Method for evaluating stone carving degradation mode by using hyperspectral imaging technology

Abstract

As a precious historical cultural heritage, the deterioration mode evaluation of the cultural relic stone sculpture is crucial to protection work. A traditional evaluation method depends on manual observation and detection and is low in efficiency and high in subjectivity. The embodiment of the invention relates to the technical field of cultural relic stone carving degradation mode evaluation, and discloses an innovative method. Firstly, collected samples are subjected to microscopic observation and mineral detection, and degradation characteristics are determined; a portable hyperspectral imaging system is used for collecting and calibrating an image of a to-be-evaluated area, and spectral features are extracted, built into a database and processed; a spectral index screening degradation mode is determined through spectral index analysis and image gray scale response correlation and optimization wavelength combination; a WOA-XGBoost algorithm is adopted to train a model, and a stone carving degradation mode is identified to draw a distribution cloud picture; and calculating a weathering coefficient by combining an X-ray fluorescence spectrum result, and evaluating the degradation degree according to the degradation model proportion and the coefficient in the region. According to the method, intelligent identification and rapid evaluation are realized, the labor cost is saved, and the automatic process of cultural relic stone carving exploration is promoted.

Description

Technical Field

[0001] The embodiment of the present invention relates to the technical field of cultural relic stone carving degradation pattern assessment, and discloses a method for assessing stone carving degradation patterns using hyperspectral imaging technology. Background Art

[0002] Rock carvings exposed outdoors for a long time are inevitably affected by rock materials, natural environment and human factors. Even if most rock carvings are well preserved, the matrix rock is damaged by deterioration modes such as flaking, crusting, salt crystallization and biological colonization. These factors make the existing deterioration more serious. That is why many scholars have been focusing on the study of deterioration modes.

[0003] However, due to the severe impact of temperature and humidity on stone carvings, the quantitative analysis of deterioration patterns is limited. In addition, the current research on different deterioration patterns is independent and the corresponding system framework is vague. Therefore, developing a deterioration assessment method is the current trend. Summary of the invention

[0004] In view of the above problems, an embodiment of the present invention relates to the technical field of cultural relic stone carving degradation pattern assessment, and discloses a method for assessing stone carving degradation patterns using hyperspectral imaging technology, which is used to better quantify the impact of different degradation patterns on rock carvings.

[0005] According to one aspect of an embodiment of the present invention, a method for evaluating the degradation pattern of stone carvings using hyperspectral imaging technology is provided. Figure 1 As shown, the method includes:

[0006] Step 1: Use polarization and stereo microscopes to observe the microstructure of the collected samples, record the morphology, distribution and crack development of mineral crystals; detect the mineral composition through X-ray diffraction to determine the characteristics of the deterioration pattern of stone cultural relics;

[0007] Step 2: Select a portable hyperspectral imaging system, which includes a near-infrared spectrometer, a movable tripod, a data transmission device, a gray panel, and a computer. According to the built-in standard operating procedures of the system, collect hyperspectral images of the area to be evaluated; after collection, calibrate the reflectivity of the collected images according to the preset reflectivity calibration algorithm;

[0008] Step 3: Extract the spectral features of different degradation modes and establish a spectral library; then normalize the spectral data, map the reflectance to a range between 0 and 1, and use the Savitzky-Golay filter algorithm to smooth the spectral data;

[0009] Step 4: Establish a spectral index and analyze the correlation between the spectral characteristics and the grayscale response of the image of the degradation pattern of the cultural relic stone carving;

[0010] Step 5: Add a third wavelength to the spectral index calculated with two wavelengths, propose a new improved spectral index about three wavelengths with reference to the similar form of the three wavelengths, and apply the entire data set to find the optimal wavelength combination and two correction factors;

[0011] Step 6: Perform a preliminary screening of the degradation mode by selecting the spectral index with the highest correlation;

[0012] Step 7: Use the Whale Optimized Extreme Gradient Boosting Algorithm (WOA-XGBoost) to train the degradation mode model;

[0013] Step 8: Use the trained model to identify the stone carving degradation pattern and draw a degradation distribution cloud map;

[0014] Step 9: Calculate the weathering coefficient of different diseases based on the X-ray fluorescence spectrum results;

[0015] Step 10: Evaluate the degree of degradation of the region based on the proportion of different degradation models in the region and the weathering coefficient.

