A system and method for identifying and evaluating cultural relics defects based on image analysis

Abstract

The present invention relates to the field of image-based cultural relic disease identification and assessment, and specifically relates to a cultural relic disease identification and assessment system and method based on image analysis. The system comprises a multispectral image acquisition module, an image processing module, a disease identification module, a restoration assessment module, a decision optimization module, and a disease database. The multispectral image acquisition module optimizes the light source by using a reflective area detection algorithm to acquire a multispectral image. The image processing module uses a multi-band collaborative filtering algorithm to eliminate noise, extract fusion features, and divide disease areas. The disease identification module uses a connected region labeling algorithm to determine the location of the disease and classifies the disease with the help of a neural symbolic model. The restoration assessment module outputs a quantitative score. The decision optimization module uses a multi-objective optimization model introduced by a genetic algorithm to weigh the restoration effect, cost, and time, and output an optimal solution. The present invention improves the accuracy of disease identification and the automation and scientific nature of restoration assessment.

Description

Technical Field

[0001] The present invention relates to the field of image cultural relic disease identification and assessment, and in particular to a cultural relic disease identification and assessment system and method based on image analysis. Background Art

[0002] In the field of cultural relics protection, stone cultural relics are susceptible to diseases such as biofilm coverage, cracks and weathering due to the influence of natural factors such as water, salt, biological erosion and weathering, as well as human factors;

[0003] Traditional manual visual inspection methods have significant limitations, such as insufficient accuracy in detecting early-stage diseases with weak spectral differences. Restoration effect evaluation relies on subjective expert experience and lacks quantitative indicators, making it difficult to scientifically compare the advantages and disadvantages of different solutions. Furthermore, they are unable to effectively integrate multi-dimensional features such as spectrum, texture, and morphology to analyze multiple types of superimposed diseases. Furthermore, the lack of data-driven optimization models makes decision-making on the trade-offs between restoration effect, cost, and time inefficient. This is especially true in the application of new technologies such as microbial remediation, where strain selection, mineralization solution formulation, and process parameter optimization require extensive trial and error.

[0004] Although existing image analysis technology can assist in detection, it faces technical bottlenecks such as difficulty in segmenting weak-contrast defects, insufficient global feature capture, a single restoration and evaluation system, and a lack of multi-objective optimization capabilities. Therefore, a cultural relic defect identification and evaluation system and method based on image analysis is needed to solve the above problems. Summary of the Invention

[0005] In order to solve the technical problems raised by the above background technology, the present invention provides a system and method for identifying and evaluating cultural relics defects based on image analysis.

[0006] The purpose of the present invention can be achieved through the following technical solutions:

[0007] A first aspect of the present invention provides a cultural relic disease identification and assessment system based on image analysis, comprising a multispectral image acquisition module, an image processing module, a disease identification module, a restoration assessment module, a decision optimization module and a disease database.

[0008] The multispectral image acquisition module uses the reflective area detection algorithm of the detection equipment to identify strong reflective points, optimize the ring light source parameters and automatically adjust the light source brightness to obtain multispectral images of cultural relics. The specific process is as follows:

[0009] The detection equipment collects pre-scan images of the cultural relics to be detected , grayscale the pre-scanned image, and then calculate the global mean of the pixels in the image and standard deviation ; Set the reflective area to the pixel value in the pre-scanned image> The corresponding area is used to segment the reflective pixel area ; If the reflective area is detected to be non-empty, adjust the brightness parameters of the ring light source , and its adjustment logic is: ,in is the initial value of the brightness parameter, is the attenuation coefficient, is the height of the pre-scanned image in pixels, is the width in pixels;

[0010] The brightness of the light source is automatically adjusted through the optimized brightness parameters, and the multispectral images of the cultural relics are collected and sent to the image processing module.

[0011] The image processing module eliminates the noise of the multispectral image through a multi-band collaborative filtering algorithm, then extracts features from the noise-eliminated image and fuses them into a disease enhancement feature set. It then sets an automatic threshold cutting and divides the image into diseased areas. The specific process is as follows:

[0012] Get Band Image Pixel , and its denoised value is , its calculation logic is: ,in Therefore Centered search window, is the normalization factor, is the pixel within the search window Pixel The weight of is calculated by the similarity of the two neighborhoods, and the formula is: ,in is an exponential function, is the number of neighborhood pixels; collaborative filtering constrains the spectral consistency of multi-band pixels, suppresses noise and retains spectral features to obtain a denoised multispectral image , its multi-band collaborative filtering calculation logic is: ,in is the mean of multi-band pixel values;

[0013] The spectral index, image texture features and morphological features of the denoised multispectral image are extracted and fused to generate a disease enhancement feature set. The spectral index is constructed by analyzing the spectral difference between diseased and normal stone, which includes the biofilm recognition index. , crack identification index Weathering index , the calculation logic of biofilm recognition index is: ,in The reflectivity value of the surface of the cultural relic under the red light band with a wavelength of 680 nanometers, Indicates the reflectivity value of the cultural relic surface under the blue light band with a wavelength of 480 nanometers; crack identification index The calculation logic is: ,in The reflectivity of the band, Logarithmic operation; Weathering index The calculation logic is: , as the spectral index feature set;

