CN104677853B - A method for assessing mural nailing disease based on near-infrared hyperspectral - Google Patents

Abstract

Onychonosus evil appraisal procedure is played the invention discloses a kind of mural painting based on near-infrared EO-1 hyperion, the mural painting plays onychonosus evil appraisal procedure and comprises the following steps：The image that particular wall is drawn under hundreds of continuous different spectral wavelengths is obtained by near-infrared Hyperspectral imager；Positioning is observed by the naked eye, the key position in the different orders of severity onychonosus evil in mural painting is searched out, and corresponding position is marked in corresponding high spectrum image, extracts spectroscopic data as the disease spectrum java standard library for playing first degree；High-spectral data is stored and is pre-processed；Spectrum characteristic data is extracted and analysis.The present invention realizes that the distribution that onychonosus evil is played to mural painting shows with the visualization of the order of severity by the acquisition and Spectra feature extraction and analysis to mural painting near-infrared high spectrum image.Operating process is simple, and the data obtained reliability is high.It is simultaneously contactless imaging, does not damage wall painting surface.

Description

A method for assessing mural nailing disease based on near-infrared hyperspectral

technical field

The invention relates to the field of computer-aided protection of cultural relics, in particular to a method for evaluating nail-on disease of murals based on near-infrared hyperspectral.

Background technique

Ancient murals are the most distinctive and important part of my country's cultural heritage. Ancient murals are preserved in most of the existing ruins in our country. It is not only a precious material for exploring the characteristics of ancient Chinese aesthetic decoration, but also an important record of the historical process of the development of human civilization, with extremely high historical, artistic and scientific value. These cultural heritages all involve protection technical issues in this industry field. Under the current situation that my country's protection technical personnel are relatively scarce and the level of protection technology is low, the protection task is very urgent.

The optical survey methods mainly used by people in the early days include X-ray photography, normal light photography, oblique light photography, infrared imaging and other means to investigate and record murals. In recent years, with the development of digital imaging technology and devices, multi-spectral imaging, ultraviolet fluorescence The application of imaging technology has entered the stage of practical application. However, the spectral range and dimension of these optical investigation methods are relatively small, and the characteristic changes of the pigment components on the surface of murals in the continuous spectral range are not well excavated. At the same time, computer technology has also begun to be used in the restoration of ancient paintings, including murals, oil paintings, etc., and computer-aided cultural relic cracks, peeling repairs and cultural relic cleaning have appeared. With the development of information technology, technologies such as digital photography, image processing, and machine learning have also been widely applied to the protection of ancient murals. In the process of using computer-aided mural protection, image processing technology is the most widely used and the most successful. However, the use of these information technologies mostly focuses on the research and analysis of the texture and color changes of mural images in the visible light band. The extracted features may not be able to represent the attributes of mural disease itself, and there is a lack of analysis of the changes in mural materials and pigments.

Contents of the invention

The present invention provides a method for assessing mural toenail disease based on near-infrared hyperspectral. The present invention realizes the distribution and severity of mural toenail disease by acquiring near-infrared hyperspectral images of murals and extracting and analyzing spectral features. Visual display, see the description below for details:

A method for evaluating nail-on disease in murals based on near-infrared hyperspectral, the method for evaluating nail-on disease in murals comprises the following steps:

Obtain images of specific murals at hundreds of consecutive different spectral wavelengths through a near-infrared hyperspectral imaging system;

Through visual observation and positioning, find the key positions of onychopathia in murals with different severity levels, mark the corresponding positions in the corresponding hyperspectral images, and extract the spectral data as the standard database of onychopathic disease spectrum;

Store and preprocess hyperspectral data; extract and analyze spectral feature data.

The near-infrared hyperspectral imaging system includes: a near-infrared hyperspectral camera, a rotating platform, a computer, a light source, a tripod and a data line,

The near-infrared hyperspectral camera is installed on the rotating platform, the near-infrared hyperspectral camera and the rotating platform are connected to the computer through the data line, and the rotating platform is controlled by the computer for a certain amount of time. Automatic rotation within an angle, while the near-infrared hyperspectral camera collects and stores images through the computer; the near-infrared hyperspectral camera and the rotating platform are fixed in position through the tripod, and the light source passes through the The tripod is fixed in position.

The steps of extracting and analyzing spectral characteristic data are specifically as follows:

Using the prediction model obtained from the training to predict the occurrence degree of nailing on the hyperspectral image of the mural pixel by pixel, and use different colors to mark the distribution of nailing disease in the mural to obtain a visual map of the distribution of nailing disease; at the same time, according to the weighted coefficient of determination in the prediction model For the maximum value and the minimum value, select the band that plays an important role in the prediction of mural nailing disease and input it into the prediction model for prediction.