[0016] In an optional manner, in step 1, an Olympus SZX16 stereo microscope is used to observe the microstructure of rock samples under different degradation conditions, an Olympus polarizing microscope BH-2 is used to measure rock thin section identification, Cu-Kα radiation is used to detect powdered rock samples by X-ray diffraction, and X-ray melting spectroscopy is used to analyze chemical composition. The Olympus SZX16 stereo microscope and the Olympus polarizing microscope BH-2 are both high-performance microscopes used in biology, materials science, geology and other fields. Cu-Kα radiation is one of the most commonly used radiation sources in X-ray diffraction analysis. Because of its stability and wide application, it has become an important analytical tool in materials science, geology, biology and other fields. X-ray melting spectroscopy is an analytical method for determining the chemical composition of materials. It identifies and quantifies the elements in a sample by measuring the wavelength and intensity of fluorescent X-rays generated by the sample under X-ray irradiation.

[0017] In an optional manner, a portable hyperspectral imaging system is used in step 2 to collect hyperspectral information of the sample, and the collection system includes a near-infrared spectrometer, a movable tripod, a data transmission device, a gray panel, and a computer. The near-infrared spectrometer completes the analysis of the sample within a few seconds to a few minutes, which is very suitable for application scenarios that require rapid detection, and does not destroy the sample during the analysis process. It is suitable for precious or non-renewable sample analysis, and simultaneously analyzes multiple chemical components in the sample, which improves the analysis efficiency. The design is compact and easy to carry and use. A 25cm×25cm gray panel with a reflectivity of 30%. The gray panel is used to calibrate the spectral image because the white panel with a reflectivity of 99% sometimes seems overexposed.

[0018] In an optional manner, the reflectivity is calibrated in step 2 to reduce the influence of dark current and noise generated by illumination. The formula for calibrating the reflectivity is:

[0019]

[0020] where i is the pixel index, M0 is the reflectance of the original hyperspectral image, and C i is the refraction image of the dark current, N i is a reference image of the grey panel.

[0021] In an optional manner, in step 3, the spectral data is normalized to map the reflectance to a range between 0 and 1, and the normalized spectral reflectance is processed by a Savitzky-Golay filter. The Savitzky-Golay filter can effectively smooth the data and remove high-frequency noise, and can adapt to different types of signals and data by adjusting the polynomial order and window size. Compared with a simple moving average filter, the Savitzky-Golay filter can better retain the peaks, valleys, and inflection points of the signal while smoothing the data.

[0022] In an optional manner, the normalization process in step 3 is a smoothing method based on local polynomial least squares fitting, and its calculation formula is expressed as:

[0023]

[0024] where y j * is the smoothed spectral data, w i is the weight of the i value in the moving window smoothing, z is the size of the smoothing window, and D is the moving window with a value of 2z+1.

[0025] In an optional manner, in step 4, the collected image is subjected to noise reduction processing to extract the image grayscale features, the grayscale value of the image is between 0-255, and the spectral reflectance corresponds to the grayscale value of the same area of ​​the image, which facilitates correlation analysis.

[0026] In an optional manner, the image is denoised using a mean filtering method in step 4. Mean filtering is a commonly used image processing technique, which is mainly used to remove noise in an image. The basic principle is to slide a window containing a number of pixels on the image and replace the value of the pixel in the center of the window with the average value of the pixels in the window.

[0027] In an optional manner, in step 5, before the correlation analysis, the image is divided into 160 regions (16×10), and the spectral reflectance and grayscale response of each region are extracted, and the correlation between the spectral characteristics and the grayscale response of the rock notch degradation pattern image is analyzed by the spectral index. Correlation analysis is a statistical method used to measure the strength and direction of the relationship between two or more variables. This analysis can help us understand how variables affect each other and whether their changes are consistent.

[0028] In an optional manner, the normalized difference index (FAL), the difference index (AL) and the ratio index (BL) are used in step 5 to analyze the correlation between the reflectances of the two wavelengths, and each index is calculated as follows:

[0029]

[0030] ASL=M λ1 -M2

[0031]

[0032] Among them, M λ1 is the reflectivity corresponding to wavelength λ1, M λ2 is the reflectivity corresponding to the wavelength λ2.