[0014] Extract image texture features and Introduce Gabor filter to get texture energy , the formula is: ,in and are the filter parameter scale and direction, is the convolution operation, is the Gabor kernel, is the phase difference, which is used as the texture feature set;

[0015] The morphological features are extracted through mathematical morphological operations, specifically: Top-Hat transformation is used to extract the small raised or depressed areas on the stone surface. , extract the crack contour through Canny edge detection and calculate the length of the contour , which is used as the morphological feature set; the spectral index feature set, image texture feature set and morphological feature set are spliced ​​into the disease enhancement feature set ;

[0016] The Otsu algorithm is used to automatically determine the segmentation threshold of the spectral index image and generate the diseased area. ,by The connected regions in the image are seeds. The disease boundary is expanded by the region growing algorithm, and the mean of the seed region is calculated. and standard deviation , traverse adjacent pixels, if the pixel value Then merge it into the diseased area and finally generate the disease mask .

[0017] The disease recognition module calculates the specific location of each independent disease based on the connected region labeling algorithm and location parameters, and then classifies the disease type through the neural symbolic model. The specific process is as follows:

[0018] Mark the independent diseased areas in the mask to obtain multiple connected areas, and then use the two-pass scanning method. The first pass is to traverse each pixel in the connected area, from left to right and from top to bottom. For the current pixel, if the foreground value is 1, check the labels of its neighboring pixels and divide them into 3 cases for processing respectively: Case 1: If the neighboring pixels are all background or unlabeled, assign a new label; Case 2: If the neighboring pixel has a label, inherit the label; Case 3: If the neighboring pixel has multiple different labels, select the smallest label and assign it to the current pixel; Use the union-find data structure to merge all equivalent pairs into a label mapping table. Specifically, the second pass traverses the image again, and maps the temporary label of each pixel to a unique real label according to the label mapping table, and finally obtains the marked connected area, and each marked connected area has a unique label;

[0019] Extract the location parameters of the disease in the marked connected area. The location parameters include the bounding box and the center of mass. The bounding box is the minimum matrix in the area, and the coordinates are , is the column index, is the row index; the centroid coordinates are the mean pixel coordinates of the region ; In this way, the bounding box and the centroid are overlapped and matched to obtain the location point of the corresponding disease;

[0020] The disease enhancement feature set is input into the neural symbolic model and calculated according to the disease probability formula ,in is the probability of each disease category, is the rule matching score, The final disease type, The execution action of integrating the two outputs is to map the disease enhancement feature set to the disease category probability. The calculation logic is: ,in, and The weight matrix and bias vector of the first fully connected layer, and The weight matrix and bias vector of the second fully connected layer, is the normalization function, is the activation function; define the symbol rule matching as ,in is the indicator function, 、 、 、 、 and is the initial value set based on expert experience; the disease type and corresponding disease location are sent to the disease database.

[0021] The restoration assessment module compares images before and after restoration and conducts quantitative analysis of mineral crystals, combining structural similarity indicators to output a comprehensive quantitative score. The specific process is as follows:

[0022] Extract the scoring indicators of the images before and after restoration and mineral crystallization, including image difference, structural similarity, crystal area change rate and crystal uniformity, including judging the pixel-level difference of the images before and after restoration, and its image difference The calculation logic is: ,in The pre-scanned image mentioned above is used as the image before restoration. is the restored image; it measures the degree of image structure preservation and its structural similarity The calculation logic is: ,in and is the image mean, and is the image variance, is the covariance, and Is the stability coefficient; it measures the degree of reduction of the crystallization area and the rate of change of the crystallization area The calculation logic is: ,in is the crystallization area before repair, The crystallization area after repair is used to measure the uniformity of the crystal size. The calculation logic is: ,in is the area of ​​each crystal particle after repair, and is the area standard deviation and area mean; the scoring index is normalized and then weighted and summed to calculate the comprehensive restoration score , its calculation logic is: ,in is the number of the scoring indicator, The total number of scoring indicators, is the indicator weight, is the normalized scoring indicator.

[0023] The decision optimization module integrates disease data and restoration scores, and uses a multi-objective optimization model to weigh restoration effects, costs, and time to output the optimal decision. The specific process is as follows:

[0024] With the goal of maximizing the repair effect, minimizing the cost and shortening the time, the optimal solution that meets the constraints is searched through a multi-objective optimization mathematical model; the maximization of the repair effect is set within the multi-objective optimization model. The objective function is based on the comprehensive repair score value , actual mineralization rate of the strain , mineralization efficiency of the formula The setting logic is: ,in is a fixed weight of 0.7, is the maximum mineralization rate of the strains in the microbial remediation knowledge layer, Maximize the efficiency of the recipe in the knowledge layer; minimize the setting cost The cost includes strain culture, mineralization solution, labor and equipment costs, and the calculation logic is: ,in is the strain culture cost, is the unit price of mineralization liquid, is the amount of mineralizing fluid, The labor unit price, is the total processing time, It is the labor depreciation cost; set the time to minimize , including strain culture time , Mineralization solution preparation time and processing time , and its overall objective function is: ;

