve the accuracy of band selection.

[0074] Random forest is an ensemble learning method, which consists of multiple decision trees. Each tree classifies the input data, and finally the final classification result is obtained through a voting mechanism. The goal of the RF classification model:

[0075]

[0076] f(X) is the finally predicted pigment category. T is the total number of decision trees. h t (X) is the classification result of the t-th decision tree, that is, 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, that is, the predicted pigment category; c t is the category predicted by the decision tree for the sample X. c is the total number of pigment categories (such as 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 an indicator function:

[0083]

[0084] The final classification result of the random forest is determined by the voting of all decision trees, and finally the category with the most votes is selected as the pigment classification result. Combining hyperspectral data can accurately distinguish different pigments in murals and provide basic data support for soot pollution assessment.

[0085] (2) Band optimization

[0086] An improved pigment partitioned Elastic Net (elastic net regression) is used to screen the spectral bands that are most sensitive to soot pollution for different pigments. The traditional Elastic Net is a regression method that combines LASSO (L1 regularization) and Ridge (L2 regularization). It takes into account both feature selection and multicollinearity problems by adding both L1 and L2 regularization terms to the loss function and has good adaptability in high-dimensional data (such as hyperspectral data analysis).

[0087] However, in the traditional Elastic Net, the L1 and L2 regularizations are unified, and all features are subject to the same constraints. However, in the case of smoke damage, certain bands are more important for smoke pollution and should be retained, while for those bands that have little correlation with the target variable, the regularization can be reduced to make them easier to be eliminated.

[0088] In the hyperspectral analysis of smoked murals, the spectral characteristics of different pigments are different. Therefore, directly using the standard Elastic Net may ignore the specificities between pigments, resulting in inaccurate band selection and affecting the identification and evaluation of smoke pollution. The Pigment-Specific Elastic Net can improve the accuracy of band selection and the robustness of pollution evaluation by first classifying pigments and then training the Elastic Net separately in each pigment region.

[0089] The optimized objective function of the Pigment-Specific Elastic Net:

[0090]

[0091] n represents the number of samples of a certain pigment (such as cinnabar), rather than the number of samples in the entire dataset. In this way, the Elastic Net for each pigment only analyzes the bands in that pigment region and will not be affected by other pigments; λ 1 p 、λ 2 p are hyperparameters set for different pigments p, and different L1 / L2 intensities can be set for different pigments; y i is the degree of smoke pollution; x ij is the value of each sample at the j-th band; β j is the regression coefficient, indicating the contribution of each band to the target variable; w j p is the adaptive band weight for a certain 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] Among them, Lasso regression is used to preliminarily fit the data to obtain the regression coefficient of each band.

[0093]

[0094] 5 Smoke Degree Evaluation

[0095] Spectral Reflectance Change (SRC) is an important indicator for evaluating smoked pollution and is commonly used to quantify the spectral reflectance difference between polluted and unpolluted areas. The larger the SRC value, the more severe the pollution level.

[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. To reduce robustness and calculation accuracy, we can optimize and improve it.

[0098] The traditional SRC calculation assumes that all wavelengths are equally important. However, in hyperspectral data, certain wavelengths may be more sensitive to smoked pollution. Therefore, a weighting factor w for the band selection step can be introduced. j p :

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

[0100] Combined with pigment partitioning Elastic Net, it improves the accuracy and adaptability of SRC calculation, and can avoid confusion of different pigments, improving the accuracy of smoked pollution assessment.

[0101] 6 Classification of Smoked Degree

[0102] First, it is necessary to classify the pollution according to the calculated SRC value. Define the pollution levels:

[0103]

[0104] τ 1 、τ 2 、τ 3 are thresholds. The pollution level numbers 0, 1, 2, 3, 4 represent no pollution, light pollution, moderate pollution, and heavy pollution respectively. Set a reflectance decrease of less than 5% as no pollution, a reflectance decrease of 5% - 10% with a slight color change as light smoking; a reflectance decrease of 10% - 25% with some pigment layers affected as moderate smoking; a reflectance decrease of > 25% with the pigment layer almost covered as heavy smoking. τ 1 = 5%, τ 2 = 10%, τ 3 = 25%.

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

[0106] 1) Non-destructive and efficient identification: This technical solution uses hyperspectral imaging technology to achieve pixel-level classification and precise pollution detection, meeting the identification requirements for smoked and polluted pigments and avoiding 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 bands of different pigments, and determine the weights of the preferred bands, improving the accuracy and robustness of smoked pollution detection.