[0033] In an optional manner, in order to improve the sensitivity of characterizing the object characteristics in step 5, a third wavelength is added to the spectral index calculated with two wavelengths. For the similar form of three wavelengths, a new improved spectral index for three wavelengths is proposed.

[0034] In an optional manner, the step 5 proposes a new improved spectral index for three wavelengths, and the entire data set is used to find the optimal wavelength combination and two correction factors. The three wavelength normalized difference index (FNF), the three wavelength difference index (FFI) and the three wavelength ratio index (FMI) are expressed as:

[0035] FFI=M λ1 -M λ2 -αM λ3

[0036]

[0037] Where α, β are correction factors, which range from -3 to 3 (increments of 0.5), M λ1 ≠M λ2 ≠M λ3 .

[0038] In an optional manner, in step 6, the degradation mode is screened a second time by comparing several spectral indices and selecting the index with the highest correlation to calculate the response of the hyperspectral image.

[0039] In an optional manner, in step 7, a stone carving degradation assessment model is constructed on the Xgboost model. The model is an optimized distributed gradient boosting decision tree algorithm that supports distributed computing, can process large-scale data sets, can automatically process missing values, and reduces the complexity of data preprocessing. When training the model, the built-in parameters of the model are automatically optimized using the whale optimization algorithm (WOA) with a global optimization solution.

[0040] In an optional manner, the parameters in step 7 include: general parameters, booster parameters, and learning task parameters.

[0041] In an optional manner, the trained model is used in step 8 to identify the degradation pattern of the stone carving and draw a degradation distribution cloud map. Each pixel needs to be classified to obtain a degradation pattern distribution cloud map, and the area proportion of each degradation pattern is calculated. After the trained model is identified, each pixel in the image seems to be different due to degradation characteristics, and each color represents a degradation pattern.

[0042] In an optional manner, the weathering index (MI) in step 9 is calculated as follows:

[0043]

[0044] Where H represents the molar ratio of the chemical components. It can be calculated as follows:

[0045]

[0046] The degradation index (DI) is calculated based on different degradation mode coefficients and the degradation mode ratio in the region. The calculation formula is as follows:

[0047]

[0048] where D is the percentage of deterioration patterns identified from the hyperspectral imagery.

[0049] In an optional manner, in step 10, the degradation degree of the region is evaluated according to the proportion of different degradation models in the region combined with the weathering coefficient.

[0050] Compared with the prior art, the advantages and beneficial effects of the present invention are: by detecting the collected samples through microscopes, ray diffraction, etc., combining microscopic and mineralogical characteristics, qualitatively screening and extracting degradation characteristics, and by introducing dynamic correction, a new spectral index is proposed to characterize the correlation between image grayscale and spectral reflectance, and the optimal wavelength combination is applied to identify the distribution of degradation patterns, which can achieve quantitative screening of degradation patterns, analyze the intelligent model used to identify the degradation pattern of stone carvings, and judge its accuracy, and by integrating the degradation index reflected by the chemical composition with the distribution ratio of the degradation pattern determined by the spectral response, quantify the impact of different degradation patterns on rock carvings. Rapid and high-precision evaluation of the degradation pattern of cultural relics and stone carvings is achieved, which saves workload and has high evaluation accuracy, providing data support for the protection of cultural relics and stone carvings.

[0051] The above description is only an overview of the technical solution of the embodiment of the present invention. In order to more clearly understand the technical means of the embodiment of the present invention, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the embodiment of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] The accompanying drawings are only used to illustrate the embodiments and are not to be considered as limiting the present invention. In addition, the same reference symbols are used to represent the same components throughout the accompanying drawings. In the accompanying drawings:

[0053] Figure 1 A schematic diagram of a flow chart of an embodiment of the present invention is shown.

[0054] Figure 2 A flowchart of establishing a degradation mode assessment model in an embodiment of the present invention is shown.

[0055] Figure 3 The figure shows the evaluation result of the stone carving degradation pattern in the embodiment of the present invention. DETAILED DESCRIPTION

[0056] Exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be limited to the embodiments set forth herein.