Compared with the prior art, the present invention has the following beneficial effects:

1. The near-infrared hyperspectral imaging technology was used in the detection and analysis of mural toenail disease, and the spectral change characteristics of murals in the near-infrared band range were obtained, and the toenail disease spectrum was established with the different severity of toenail disease as a reference Standard library of traits. The operation process is simple, and the obtained data is highly reliable. At the same time, it is non-contact imaging, without damaging the surface of the mural.

2. Using the PLSR prediction model, the location and severity distribution map of mural nailing disease was obtained. Through this distribution map, users can intuitively see the location of the onycholysis disease in the mural, and according to the color change, they can find different severity of onychomycosis, so as to take corresponding protective measures. At the same time, it recorded the spectral change law of the material and state of the mural's nailing disease over time, which provided reference data for the protection and restoration of the mural in the later stage.

3. The extraction of important bands reduces the calculation cost and overhead of the system and simplifies the prediction model on the premise of ensuring the prediction accuracy and stability of the prediction model.

Description of drawings

Fig. 1 is the overall flowchart of the present invention;

Figure 2 is an example diagram of a hyperspectral image;

Fig. 3 is a schematic diagram of hyperspectral image storage format;

4 is a schematic diagram of a hyperspectral image acquisition device;

Figure 5 is a schematic diagram of different degrees of armor;

Fig. 6 is the spectrum graph of four kinds of different forms of raising nails;

Fig. 7 is a schematic diagram of a mural painting risk estimation map and its filtered image.

detailed description

In order to make the purpose, technical solution and advantages of the present invention clearer, the implementation manners of the present invention will be further described in detail below.

As a non-destructive optical investigation method, near-infrared hyperspectral technology has an important application prospect in the investigation and diagnosis of murals, especially when the image can quickly and intuitively distinguish a variety of information about the restoration and protection process of murals. It is of great significance for the analysis and identification of restoration materials in the scientific protection of murals and the evaluation of the preservation status of murals.

There are two problems in the mural disease early warning and evaluation technology: one is the imaging and feature extraction of murals; the other is the feature analysis and visualization of mural disease changes over time. The existing mural painting onycholysis analysis technology has shortcomings such as narrow spectral range and small spectral dimension in mural imaging, and it is difficult to analyze the change of mural painting from the chemical aspect. The present invention proposes a model that uses near-infrared hyperspectral technology to analyze and evaluate the onycholysis of murals, and analyzes and predicts the degree of onychomycosis at each pixel position in the imaging mural. The overall process is shown in Figure 1 .

101: Information collection of hyperspectral murals and establishment of a spectral standard database for nail-leaching diseases;

(1) Near-infrared hyperspectral imaging system

The scanning method of the imaging spectrometer is line scanning, and the light is split so that each element component corresponds to a pixel on the line array, and multiple lines are collected continuously to form a hyperspectral image. Therefore, each image structure from a spectral camera consists of linear pixels in one dimension (spatial axis) and the spectral distribution (the intensity of light at spectral elements) in another dimension (spectral axis), as shown in Figure 2 .

A hyperspectral information collection system specially for cultural relics is built by a three-dimensional displacement platform, a near-infrared hyperspectral imaging system and other auxiliary units (light source, control software, etc.), and its spectral range covers near-infrared (900-1700nm) and other bands. The spectral resolution is around 5nm. Use this system to acquire hyperspectral images of specific murals, as shown in Figure 4, 1 represents the near-infrared hyperspectral camera, 2 represents the rotating platform, 3 represents the computer, 4 represents the light source, 5 represents the mural samples taken, 6 represents Tripod, 7 represents the data line. Wherein the near-infrared hyperspectral camera 1 is installed on the rotating platform 2, the near-infrared hyperspectral camera 1 and the rotating platform 2 are connected to the computer 3 through the data line 7, and the rotating platform 2 can automatically rotate within a certain angle through the control of the computer 3, At the same time, the near-infrared hyperspectral camera 1 can collect and store images through the computer 3; the near-infrared hyperspectral camera 1 and the rotating platform 2 are fixed by the tripod 6, and the position of the light source 4 is fixed by the tripod 6, which can enhance the near-infrared in the environment. band strength.

(2) Establishment of disease spectrum standard library

Through the near-infrared hyperspectral imaging system, it is possible to obtain images of a specific mural at hundreds of continuous different spectral wavelengths, that is, its hyperspectral image. The storage form of the hyperspectral image is shown in Figure 3. The storage form of the hyperspectral image is Three-dimensional matrix, the image in the figure contains B bands in total. The x-axis direction represents the column of the image, the y-axis direction represents the row of the image, and the z-axis direction represents the band of the image. Each pixel on the image can be regarded as a B-dimensional vector. The image information is rich, the recognition degree is high, and there are many data description models. Since the reflectance spectrum of an object has a "fingerprint" effect, different objects have different spectra, and the same object must have the same spectrum to distinguish different material information, providing a technical basis for further risk assessment, protection, and restoration of cultural relics.