[0108] 3) Based on the evaluation of the degree of smoked pollution, it can comprehensively evaluate the smoked pollution, provide support for the subsequent digital cleaning of smoked murals, and assist in the decision-making of mural restoration.

[0109] Although the embodiments of the present invention have been disclosed as above, they are not limited to the applications listed in the specification and embodiments. It can be fully applied to various fields suitable for the present invention. For those familiar with the field, additional modifications can be easily made. 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 the illustrated examples here.

Claims

1. A method for evaluating the smoke level of a mural based on hyperspectral, characterized in that: include: Perform pigment identification on the sample mural, and divide the sample mural into a plurality of sample areas, wherein two adjacent sample areas correspond to different pigments, and the same pigment corresponds to multiple sample areas; Obtaining the hyperspectral reflectance and smoke pollution index of each sample area, wherein the hyperspectral data of each sample area includes the reflectance of multiple bands; The elastic regression network is used to process the hyperspectral reflectance and smoke pollution index of multiple sample areas corresponding to the same pigment, and the preferred band of each pigment and the adaptive band weight of the preferred band are determined; Perform pigment identification on the mural to be tested, dividing the mural to be tested into a plurality of test areas, wherein two adjacent test areas correspond to different pigments; Obtaining the high spectral reflectance of each area to be measured; The smoke pollution level SRC of each area to be tested is calculated by the spectral reflectance change formula λ p , wherein the spectral reflectance change formula is as follows: SRC λ p =w j p ·(R clean (λ)-R smoke (λ)) Among them, R smoke (λ) represents the high spectral reflectance of the measured area λ, R clean (λ) represents the high spectral reflectance of the uncontaminated area in the mural to be tested, w j p represents the adaptive band weight of the preferred band of the pigment corresponding to the area to be tested λ, and p represents the number of bands of the pigment corresponding to the area to be tested λ.

2. The method for evaluating the smoke level of a mural based on hyperspectral according to claim 1, characterized in that: The optimization objective function of the elastic regression network is as follows: Among them, n p Represents the number of sample areas corresponding to a certain pigment; λ1 p ,λ2 p is a hyperparameter set for the corresponding pigment; y i is the smoke pollution index of the i-th sample area corresponding to the corresponding pigment; x ij is the reflectance of the i-th sample area corresponding to the corresponding pigment in the j-th band; β j is the regression coefficient of the pigment corresponding to the jth band, β=[β1,β2,...,β p ], is the regression coefficient of all bands; w j p is the adaptive band weight of the corresponding pigment in the jth band, ε is a small value.

3. The method for evaluating the smoke level of a mural based on hyperspectral according to claim 2, characterized in that: The elastic regression network is fitted using the Lasso regression formula to obtain the regression coefficient β for each band j , where the Lasso regression formula is as follows: n is the number of sample areas for a certain pigment, p is the number of bands, β = [β1, β2, ..., βp] is the regression coefficient of all bands, β j is the regression coefficient of the jth band, X i is the high spectral reflectance of the i-th sample, y i is the smoke pollution level index of the i-th sample, and λ is the regularization strength.

4. The method for evaluating the smoke level of a mural based on hyperspectral according to claim 1, characterized in that: The random forest algorithm is used to identify the pigments of the sample murals and the murals to be tested.

5. The method for evaluating the smoke level of a mural based on hyperspectral according to claim 1, characterized in that: Based on the pollution level formula, pollution level is graded for each area to be tested according to the smoke pollution level value of each area to be tested, wherein the pollution level formula is as follows: Among them, Pollution Level is the smoke pollution index of the corresponding area to be tested, and τ1, τ2, and τ3 are thresholds.

6. The method for evaluating the smoke level of a mural based on hyperspectral according to claim 5, characterized in that: τ1=5%, τ2=10%, τ3=25%.

7. The method for evaluating the smoke level of a mural based on hyperspectral according to claim 1, characterized in that: The hyperspectral reflectance of each sample area and each area to be tested is obtained by the following method: radiation correction is performed on the original hyperspectral reflectance of each sample area and each area to be tested. The correction formula is as follows: R is the corrected reflectivity, Rraw is the original hyperspectral data; Rwhite is the standard reflector data obtained on site; Rdark is the dark current noise data obtained after the light source is turned off and the lens is covered. The reflectivity of the standard reflector is 99%.

8. The method for evaluating the smoke level of a mural based on hyperspectral according to claim 7, characterized in that: The minimum noise separation transform is used to reduce the dimension of the corrected reflectivity.

Images