[0057] A method for assessing the degradation pattern of stone carvings using hyperspectral imaging technology, as shown in the attached Figure 1 and attached Figure 2 As shown, the following steps are included:

[0058] Step 1: Use polarization and stereo microscopes to observe the microstructure of the collected samples, record the mineral crystal morphology, distribution and crack development; detect the mineral composition through X-ray diffraction to determine the characteristics of the deterioration pattern of stone cultural relics.

[0059] In this step, an Olympus SZX16 stereo microscope was used to observe the microstructure of rock samples under different degradation conditions, and an Olympus polarizing microscope BH-2 was used to measure rock thin section identification. For the mineral composition of rock samples, powdered rock samples were detected by X-ray diffraction (XRD) using Cu-Kα radiation. During the test, the 2θ angle was set from 10° to 50° with a step size of 0.02° / s. The chemical composition was analyzed using X-ray melting spectrometry, and the FeO content was supplemented by wet chemical analysis.

[0060] Step 2: Select a portable hyperspectral imaging system, which includes a near-infrared spectrometer, a movable tripod, a data transmission device, a gray panel, and a computer. According to the built-in standard operating procedures of the system, collect hyperspectral images of the area to be evaluated; after collection, calibrate the reflectance of the collected images according to the preset reflectance calibration algorithm.

[0061] Hyperspectral images were collected in the wavelength range of 400-1000nm by a portable hyperspectral imaging system. The system includes a near-infrared spectrometer, a movable tripod, a data transmission device, a 25cm×25cm gray panel with a reflectivity of 30%, and a computer supported by image acquisition software. The effective resolution of the spectrum is 2048×2046 pixels, and the spectral resolution is 5nm. During the hyperspectral image acquisition process, it should be noted that data acquisition should be carried out in an environment with sufficient and stable light. In addition, the gray panel is used to calibrate the spectral image because the white panel with a reflectivity of 99% sometimes seems to be overexposed.

[0062] The reflectivity is calibrated to reduce the noise effect caused by dark current and illumination. The formula for calibrating the reflectivity is:

[0063]

[0064] where i is the pixel index, M0 is the reflectance of the original hyperspectral image, and C i is the refraction image of the dark current, N i is a reference image of the grey panel.

[0065] Step 3: Extract the spectral features of different degradation modes and establish a spectral library; then normalize the spectral data, map the reflectance to the range between 0 and 1, and use the Savitzky-Golay filtering algorithm to smooth the spectral data.

[0066] This operation improves the convergence speed of model training and can eliminate the influence of data dimensions and differences in value ranges. The normalized spectral reflectance is processed by a Savitzky-Golay filter. The Savitzky-Golay filter can effectively smooth data and remove high-frequency noise. It can adapt to different types of signals and data by adjusting the polynomial order and window size. Compared with a simple moving average filter, the Savitzky-Golay filter can better retain the peaks, valleys, and inflection points of the signal while smoothing the data.

[0067] Normalization is a smoothing method based on local polynomial least squares fitting, and its calculation formula is expressed as:

[0068] where y j * is the smoothed spectral data, w i is the weight of the i value in the moving window smoothing, z is the size of the smoothing window, and D is the moving window with a value of 2z+1.

[0069] Step 4: Establish a spectral index and analyze the correlation between the spectral characteristics and the grayscale response of the image of the cultural relic stone carving degradation pattern.

[0070] Considering that image acquisition is usually affected by ambient lighting, grayscale processing is required when acquiring image information to avoid interference from unstable lighting. The acquired image is denoised and the image grayscale features are extracted. The grayscale value of the image is between 0 and 255, and the spectral reflectance corresponds to the grayscale value of the same area of ​​the image, which is convenient for correlation analysis.

[0071] The image is denoised using the mean filtering method. The function can be expressed as:

[0072]

[0073] Where P(x, y) is the color feature value of the pixel at the coordinate (x, y) after filtering; M is the neighborhood radius; f(x', y') is the color feature value of the pixel at the coordinate (x', y').

[0074] Mean filtering is a commonly used image processing technique, mainly used to remove noise from images. Its basic principle is to slide a window containing several pixels on the image and replace the value of the pixel in the center of the window with the average value of the pixels in the window.