[0025] Genetic algorithm is introduced into the multi-objective optimization model, and each repair solution Convert to triple individuals, select any combination from the microbial remediation knowledge layer, set the population to 80 individuals, and calculate the adaptation evaluation value of the three function objectives of each individual. The calculation logic is as follows: 、 and ,in and is the minimum / maximum value of the repair effect in the current population, and is the minimum / maximum cost in the current population, and is the minimum / maximum value of time in the current population; select the top 20% individuals in fitness, randomly exchange the strain ID or recipe ID of two individuals, and randomly replace the process ID of an individual with a probability of 5%; stop and output the optimal solution when the population fitness converges or reaches the preset maximum number of iterations;

[0026] The disease database is equipped with various preset thresholds and a microbial remediation knowledge layer. The microbial remediation knowledge layer stores structured data related to microbial remediation and supports rapid retrieval and rule matching. Its categories include strain characteristics, mineralization liquid formula and process parameters. The strain characteristics determine whether the strain can adapt to the cultural relic environment and the mineralization efficiency. The mineralization liquid formula affects the mineralization reaction rate, material cost and operational feasibility. The process parameters directly determine the repair time and operation complexity.

[0027] In a second aspect, the present invention provides a method for identifying and evaluating cultural relics damage based on image analysis, the specific steps of which are as follows:

[0028] Step 1: Multispectral image acquisition: The reflective area detection algorithm identifies strong reflective points, automatically adjusts the light source brightness and optimizes the ring light source angle to obtain multispectral images of the cultural relics, and then verifies the image validity using a quality assessment algorithm;

[0029] Step 2: Image processing: Use a multi-band collaborative filtering algorithm to eliminate noise from the multispectral image, extract features from the noise-eliminator image, and fuse them into a disease enhancement feature map. Then, set an automatic threshold to divide the image into diseased areas.

[0030] Step 3: Disease identification: Calculate the specific location of each independent disease using the connected component labeling algorithm and location parameters, and then classify the disease type using a neural symbolic model;

[0031] Step 4: Restoration Assessment: Through image comparison before and after restoration and quantitative analysis of mineral crystals, combined with structural similarity indicators, a quantitative score is output;

[0032] Step 5: Decision optimization: Based on the integrated disease data and restoration scores, a multi-objective optimization model is used to weigh the restoration effect, cost, and time to output the optimal solution.

[0033] Compared with the existing technology, the present invention has the following advantages: the disease recognition module uses a two-pass scanning method combined with a query data structure to accurately mark connected areas, obtain position parameters such as bounding boxes and centroids, and then uses a neural symbolic model to integrate the feature learning ability of neural networks with the logical reasoning ability of symbolic rules to achieve accurate positioning of disease locations and interpretable classification of types, effectively solving the problem of identifying multiple types of superimposed diseases;

[0034] The restoration assessment module outputs a scientific and objective comprehensive restoration score by normalizing and weighting the quantitative indicators such as the image difference before and after restoration, structural similarity, and crystal area change rate, avoiding the subjectivity of traditional manual evaluation and providing a traceable quantitative basis for the restoration effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. The following drawings are not intentionally scaled to the actual size, and the focus is on illustrating the main purpose of the present invention.

[0036] Figure 1 This is a principle block diagram of the present invention.

[0037] Figure 2 A diagram showing the steps of the method of the present invention. DETAILED DESCRIPTION

[0038] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts also fall within the scope of protection of the present invention.

[0039] Please refer to Figure 1 As shown, the first aspect of the present invention provides a cultural relic disease identification and evaluation system based on image analysis, including a multispectral image acquisition module, an image processing module, a disease identification module, a restoration evaluation module, a decision optimization module and a disease database.

[0040] The multispectral image acquisition module uses the reflective area detection algorithm of the detection equipment to identify strong reflective points, optimize the ring light source parameters and automatically adjust the light source brightness to obtain multispectral images of cultural relics. The specific process is as follows:

[0041] The detection equipment collects pre-scan images of the cultural relics to be detected , grayscale the pre-scanned image, and then calculate the global mean of the pixels in the image and standard deviation ; Set the reflective pixel area to the pixel value in the pre-scan image> The corresponding area is used to segment the reflective pixel area ; If the reflective area is detected to be non-empty, adjust the brightness parameters of the ring light source , it should be noted that the reflective pixel area This refers to areas of strong reflection on the surface of cultural relics due to improper light source illumination angles or the reflective properties of the material, such as smooth areas on stone surfaces or metal parts. In the pre-scanned image, these areas show significantly higher pixel values ​​than the surrounding areas. The ring light source is an auxiliary light source in the detection equipment. It consists of multiple LED lamp beads evenly arranged around the detection equipment lens to provide surround lighting, ensuring consistent brightness in all areas of the multispectral image. Its adjustment logic is as follows: ,in is the initial value of the brightness parameter, is the attenuation coefficient, which ranges from 0.1 to 0.3. The larger the reflective area, the greater the brightness attenuation. is the height of the pre-scanned image in pixels, is the width in pixels;

