JP2011180120A - Method and device for diagnosis of waterproof sheet - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To diagnose a diagnostic object without deterioration by the use of nondestructive inspection, to perform easily on site, and to evaluate degradation level quantitatively with high accuracy. <P>SOLUTION: The data of original imagery are processed to extract multiple kinds of feature values of imageries defined as elements which appear with deterioration of a waterproof sheet. A variety of feature values extracted are analyzed with multiple regression analysis to derive the residual quality or a retention of coefficient of extension at the rupture time of the waterproof sheet. At image processing, domain dividing is carried out through a binarization process to crack domain and non-crack domain. As multiple kinds of feature values, a rate of the area occupied by crack domain that is a feature value related to the size of a crack and an average width obtained from distance transform are used and, moreover, a feature value according to the local maximum value and a feature value obtained by comparing adjacent pixels are also used. <P>COPYRIGHT: (C)2011,JPO&INPIT

Description

  The present invention relates to a waterproof sheet diagnostic method and a diagnostic apparatus for diagnosing the degree of deterioration of a waterproof sheet used for a roof of a building, a balcony floor, and the like by nondestructive inspection.

Many buildings were built in Japan during the period of high economic growth and the bubble period, but many of them are now obsolete. On the other hand, due to the recent deterioration in economic conditions and the increase in environmental awareness, there is an increasing momentum to use buildings for renovation rather than rebuilding with a large cost and environmental burden.
High durability of buildings is strongly screamed from the viewpoint of global environmental conservation. When a building is used for a long time, deterioration diagnosis of a waterproof sheet can be an important factor in judging the value of the building. As the waterproof sheet ages, it gradually deteriorates under the influence of deterioration external forces such as ultraviolet rays, heat, ozone, rainwater, moisture, and repeated fatigue. Eventually, the time will come when repairs and re-working are required. In order to accurately determine the time, it is necessary to confirm the degree of deterioration of the waterproof sheet in each building.

  As an index for evaluating the degree of deterioration of a waterproof sheet, the elongation retention at break in a tensile strength test is generally used. However, although the elongation retention at break is suitable as an evaluation method in a laboratory, it is necessary to measure by cutting and collecting a waterproof sheet, and it is not suitable as a deterioration diagnosis method in an existing building site. Therefore, there is a need to inspect by some method and measure the residual strength of the waterproof sheet without actually destroying a part of the target building.

  As a conventional technique for diagnosing non-destructively or substantially non-destructively, a deterioration determination method using a hardness meter of a waterproof sheet (Patent Document 1), a deterioration determination method using a crack photograph of the waterproof sheet (Patent Document 2), a synthetic resin sheet A degradation evaluation method, a degradation evaluation scale (Patent Document 3), and the like have been proposed. In addition, regarding the determination of the strength deterioration of the membrane material covered with resin, the determination device, the determination system, the determination method, etc. by color analysis using the personal computer connected microscope, the shape analysis of the area sum of the crack area, etc. (Patent Document 4) has been proposed.

  In addition, as a past study on the estimation of the residual strength of degraded materials using image processing, Yanagimuro (Non-Patent Document 1) uses image analysis and image recognition technology as one of the more objective and effective non-destructive inspections. We have proposed a method for estimating the residual intensity from the color and morphological features of digital images on the surface of membrane materials. Enomoto et al. (Non-Patent Document 2) proposes a technique for quantifying the degree of surface cracks by image processing as one of the “suggestion of a simple deterioration diagnosis method for a polyvinyl chloride roofing sheet”. However, as a pretreatment, dust or the like adhering to the surface is removed using ethanol, and silicone grease mixed with carbon black is rubbed in order to make the crack portion stand out.

JP-A-6-27010 JP-A-6-27046 JP 2003-254893 A JP 2008-190880 A

Jun Yanagimuro, Ryu Gugen, Shigeshi Serizawa, Naoki Kato, Hiroshi Toyoda, Satoshi Fujiwara, Kenji Oda: Development of automatic diagnosis technology for film material degradation by image analysis and recognition, Membrane Structure Research Papers 2006, pp. 71-82, 2007. Yoshihiro Enomoto, Tetsuya Sudo, Nobukazu Ohba, Proposal of simple deterioration diagnosis method for polyvinyl chloride roofing sheet. Architectural Institute of Japan, No.489, pp.17-24, 1996. Supervised by Ministry of Construction Secretariat Technology Research Office, National Land Development Technology Research Center Building Durability Improvement Technology Promotion Committee, Durability Improvement Technology for Building Waterproofing. Gihodo Publishing, 1987. Vincent Luc, Pierre Soille, Watersheds in Digital Spaces: AnE.cient Algorithm Based on Immersion Simulations. IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol.13, No.6, pp.583-598, 1991.