[0075] The image after noise removal will be grayed out as follows:

[0076] G(x,y)=aR(x,y)+bG(x,y)+aB(x,y)

[0077] Where G(x,y) is the grayscale value of the image; a, b and c are constants that can be processed by arithmetic averaging and the value can be set to 1 / 3.

[0078] Step 5: Add a third wavelength to the spectral index calculated with two wavelengths, refer to the similar form of the three wavelengths, propose a new improved spectral index about the three wavelengths, and apply the entire data set to find the optimal wavelength combination and two correction factors. Before the correlation analysis, the image was divided into 160 regions (16×10), and the spectral reflectance and grayscale response of each region were extracted. The correlation between the spectral characteristics and the grayscale response of the rock scratch degradation pattern image was analyzed by the spectral index. Correlation analysis is a statistical method used to measure the strength and direction of the relationship between two or more variables. This analysis can help us understand how variables affect each other and whether their changes are consistent.

[0079] In this step, the normalized difference index (FAL), difference index (AL) and ratio index (BL) are used to analyze the correlation of the reflectance of the two wavelengths. Each index is calculated as follows:

[0080]

[0081] ASL=M λ1 -M λ2

[0082]

[0083] Among them, M λ1 is the reflectivity corresponding to wavelength λ1, M λ2 is the reflectivity corresponding to the wavelength λ2.

[0084] In order to improve the sensitivity of characterizing the characteristics of the object, a third wavelength is added to the spectral index calculated with two wavelengths. A new improved spectral index about three wavelengths is proposed for the similar form of three wavelengths. The entire data set is used to find the optimal wavelength combination and two correction factors to improve the estimation accuracy of the index and enhance the stability of the factor correlation. The three-wavelength normalized difference index (FNF), three-wavelength difference index (FFI) and three-wavelength ratio index (FMI) are expressed as:

[0085]

[0086] FFI=M λ1 -M λ2 -αM λ3

[0087]

[0088] Where α, β are correction factors, which range from -3 to 3 (increments of 0.5), M λ1 ≠M λ2 ≠M λ3 .

[0089] FNF characterizes the relative changes in spectral reflectance at different wavelengths by calculating the ratio of the difference in reflectance at two specific wavelengths to the sum of the reflectances, thereby reflecting the correlation with a certain feature (such as the degradation pattern of rock carvings). FFI calculates the difference in reflectance at two wavelengths to measure the absolute difference in spectral reflectance at these two wavelengths. FMI uses the ratio of the reflectance of two wavelengths as a measurement standard to reflect the proportional relationship between the reflectances of different wavelengths.

[0090] By comparing the correlation coefficients under different correction factor values, the wavelength combination that makes the spectral index most correlated with the image grayscale and spectral reflectance is determined (for example, when FNF is α = -1, β = -1.5, the corresponding wavelengths of 745nm, 750nm and 670nm are the optimal combination). This shows that the correction factor helps to screen out wavelengths that are more sensitive to degradation patterns, thereby improving the accuracy of spectral index in identifying degradation patterns.

[0091] Step 6: Perform a second screening of degradation patterns by comparing several spectral indices and selecting the one with the highest correlation to calculate the response of the hyperspectral image.

[0092] Step 7: Build a stone carving degradation assessment model based on the XGboost model. The model supports distributed computing, can handle large-scale data sets, can automatically handle missing values, and reduces the complexity of data preprocessing. When training the model, the built-in parameters of the model are automatically optimized using the Whale Optimization Algorithm (WOA) with a global optimization solution.

[0093] The parameters include: general parameters, booster parameters, and learning task parameters.

[0094] Whale Optimized Extreme Gradient Boosting (WOA-XGBoost) is a machine learning method that combines the Whale Optimization Algorithm (WOA) and the Extreme Gradient Boosting (XGBoost) algorithm to optimize the parameters of classification models to improve classification accuracy.

[0095] Step 8: Use the trained model to identify the degradation pattern of the stone carving and draw a degradation distribution cloud map. It is necessary to classify each pixel to obtain the degradation pattern distribution cloud map and calculate the area proportion of each degradation pattern. After the trained model is identified, each pixel in the image seems to be different due to degradation characteristics, and each color represents a degradation pattern.