[0042] It should be noted that the specific value of k is determined by technical personnel in this field through routine experiments based on the actual application scenario and the expected brightness attenuation characteristics; the lower limit k=0.1: is suitable for scenarios with a relatively small reflective area. At this time, the interference of reflection on the characteristics of the disease is relatively weak, and a small attenuation of brightness can avoid excessive reduction of the overall brightness of the image, thereby ensuring the visibility of weak-contrast diseases; the upper limit k=0.3: is suitable for scenarios with a relatively large reflective area; at this time, the reflective area may cover a large number of potential disease areas, and the reflection needs to be suppressed through a large-scale brightness attenuation, but excessive attenuation should be avoided to cause loss of dark details in the image, such as textures deep in the cracks. The brightness of the light source is automatically adjusted through the optimized brightness parameters, and the multispectral image of the cultural relics is collected and sent to the image processing module.

[0043] The image processing module eliminates the noise of the multispectral image through a multi-band collaborative filtering algorithm, extracts features from the noise-eliminated image, and fuses them into a disease enhancement feature set. It then sets an automatic threshold cutting to divide the image into diseased areas. The specific process is as follows:

[0044] Multispectral images contain raw light intensity data of multiple discrete spectral bands, which are usually expressed as multiple independent single-band images or unseparated composite data. In order to improve the accuracy of feature extraction and positioning of image disease areas, a multi-band collaborative filtering algorithm is introduced to eliminate noise during image acquisition. The multi-band collaborative filtering process is divided into two steps: single-band denoising and collaborative filtering. The goal of single-band denoising is to suppress random noise in the band, such as sensor thermal noise and readout noise, while retaining the texture details of the band, such as crack edges and biofilm boundaries; Acquire the band Image Pixel , and its denoised value is , its calculation logic is: ,in Therefore Centered search window, is the normalization factor, is the pixel within the search window Pixel The weight of Indicates that the band is In a single-band image, the coordinates are pixels The original grayscale value of ; its weight is calculated by the similarity of the two neighborhoods, the formula is: ,in is an exponential function, is the number of neighborhood pixels, h is a smoothing parameter that controls the decay rate of the weight; after single-band denoising, pixels at the same spatial position in different bands may have spectral inconsistencies due to residual noise. For example, the reflectance of normal stone at 550nm and 650nm should be positively correlated, but after denoising, a negative correlation may appear. Collaborative filtering further suppresses noise and retains spectral features by constraining the spectral consistency of multi-band pixels, thus obtaining a denoised multispectral image. , its multi-band collaborative filtering calculation logic is: ,in is the mean of multi-band pixel values;

[0045] The spectral index, image texture features and morphological features of the denoised multispectral image are extracted and fused to generate a disease enhancement feature set. The spectral index is constructed by analyzing the spectral difference between diseased and normal stone, which includes the biofilm recognition index. , crack identification index Weathering index Biofilms such as cyanobacteria or lichens contain chlorophyll and carotenoids, and their reflectivity at the blue light absorption peak of 480nm and the red light absorption peak of 680nm is lower than that of normal stone. The calculation logic of the biofilm identification index is: ,in The reflectivity value of the surface of the cultural relic under the red light band with a wavelength of 680 nanometers, It represents the reflectivity value of the cultural relic surface under the blue light band with a wavelength of 480 nanometers. It should be noted that the biofilm identification index is essentially amplifying the spectral difference between biofilm and normal stone by calculating the ratio of the reflectivity difference and the sum of two characteristic bands. The cracks are often filled with soil or water, and the reflectivity at 900nm is significantly lower than that of the surrounding stone. The calculation logic of the crack identification index is: ,in The reflectivity of the band, It is a logarithmic operation; weathering will cause the minerals on the stone surface to decompose, and the reflectivity at 2300nm will decrease. The calculation logic is: , as the spectral index feature set;

[0046] Image texture features reflect the spatial distribution pattern of diseases, and the visible light band image Introduce Gabor filter to get texture energy , which is suitable for the multi-scale disease characteristics of stone cultural relics, and the formula is expressed as: ,in and are the filter parameter scale and direction, is the convolution operation, is the Gabor kernel, is the phase difference, which is used as the texture feature set;

[0047] Morphological characteristics describe the geometric properties of the disease, such as the length of the crack and the area of ​​the biofilm, which are extracted through mathematical morphological operations, specifically: Top-Hat transformation to extract the area of ​​small protrusions or depressions on the stone surface , extract the crack contour through Canny edge detection and calculate the length of the contour , which is used as the morphological feature set; the spectral index feature set, image texture feature set and morphological feature set are spliced ​​into the disease enhancement feature set ;

[0048] The Otsu algorithm is used to automatically determine the segmentation threshold of the spectral index image and generate the diseased area. , based on the diseased area The connected regions in the image are seeds. The disease boundary is expanded by the region growing algorithm, and the mean of the seed region is calculated. and standard deviation , traverse adjacent pixels, if the pixel value Then merge it into the diseased area and finally generate the disease mask , 0 represents the background area, and 1 represents the diseased area.