Cracking disease refers to the phenomenon that the base color layer or pigment layer of murals cracks, and then curls up in scales. The murals show different forms in different stages of nailing, and there may be various degrees of nailing in the same mural, as shown in Figure 5. Through visual observation and positioning, it is possible to find nails with different degrees of severity in the murals. The key position is marked in the corresponding hyperspectral image, and its spectral data is extracted as the disease spectrum standard library of the degree of infestation, as shown in Figure 6. According to the different forms of nail-leaching disease on the murals, the nail-leaching disease is divided into four different degrees. Among them, no nail-leaching occurs at all as state 1, which is represented by 0; slight nail-leafing occurs as state 2, and 0.3 is represented; Severe armor-raising is state 3, represented by 0.6; state 4 occurs when armor-leafing occurs and falls off, and 1 is represented.

102: Storage and preprocessing of hyperspectral data;

Due to the influence of the imaging system itself and the external environment, there is a certain amount of noise in the image. Firstly, the spectral data is corrected and registered, so as to obtain the correct spectral information and geometric information of the relevant cultural relics. Then the minimum noise separation method is used to concentrate the energy of the hyperspectral image to achieve the purpose of removing the noise in the band.

103: Spectral feature data extraction and analysis.

(1) Prediction model verification

Carry out the establishment of prediction model by the spectral feature that shows in the different degrees of severity in the onychopathic characteristic standard library of onychopathic disease, what adopt among the present invention is partial least squares method (partial leastsquares regressions, PLSR) to carry out The fitting accuracy of the model is evaluated by the error sum of squares (PRESS), the mean error (RMSE), and the coefficient of determination (R-squared). The coefficient of determination is used to explain the fitting accuracy of the prediction model. Goodness of fit value, the closer R-squared is to 1, the better the fitting degree, the average error and the sum of squared errors measure the deviation between the predicted value and the true value, the smaller the PRESS and RMSE, the higher the fitting accuracy . Adjust the parameters in the PLSR algorithm so that the R-squared of the prediction result is the largest, and the PRESS and RMSE are the smallest, so that the PLSR model that is most suitable for the prediction of mural toenail disease is obtained. The relevant formulas are defined as:

PRESS＝Σ(y _pred -y _act ) ²

Among them, y _pred is the predicted value obtained through the prediction model, y _act is the risk value of the hand-marked murals, and n is the number of samples. is the average value of y _act .

(2) Establishment of visualization map of onychopathia distribution

In order to obtain the overall risk assessment map of the mural, it is necessary to use the PLSR prediction model to predict each pixel of the hyperspectral image corresponding to the mural. The predicted results are normalized and range between 0-1, which can be regarded as The degree of severity of nail damage, where 0 means no nail lift and 1 means shedding occurs. Different colors are used to calibrate the results obtained, and a visual map of the distribution of nail nail damage in murals can be drawn. According to the distribution of toenail disease, the parts with different toenail degrees can be properly detected and maintained. As shown in Figure 7, it shows the results obtained by the near-infrared hyperspectral toenail disease assessment method for two mural images. Each row represents a mural image. The first column is the image of the mural at a wavelength of 1200nm, the second column is the value of the incidence of onychopathia in the corresponding position of the mural obtained through the PLSR prediction model, and the third column is the result obtained after the second column is filtered by the mean value, the purpose is to make The resulting graph is smoother.

(3) Extraction and analysis of important bands

The hyperspectral image is a high-dimensional data format, and has a high correlation between its bands, which leads to the redundancy of the hyperspectral image and the instability of the convergence of the prediction results. Selecting fewer important bands in the spectral region of the hyperspectral image for prediction can not only reduce the computational cost of the system, but also is very important for the stability and scalability of the system. In the present invention, the PLSR prediction model is used to predict the degree of onset of murals, so select important bands related to mural diseases and use PLSR to predict again, and compare with the results obtained by using all the band space mentioned above, and verify The validity and accuracy of the selected important bands. In the present invention, the selection of important bands is determined based on the weighted regression coefficients in the PLSR prediction model. The weighted regression coefficient represents the weight value of each band in the prediction in the entire band space, and the maximum and minimum values play a major role in the prediction, so select the bands corresponding to the maximum and minimum points to complete The selection of important bands.