[0096] Step 9: Calculate the weathering coefficient of different diseases based on the X-ray fluorescence spectrum results.

[0097] The weathering index (MI) is calculated as follows:

[0098]

[0099] Where H represents the molar ratio of the chemical components. It can be calculated as follows:

[0100]

[0101] The degradation index (DI) is calculated based on different degradation mode coefficients and the degradation mode ratio in the region. The calculation formula is as follows:

[0102]

[0103] where D is the percentage of deterioration patterns identified from the hyperspectral imagery.

[0104] Step 10: Evaluate the degree of degradation of the region based on the proportion of different degradation models in the region and the weathering coefficient.

[0105] Below is an example of how hyperspectral imaging can be used to assess degradation patterns:

[0106] The study area is the Dazu Rock Carvings in the humid southern region of China. The region has a subtropical warm and humid monsoon climate, with an average annual temperature of 17.2℃ and an average annual rainfall of 200mm more than evaporation. The region is usually characterized by narrow valleys and deep hills, with an altitude range of 470 to 530m and a terrain height difference of about 60m. The geological characteristics of the region are that the overburden is mainly loose miscellaneous fill accumulation, and all exposed strata are sedimentary rocks; in the bedrock distribution, the upper 20m is mainly fine-grained sandstone, and the lower layer is mainly mudstone and medium-thick interlayered sandstone. There is no large area of ​​surface water in the region, and the groundwater mainly comes from atmospheric rainfall infiltration. It can be divided into pore water in the overburden and fissure water in the bedrock cracks according to the location of existence. Sampling was carried out in the area within 500m of the rock carvings and in the same strata as the Jurassic sandstone and mudstone. The selected site needs to meet the requirements of no rock carvings and obvious deterioration. Considering that the Dazu Rock Carvings have been restored and preserved many times in history, the use of materials was avoided as much as possible due to the metamorphic phenomena of the original rock mass and the large number of preserved rock faces. A total of 50 samples were collected for testing according to the four defined degradation modes and the original state. The different degradation modes of the magma were analyzed using microlithology, mineralogy and major elements. The microstructure of the rock samples under different degradation conditions was observed using an Olympus SZX16 stereo microscope. The rock thin section identification was measured using an Olympus polarizing microscope BH-2. For the minimum composition of the rock samples, the powdered rock samples were detected by X-ray diffraction using Cu-Kα radiation. During the test, the 2θ angle was set from 10° to 50° with a step size of 0.02° / s. The chemical composition was analyzed using the alkali fusion method of X-ray fluorescence spectroscopy. The content of FeO was supplemented by wet chemical analysis. The same type of samples were tested three times.

[0107] The test equipment is a portable hyperspectral imaging system, and the hyperspectral images are collected by the portable hyperspectral imaging system in the wavelength range of 400-1000nm. The effective resolution of the spectrometer is 2048×2046 pixels, and the spectral resolution is 5nm. During the acquisition of hyperspectral images, it should be noted that data acquisition should be carried out in an environment with sufficient and stable light. A 25cm×25cm gray panel with a reflectivity of 30% is used to calibrate the spectral image, because a white panel with a reflectivity of 99% sometimes seems to be overexposed. In order to make the spectral curves comparable, a series of preprocessing steps are performed on the original spectral reflectance. The reflectance of the original spectral image is calibrated, the reflectance of the calibrated hyperspectral image is extracted, and the average reflectance of all pixels corresponding to each feature is calculated. The spectral data is normalized to map the reflectance to a range between 0 and 1.

[0108] Each location of the degraded feature was tested multiple times by the same procedure, and spectral images were collected according to strict quality assurance and quality control protocols. Considering that image acquisition is usually affected by ambient lighting, grayscale processing is required when acquiring image information to avoid interference from unstable lighting.