[0049] The disease recognition module calculates the specific location of each independent disease based on the connected region labeling algorithm and location parameters, and then classifies the disease type through the neural symbolic model. The specific process is as follows:

[0050] The binary mask generated by the Otsu algorithm and morphological operations only marks the boundary between the disease and the background. The disease location needs to be further obtained through the connected component labeling algorithm; mark the independent disease areas in the mask to obtain multiple connected components, and then use the two-pass scanning method. The first pass is to traverse each pixel in the connected component, from left to right and from top to bottom. For the current pixel, if the foreground value is 1, check the labels of its neighboring pixels and divide them into three cases for processing respectively: Case 1: If all neighboring pixels are background or unlabeled, a new label is assigned; Case 2: If the neighboring pixel has a label, it inherits the label; Case 3: If the neighboring pixel has multiple different labels, such as label A and label B → record (A, B) as an equivalent pair, and select the smallest label to assign to the current pixel. It should be noted that the purpose of the first pass is to assign a temporary label to each pixel and record the equivalent pairs; use the union-find data structure to merge all equivalent pairs into a label mapping table. The second pass traverses the image again, and maps the temporary label of each pixel to a unique real label according to the label mapping table, and finally obtains a labeled connected component, each labeled connected component has a unique label;

[0051] Extract the location parameters of the disease in the marked connected area. The location parameters include the bounding box and the center of mass. The bounding box is the minimum matrix in the area, and the coordinates are , is the column index, is the row index; the centroid coordinates are the mean pixel coordinates of the region , reflecting the geometric center of the disease; thereby, the bounding box and the centroid are overlapped and matched to obtain the location point of the corresponding disease;

[0052] The disease enhancement feature set is input into the neural symbolic model and calculated according to the disease formula ,in is the probability of each disease category, is the rule matching score, The final disease type, The execution action is to integrate the outputs of the two. Specifically, the disease enhancement feature set is mapped to the disease category probability. The disease category probability calculation logic is: ,in, and The weight matrix and bias vector of the first fully connected layer, and The weight matrix and bias vector of the second fully connected layer, is the normalization function, is the activation function; define the symbol rule matching as ,in is the indicator function, 、 、 、 、 and is the initial value set based on expert experience; the disease type and corresponding disease location are sent to the disease database.

[0053] It should be noted that the rule matching score is the degree of consistency between the symbolic rules constructed based on domain knowledge and the input features, reflecting the constraints of expert experience on disease characteristics; the disease diagnosis rule base is formulated by agricultural experts, and each rule contains premise conditions (such as symptom combination) and conclusions (disease type), and the uncertainty of the rules is described by parameters. The probability distribution output by the neural symbolic model is converted into a possibility distribution as the input of the symbolic rule, and the possibility of the rule premise is calculated through max-min matrix operations. The rule results of the disease diagnosis rule base are matched, and all rule matching results are standardized to obtain a rule matching score in the range of 0-1; the category probability is the probability prediction of the neural network model for the input features, and the probability distribution of each disease category is output through the softmax activation function. Specifically, through neural network training, the disease enhancement feature set is used as input, and the parameters are optimized through back propagation, so that the model learns the mapping relationship between features and disease categories; the final disease type is the decision result after integrating the rule matching score and the category probability, which is a double judgment result. For example, when the category probability of downy mildew is 0.8 and the corresponding rule matching score is 0.9, downy mildew is determined to be the final disease type. The final disease types include biofilm, cracks and weathering; the execution action is the specific operation instruction generated by the model according to the final disease type, and a mapping table from disease type to action is specifically established. For example, if the disease type is biofilm, the repair target is to remove biofilms such as cyanobacteria and lichens to inhibit regeneration. The actions are spraying microbial inhibitors, local cleaning with weak acid solution and ultraviolet light inactivation treatment;

[0054] The restoration assessment module compares images before and after restoration and conducts quantitative analysis of mineral crystals, combining structural similarity indicators to output a comprehensive quantitative score. The specific process is as follows:

[0055] Extract the scoring indicators of the images before and after restoration and mineral crystallization, including image difference, structural similarity, crystal area change rate and crystal uniformity, including judging the pixel-level difference of the images before and after restoration, and its image difference The calculation logic is: ,in The pre-scanned image mentioned above is used as the image before restoration. is the restored image, The larger the value, the more obvious the visual difference; it measures the degree of image structure preservation. The closer the structural similarity is to 1, the more similar the structures are. The calculation logic is: ,in and is the image mean, and is the image variance, is the covariance, and Is the stability coefficient; it measures the degree of reduction of the crystallization area and the rate of change of the crystallization area The calculation logic is: ,in is the crystallization area before repair, The crystal area after repair is used to measure the uniformity of the crystal size. The more uniform the structure is, the more stable it is after repair. The calculation logic is: ,in is the area of ​​each crystal particle after repair, and are the area standard deviation and area mean, The closer it is to 1, the more uniform the crystal size is; the comprehensive repair score is calculated by normalizing the scoring index and then weighting and summing it. , its calculation logic is: ,in is the number of the scoring indicator, The total number of scoring indicators, is the indicator weight, is the normalized scoring index;