The operation process of the present invention is illustrated below in conjunction with specific tests, see the following description for details:

The near-infrared hyperspectral imager has a spectral range of 900-1700nm, a spectral resolution of 5nm, and a total of 256 bands. Under a suitable environment, the murals with onychopathia occur, and the spatial resolution of the image is 320* 400. The spectral data of four distinct stages of onychomycosis were extracted and saved by manual labeling.

Affected by the environment and the instrument itself, the obtained mural hyperspectral images are doped with certain noises. In the present invention, the first 20 and the last 20 bands with larger noise in the image bands are removed, and the remaining 216 bands are simultaneously The bands are subjected to a minimum noise separation transformation to reduce the impact of noise in the useful bands on the results.

The partial least squares method was used to analyze the selected spectral data of the disease. The partial least squares method was used to analyze the selected spectral data of the disease. By adjusting the parameters in the PLSR model, the sum of squared errors and the average error between the predicted results and the real results are minimized, and the coefficient of determination is maximized, and the PLSR prediction model is trained and generated.

Using the trained PLSR prediction model to predict the incidence of onycholysis on a pixel-by-pixel basis for hyperspectral images of murals, and use different colors to mark the distribution of onychomycosis in murals. At the same time, according to the maximum value and minimum value of the weighted determination coefficient in the PLSR model, select the band that plays an important role in the prediction of mural nail damage and input it into the PLSR model for prediction instead of inputting the entire band space, which is effective while ensuring the results The calculation amount and calculation time in the forecasting process are reduced.

There are various types of mural diseases. The common and serious diseases in other murals include brittleness, discoloration, cracking, etc. These are the same as the onychopathic diseases, which are caused by the change of the material and chemical composition of the mural itself. The invention can be directly used for detection and analysis of other various diseases only through a small amount of modification.

In the embodiments of the present invention, unless otherwise specified, the models of the devices are not limited, as long as they can complete the above functions.

Those skilled in the art can understand that the accompanying drawing is only a schematic diagram of a preferred embodiment, and the serial numbers of the above-mentioned embodiments of the present invention are for description only, and do not represent the advantages and disadvantages of the embodiments.

The above descriptions are only preferred embodiments of the present invention, and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included in the protection of the present invention. within range.

Claims (2)

1. a kind of mural painting based on near-infrared EO-1 hyperion plays onychonosus evil appraisal procedure, it is characterised in that the mural painting plays onychonosus evil Appraisal procedure is comprised the following steps：

The image that particular wall is drawn under hundreds of continuous different spectral wavelengths is obtained by near-infrared Hyperspectral imager；

Positioning is observed by the naked eye, the key position in the different orders of severity onychonosus evil in mural painting is searched out, and in correspondence High spectrum image in corresponding position be marked, extract spectroscopic data as the disease spectrum java standard library for playing first degree；

High-spectral data is stored and is pre-processed；Spectrum characteristic data is extracted and analysis；

Wherein, above-mentioned near-infrared is 900-1700nm, and spectral resolution has 256 wave bands, the spatial discrimination of image in 5nm Rate is 320*400；

The onychonosus evil high-spectral data that rises for selecting is analyzed using PLS；By adjusting offset minimum binary mould Parameter in type so that error sum of squares, the mean error minimum predicted the outcome with legitimate reading, the coefficient of determination are maximum, carry out The training and generation of offset minimum binary forecast model；

What the forecast model obtained using training was put pixel-by-pixel to mural painting high spectrum image plays the prediction of first occurrence degree, and makes It is labeled with different colors, obtains the first disease distribution visualization figure in mural painting；

The selection of important wave band is determined based on the weighted regression coefficient in offset minimum binary forecast model, weighted regression coefficient Represent weighted value of each wave band in prediction in whole wave band spaces；

According to the maximum and minimum of weighted regression coefficient in forecast model, it is chosen at during mural painting plays first plant disease prevention and plays weight The wave band input prediction model to be acted on is predicted.

2. a kind of mural painting based on near-infrared EO-1 hyperion according to claim 1 plays onychonosus evil appraisal procedure, and its feature exists In the near-infrared Hyperspectral imager includes：Near-infrared EO-1 hyperion camera, rotation platform, computer, light source, tripod and Data wire,

The near-infrared EO-1 hyperion camera is arranged on the rotation platform, the near-infrared EO-1 hyperion camera and the rotary flat Platform is connected by the data wire with the computer, and the rotation platform is carried out in certain angle by the control of the computer Automatic rotation, while the near-infrared EO-1 hyperion camera carries out image collection with storage by the computer；The near-infrared is high Spectrum camera carries out position and fixes with the rotation platform by the tripod, and the light source enters line position by the tripod Put fixation.