[0109] Before the correlation analysis, the image was divided into 160 regions (16×10), and the spectral reflectance and grayscale response of each region were extracted. The correlation between the spectral features and the grayscale response of the rock carving degradation pattern image was analyzed by spectral indicators. The normalized difference index, difference index and ratio index were used to analyze the correlation by the reflectance of two wavelengths. A third wavelength was added to the spectral index calculated with two wavelengths. With reference to the similar form of the three wavelengths, a new improved spectral index about three wavelengths was proposed, and the entire data set was applied to find the optimal wavelength combination and two correction methods. The degradation pattern was screened for the second time by comparing several spectral indices and selecting the index with the highest correlation to calculate the response of the hyperspectral image.

[0110] This study used three classification models: extreme gradient boosting, random forest, and support vector machine. The built-in parameters of the classification model were optimized using the whale optimization algorithm, gray wolf optimizer, and particle swarm optimization, and the classification accuracy of various hybrid models was compared. The specific steps include: constructing a data set containing 121 bands, and its correlation with spectral reflectance and image grayscale was established. The data set was divided into a training set and a test set in a ratio of 7:3, and the classification model was constructed using the training set. The accuracy of the model for classifying new samples was verified by the test set. An objective function was constructed to optimize the built-in parameters of the classification model, and the best fitness was obtained through iteration. After ten training and testing operations, the performance of the classification model was evaluated based on the accuracy. In terms of results, among the optimized models, the correlation coefficient of the three-wavelength normalized difference index (FNF) corresponding to the WOA-XGBoost model was the highest, reaching 0.966. Compared with other indicators such as FAL (correlation coefficient of 0.845), AL (correlation coefficient of 0.761), and BL (correlation coefficient of 0.742), its correlation was significantly improved. This shows that the WOA-XGBoost model can better capture the relationship between image grayscale values ​​and spectral reflectance when using spectral index to classify degradation patterns, thereby showing better performance. After this, the WOA-XGBoost model showed better performance in degradation pattern classification. Finally, by integrating the degradation index reflected by the chemical composition with the degradation pattern distribution ratio determined by the spectral response, the impact of different degradation patterns on rock carvings was quantified. In the regional degradation evaluation of Dazu Rock Carvings, biological colonization and surface morphological changes have the highest degradation ratio and degree, which deserves attention in the protection of stone carvings in the area. The results of the stone carving degradation assessment are shown in the figure below. Figure 3 .

[0111] In the description provided herein, a large number of specific details are described. However, it is understood that embodiments of the present invention can be practiced without these specific details. Similarly, in order to simplify the present invention and help understand one or more of the various inventive aspects, in the above description of exemplary embodiments of the present invention, the various features of the embodiments of the present invention are sometimes grouped together into a single embodiment, figure, or description thereof. Wherein, the claims that follow the specific embodiment are hereby expressly incorporated into the specific embodiment, wherein each claim itself is a separate embodiment of the present invention.

[0112] Those skilled in the art will appreciate that the modules in the devices in the embodiments may be adaptively changed and arranged in one or more devices different from the embodiments. The modules or units or components in the embodiments may be combined into one module or unit or component, and further may be divided into a plurality of submodules or subunits or subcomponents. Except that at least some of such features and / or processes or units are mutually exclusive.

[0113] It should be noted that the above embodiments illustrate the present invention rather than limit it, and that those skilled in the art may devise alternative embodiments without departing from the scope of the appended claims. In the claims, any reference symbol between brackets shall not be construed as a limitation on the claims. The word "comprising" does not exclude the presence of elements or steps not listed in the claims. The word "one" or "an" preceding an element does not exclude the presence of a plurality of such elements. The present invention may be implemented by means of hardware comprising a number of different elements and by means of a suitably programmed computer. In a unit claim enumerating a number of devices, several of these devices may be embodied by the same hardware item. The use of the words first, second, and third, etc. does not indicate any order. These words may be interpreted as names. The steps in the above embodiments, unless otherwise specified, should not be understood as limitations on the order of execution.