[0056] It should be noted that The indicator weights are determined by the hierarchical analysis method. A hierarchical structure model is specifically established to clarify the hierarchical relationship between the target layer and the criterion layer. The target layer is used to determine the weights of the restoration evaluation indicators; the criterion layer is the evaluation indicators of image difference, structural similarity, crystal area change rate and crystal uniformity. The relative importance of the indicators is quantified by the 1-9 scaling method, and a 4×4 judgment matrix AH is constructed. The matrix elements represent the importance of any indicator relative to other indicators. If the meaning is equally important, the influence of crystal uniformity and image difference on the restoration effect is equivalent. If it is obviously important, the crystal area change rate is obviously more important than structural similarity. Then, the geometric mean of each row of the judgment matrix is ​​calculated and the geometric mean is normalized to obtain the weight of each indicator.

[0057] The normalized scoring indicators are expressed as the normalized processing of image difference, structural similarity, crystal area change rate and crystal uniformity. By unifying the dimension [0,1] and direction, the comparability of different indicators is achieved.

[0058] The decision optimization module integrates disease data and restoration scores, and uses a multi-objective optimization model to weigh restoration effects, costs, and time, and outputs the optimal solution that takes multiple objectives into account. The specific process is as follows:

[0059] With the goal of maximizing the repair effect, minimizing the cost and shortening the time, the optimal solution that meets the constraints is searched through a multi-objective optimization mathematical model; the maximization of the repair effect is set within the multi-objective optimization mathematical model. The objective function is based on the comprehensive repair score value , actual mineralization rate of the strain , mineralization efficiency of the formula The setting logic is: ,in is a fixed weight with a value of 0.7. is the maximum mineralization rate of the strains in the microbial remediation knowledge layer, Maximize the efficiency of the recipe in the knowledge layer; minimize the setting cost The cost includes strain culture, mineralization solution, labor and equipment costs, and the calculation logic is: ,in is the strain culture cost, is the unit price of mineralization liquid, is the amount of mineralizing fluid, The labor unit price, is the total processing time, It is the labor depreciation cost; set the time to minimize , including strain culture time , Mineralization solution preparation time and processing time , its calculation logic is: , and its overall objective function is: ;

[0060] It should be noted that a large number of actual case data on microbial remediation in the past were collected, sorted and deeply analyzed; a detailed database was established to record the strains, formulas and corresponding mineralization rate and efficiency data used in each case; through statistical analysis of these historical data, data mining and statistical analysis methods were used, such as calculating statistical quantities such as average, maximum and minimum values, to find the distribution of mineralization rates and formula efficiencies of strains under different conditions, and then determine the maximum mineralization rate and maximum formula efficiency. For example, data on all cases of microbial remediation of stone cultural relics in the past were analyzed to screen out the highest mineralization rates and efficiencies under conditions such as different types of cultural relics and degree of disease; at the same time, machine learning algorithms were used to model historical data to predict the mineralization rates and efficiencies that different strains and formulas may achieve under specific conditions, providing a more accurate reference for determining the maximum mineralization rate and efficiency;

[0061] The comprehensive restoration score is obtained by weighted summation of the scoring indicators image difference, structural similarity, crystal area change rate, and crystal uniformity. The actual mineralization rate of the strain is the core indicator for measuring the ability of microorganisms (such as urease-producing bacteria) to induce mineralization. It refers to the mass of mineralized products generated per unit time under the restoration environment. In the laboratory, the on-site environment of cultural relics restoration is simulated, and the strain is inoculated into a culture medium containing mineralized substrates. Samples are taken regularly, and the amount of mineralized products is calculated by weight or spectrophotometry. The specific calculation logic is as follows: ,in For cultivation time The quality of the mineralized products generated within is the volume of the reaction system;

[0062] The mineralization efficiency of a formula is a comprehensive indicator that measures the effectiveness of the mineralization solution in promoting the mineralization reaction. Combining the relationship between mineralization effect, cost, and time, the calculation logic is as follows: ,in is the total mass of mineralized products generated after using this formula, is the volume of mineralization fluid used, is the unit volume cost of mineralized fluid;

[0063] Genetic algorithm is introduced into the multi-objective optimization mathematical model. Convert to triple individuals, select any combination from the microbial remediation knowledge layer, set the population to 80 individuals, and calculate the adaptation evaluation value of the three function objectives of each individual. The calculation logic is as follows: 、 and ,in and is the minimum / maximum value of the repair effect in the current population, and is the minimum / maximum cost in the current population, and is the minimum / maximum value of time in the current population; select the top 20% individuals in fitness, randomly exchange the strain ID or recipe ID of two individuals, and randomly replace the process ID of an individual with a probability of 5%; stop and output the optimal solution when the population fitness converges or reaches the preset maximum number of iterations;

[0064] The disease database is equipped with various preset thresholds and a microbial remediation knowledge layer. The microbial remediation knowledge layer stores structured data related to microbial remediation and supports rapid retrieval and rule matching. Its categories include strain characteristics, mineralization liquid formula and process parameters. The strain characteristics determine whether the strain can adapt to the cultural relic environment and the mineralization efficiency. The mineralization liquid formula affects the mineralization reaction rate, material cost and operational feasibility. The process parameters directly determine the repair time and operation complexity.

[0065] Please refer to Figure 2 As shown, the second aspect of the present invention provides a method for identifying and evaluating cultural relics damage based on image analysis, and the specific steps are as follows:

[0066] Step 1: Multispectral image acquisition: Use the reflective area detection algorithm to identify strong reflective points, automatically adjust the light source brightness and optimize the annular light source angle to obtain multispectral images of cultural relics;

[0067] Step 2: Image processing: Use a multi-band collaborative filtering algorithm to eliminate noise from the multispectral image, extract features from the noise-eliminator image, and fuse them into a disease enhancement feature map. Then, set an automatic threshold to divide the image into diseased areas.

[0068] Step 3: Disease identification: Calculate the specific location of each independent disease using the connected component labeling algorithm and location parameters, and then classify the disease type using a neural symbolic model;

[0069] Step 4: Restoration Assessment: Through image comparison before and after restoration and quantitative analysis of mineral crystals, combined with structural similarity indicators, a quantitative score is output;

[0070] Step 5: Decision optimization: Based on the integrated disease data and restoration scores, a multi-objective optimization model is used to weigh the restoration effect, cost, and time to output the optimal solution.

[0071] The above is an illustration of the present invention and should not be considered as limiting thereof. Although several exemplary embodiments of the present invention have been described, it will be readily understood by those skilled in the art that many modifications may be made to the exemplary embodiments without departing from the novel teachings and advantages of the present invention. Therefore, all such modifications are intended to be included within the scope of the present invention as defined by the claims. It should be understood that the above is an illustration of the present invention and should not be considered as being limited to the specific embodiments disclosed, and modifications to the disclosed embodiments and other embodiments are intended to be included within the scope of the appended claims. The present invention is defined by the claims and their equivalents.

Claims

1. A cultural relic damage identification and assessment system based on image analysis, comprising an image processing module, a damage identification module, a restoration assessment module, a decision optimization module, and a damage database, characterized by: The image processing module eliminates noise from multispectral images through a multi-band collaborative filtering algorithm, extracts features from the noise-eliminated image, and fuses them into a disease enhancement feature map. It then sets an automatic threshold cutting to divide the image into diseased areas. The restoration assessment module compares images before and after restoration and conducts quantitative analysis of mineral crystals, combining structural similarity indicators to output a quantitative score. The decision optimization module integrates disease data and restoration scores, and uses a multi-objective optimization model to weigh restoration effects, costs, and time to output the optimal solution. The disease recognition module calculates the specific location of each independent disease according to the connected area marking algorithm and position parameters. Specifically, the independent disease area in the mask is marked to obtain multiple connected areas. Then, a two-pass scanning method is used. The first pass is to traverse each pixel in the connected area from left to right and from top to bottom. For the current pixel, if the foreground value is 1, the labels of its neighboring pixels are checked and divided into three cases for processing respectively: Case 1: If the neighboring pixels are all background or unlabeled, a new label is assigned; Case 2: If the neighboring pixel has a label, the label is inherited; Case 3: If the neighboring pixel has multiple different labels, the smallest label is selected and assigned to the current pixel; Use the union-find data structure to merge all equivalent pairs into a label mapping table. Specifically, the second pass traverses the image again, and maps the temporary label of each pixel to a unique real label according to the label mapping table, and finally obtains a marked connected area, and each marked connected area has a unique label. Extract the location parameters of the disease in the marked connected area. The location parameters include the bounding box and the center of mass. The bounding box is the minimum matrix in the area, and the coordinates include column index and row index. The centroid coordinates are the mean pixel coordinates of the region; the bounding box and the centroid are matched to obtain the location point of the corresponding disease.

2. The cultural relics disease identification and assessment system based on image analysis according to claim 1, characterized in that: The disease recognition module classifies disease types through a neural symbolic model, specifically: The disease enhancement feature set is input into the neural symbolic model, and the final disease type is calculated according to the disease probability formula; the disease enhancement feature set is mapped to the disease category probability, and the symbol rule matching is defined; the disease type and the corresponding disease location are sent to the disease database.

3. The cultural relics disease identification and assessment system based on image analysis according to claim 1, characterized in that: The image processing module eliminates the noise of the multispectral image through a multi-band collaborative filtering algorithm. The specific process is as follows: The image pixels of the band are obtained, their denoised values ​​are calculated, and collaborative filtering is used to constrain the spectral consistency of multi-band pixels, suppress noise and retain spectral characteristics to obtain a denoised multispectral image.