Claims

1. A method for evaluating the degradation pattern of stone carvings using hyperspectral imaging technology, characterized in that: The method comprises: Step 1: Use polarization and stereo microscopes to observe the microstructure of the collected samples, record the morphology, distribution and crack development of mineral crystals; detect the mineral composition through X-ray diffraction to determine the characteristics of the deterioration pattern of stone cultural relics; Step 2: Select a portable hyperspectral imaging system, which includes a near-infrared spectrometer, a movable tripod, a data transmission device, a gray panel, and a computer. According to the built-in standard operating procedures of the system, collect hyperspectral images of the area to be evaluated; after collection, calibrate the reflectivity of the collected images according to the preset reflectivity calibration algorithm; Step 3: Extract the spectral features of different degradation modes and establish a spectral library; then normalize the spectral data, map the reflectance to a range between 0 and 1, and use the Savitzky-Golay filter algorithm to smooth the spectral data; Step 4: Establish a spectral index and analyze the correlation between the spectral characteristics and the grayscale response of the image of the degradation pattern of the cultural relic stone carving; Step 5: Add a third wavelength to the spectral index calculated with two wavelengths, propose a new improved spectral index about three wavelengths with reference to the similar form of the three wavelengths, and apply the entire data set to find the optimal wavelength combination and two correction factors; Step 6: Perform a preliminary screening of the degradation mode by selecting the spectral index with the highest correlation; Step 7: Use the Whale Optimized Extreme Gradient Boosting Algorithm (WOA-XGBoost) to train the degradation mode model; Step 8: Use the trained model to identify the stone carving degradation pattern and draw a degradation distribution cloud map; Step 9: Calculate the weathering coefficient of different diseases based on the X-ray fluorescence spectrum results; Step 10: Evaluate the degree of degradation of the region based on the proportion of different degradation models in the region and the weathering coefficient.

2. The method according to claim 1, characterized in that The formula for calibrating the reflectivity in step 2 is: where i is the pixel index, M0 is the reflectance of the original hyperspectral image, and C i is the refraction image of the dark current, N i is a reference image of the grey panel.

3. The method according to claim 1, characterized in that: In step 3, the normalized spectral reflectance is processed by a Savitzky-Golay filter. The Savitzky-Golay filter is a smoothing method based on local polynomial least squares fitting. It can effectively smooth data and remove high-frequency noise by performing weighted fitting on data points in a moving window. The calculation formula is expressed as follows: where y j * is the smoothed spectral data, w i is the weight of the i value in the moving window smoothing, z is the size of the smoothing window, and D is the moving window with a value of 2z+1.

4. The method according to claim 1, characterized in that: In step 4, the grayscale feature of the image is extracted. The grayscale value of the image is between 0 and 255, and the spectral reflectance corresponds to the grayscale value of the same area of ​​the image.

5. The method according to claim 1, characterized in that In step 5, before the correlation analysis, the image is divided into 160 regions (16×10), and the spectral reflectance and grayscale response of each region are extracted. The correlation between the spectral characteristics and the grayscale response of the rock notch degradation pattern image is analyzed by the spectral index.

6. The method according to claim 1, characterized in that In step 5, the normalized difference index (FAL), the difference index (AL) and the ratio index (BL) are used to analyze the correlation between the reflectances of the two wavelengths, and each index is calculated as follows: ASL=M λ1 -M λ2 Among them, M λ1 is the reflectivity corresponding to wavelength λ1, M λ2 is the reflectivity corresponding to the wavelength λ2.

7. The method according to claim 1, characterized in that In step 5, a third wavelength is added to the spectral index calculated with two wavelengths. For the similar form of the three wavelengths, a new improved spectral index about the three wavelengths is proposed. The entire data set is used to find the optimal wavelength combination and two correction factors. The three-wavelength normalized difference index (FNF), the three-wavelength difference index (FFI) and the three-wavelength ratio index (FMI) are expressed as: FFI=M λ1 –M λ2 –αM λ3 Where α, β are correction factors, which range from -3 to 3 (increments of 0.5), M λ1 ≠M λ2 ≠M λ3 .

8. The method according to claim 1, characterized in that: In step 9, each pixel needs to be classified to obtain a degradation mode distribution cloud map, and the area proportion of each degradation mode needs to be calculated.

9. The method according to claim 1, characterized in that: The weathering index (MI) in step 9 is calculated as follows: Where H represents the molar ratio of the chemical components. It can be calculated as follows:

10. The degradation index (DI) is calculated based on different degradation mode coefficients and the degradation mode ratio in the region. The calculation formula is as follows: where D is the percentage of deterioration patterns identified from the hyperspectral imagery.

Images