4. The cultural relics disease identification and assessment system based on image analysis according to claim 3 is characterized in that: The image processing module extracts the spectral index, image texture features and morphological features of the denoised multispectral image, and generates a disease enhancement feature set after fusion, specifically: The spectral index is constructed by analyzing the spectral differences between diseased and normal stones. It includes the biofilm identification index, crack identification index and weathering degree index as the spectral index feature set. Extract image texture features, introduce Gabor filter into visible light band image, obtain texture energy, and use it as texture feature set; Morphological features are extracted through mathematical morphological operations, specifically: Top-Hat transformation is used to extract the tiny raised or depressed areas on the stone surface, crack contours are extracted through Canny edge detection, and the length of the contours is calculated as the morphological feature set; the spectral index feature set, image texture feature set and morphological feature set are spliced ​​into a disease enhancement feature set.

5. The cultural relics disease identification and assessment system based on image analysis according to claim 4 is characterized in that: The image processing module uses the Otsu algorithm to automatically determine the segmentation threshold for the spectral index image, generates the diseased area, uses the connected area in the diseased area as the seed, expands the diseased boundary through the region growing algorithm, and calculates the mean of the seed area. and standard deviation , traverse adjacent pixels, if the pixel value , then it is merged into the diseased area and finally a disease mask is generated.

6. The cultural relics disease identification and assessment system based on image analysis according to claim 1, characterized in that: The restoration assessment module uses image comparison before and after restoration and quantitative analysis of mineral crystals, combined with structural similarity indicators to output a comprehensive quantitative score. The specific process is as follows: Extract scoring indicators for images and mineral crystals before and after restoration. The scoring indicators include image difference, structural similarity, crystal area change rate, and crystal uniformity. This includes determining the pixel-level difference between images before and after restoration, measuring the degree of image structure preservation, calculating the crystal area change rate, and measuring the uniformity of crystal size. The scoring indicators are normalized and then weighted summed to calculate the comprehensive restoration score.

7. The cultural relics disease identification and assessment system based on image analysis according to claim 1, characterized in that: The decision optimization module takes maximizing the repair effect, minimizing the cost and shortening the time as the objective function, and its specific process is as follows: The objective function of maximizing the remediation effect is set in the multi-objective optimization model, which is based on the comprehensive remediation score, the actual mineralization rate of the strain, and the mineralization efficiency of the formula. The cost is minimized, and the cost includes strain culture, mineralization solution, labor and equipment costs, and the total objective function is calculated.

8. The cultural relics disease identification and assessment system based on image analysis according to claim 7, characterized in that: The decision optimization module introduces a genetic algorithm into the multi-objective optimization model to search for the optimal solution that meets the constraints, specifically: Each remediation scheme is converted into a triplet of individuals, and any combination is selected from the microbial remediation knowledge layer. The population is set to 80 individuals, and the fitness evaluation value of the three function objectives of each individual is calculated; The top 20% individuals in fitness are selected, the strain ID or recipe ID of two individuals are randomly exchanged, and the process ID of an individual is randomly replaced with a probability of 5%; when the fitness of the population converges or reaches the preset maximum number of iterations, the algorithm stops and outputs the optimal solution.

9. The cultural relics disease identification and assessment system based on image analysis according to claim 1, characterized in that: The multispectral image acquisition module is also included. The multispectral image acquisition module uses the reflective area detection algorithm of the detection equipment to identify strong reflective points, optimize the parameters of the ring light source and automatically adjust the brightness of the light source to obtain multispectral images of the cultural relics. The specific process is as follows: The detection equipment collects the pre-scanned image of the cultural relic to be detected, grayscales the pre-scanned image, and then calculates the global mean of the pixels in the image. and standard deviation ; Set the reflective area to the pixel value in the pre-scanned image> The corresponding area is segmented to separate the reflective pixel area; If it is detected that the reflective area is not empty, the brightness parameters of the ring light source are adjusted; the brightness of the light source is automatically adjusted according to the optimized brightness parameters, and the multispectral image of the cultural relic is collected and sent to the image processing module.

10. A method for identifying and evaluating cultural relics defects based on image analysis, used to implement a system for identifying and evaluating cultural relics defects based on image analysis as claimed in any one of claims 1 to 9, characterized in that: include: Step 1: Multispectral image acquisition: Use the reflective area detection algorithm to identify strong reflective points, automatically adjust the light source brightness and optimize the annular light source angle to obtain multispectral images of cultural relics; Step 2: Image processing: Use a multi-band collaborative filtering algorithm to eliminate noise from the multispectral image, extract features from the noise-eliminator image, and fuse them into a disease enhancement feature map. Then, set an automatic threshold to divide the image into diseased areas. Step 3: Disease identification: Calculate the specific location of each independent disease using the connected component labeling algorithm and location parameters, and then classify the disease type using a neural symbolic model; Step 4: Restoration Assessment: Through image comparison before and after restoration and quantitative analysis of mineral crystals, combined with structural similarity indicators, a quantitative score is output; Step 5: Decision optimization: Based on the integrated disease data and restoration scores, a multi-objective optimization model is used to weigh the restoration effect, cost, and time to output the optimal solution.

Images