In the deterioration determination method using a hardness meter of the waterproof sheet (Patent Document 1), it is necessary to make a hole in the sheet with a needle. Although this hole is repaired after inspection, there is a problem.
In the degradation determination method based on the crack photograph of the waterproof sheet (Patent Document 2), coloring is necessary to photograph the crack, and there are few verification samples and the accuracy is low. It is difficult to make a precise determination because a film camera is used.
In the degradation evaluation method of the synthetic resin sheet and the like (Patent Document 3), there is a problem that the waterproof sheet is colored, and the target for degradation evaluation is limited to a horizontal plane.
The determination method based on the color analysis or the shape analysis of the sum of the areas of the crack regions (Patent Document 4) cannot make a highly accurate diagnosis. Moreover, the determination object by the method of the same literature is a film | membrane material which coat | covered the fiber with resin, Comprising: With a polyvinyl chloride waterproof sheet etc., a precise diagnosis is difficult.

  In addition, none of the above previous studies (Non-Patent Documents 1 and 2) has been able to quantitatively evaluate the degree of deterioration with high accuracy, or requires an accompanying pretreatment, etc. It is difficult to make a diagnosis.

  An object of the present invention is to perform a deterioration diagnosis of a waterproof sheet by a non-destructive inspection that does not change the property of the diagnosis object, can be performed on site, can be easily performed in a short time, and has a high degree of deterioration. It is an object to provide a waterproof sheet diagnostic method and diagnostic apparatus that can quantitatively evaluate the accuracy.

The waterproof sheet deterioration diagnosis method of this invention is
An input processing step of storing in the original image storage means original image data consisting of a digital image obtained by photographing a part of the surface of the waterproof sheet to be determined;
An input processing process for storing the original image data in the original image storage means, and processing of the original image data stored in the original image storage means, and an image defined as an element appearing with the deterioration of the waterproof sheet Image processing and feature extraction process to extract multiple types of features,
And analyzing the extracted feature quantities of each type according to a predetermined rule to obtain a residual quality calculation process for obtaining a residual quality, which is a retention rate of elongation at break of the waterproof sheet.
In the image processing / feature amount extraction process, each pixel of the original image is divided into a pixel indicating a crack region and a pixel indicating a non-crack region by a binarization process, and the region divided 2 And a feature amount extraction process for extracting a feature amount from the value image.
As the plurality of types of feature amounts extracted in the feature amount extraction process, any one of an area ratio occupied by a crack region and an average width of the crack region in the binary image, which is a feature amount related to the size of a crack One or both feature amounts and the feature amount related to the width of the crack obtained by weighting using distance conversion are used.
The weighting by the distance conversion is a process of generating an image in which the value of each pixel in the crack region is weighted by the distance value assuming that the shortest distance from each pixel in the crack region to the pixel in the non-crack region is the distance value of the pixel. And at least one of the sum, median, maximum value, and standard deviation of the values of each pixel in the weighted image is used as a feature amount related to the width of the crack region.

According to this degradation diagnosis method, a part of the surface of the waterproof sheet to be judged is photographed, image processing is performed to extract feature quantities, and residual quality is obtained from the feature quantities. It can be performed by non-destructive inspection that does not change the properties of the object, can be performed in the field, is simple, and can be performed in a short time.
In this case, as the feature amount of the image, the feature amount related to the size of the crack, the ratio of the area occupied by the crack region, the feature amount of one or both of the average width of the crack region, and distance conversion were used. Since the feature amount related to the crack width obtained by weighting is used, the degree of deterioration can be quantitatively evaluated with high accuracy.
In many cases, the waterproof sheet is used for a flat roof. In this case, the waterproof sheet is considered to deteriorate due to the influence of ultraviolet rays, heat, moisture, and the like. In PVC waterproof sheets and the like, deterioration due to ultraviolet rays is the largest, and irradiation with ultraviolet rays due to sunlight leads to a decrease in molecular weight from the surface side, and this phenomenon appears in appearance changes such as generation of cracks on the surface. Therefore, an image sample of the deterioration process of the waterproof sheet was observed. As a result, it was found that as the residual quality (retention rate of elongation at breakage of the waterproof sheet) deteriorates, the width of the crack increases and the crack area widens. Although the progress of deterioration can be diagnosed to some extent only by the sum of the areas of the crack regions, it is insufficient for accurate diagnosis. In contrast, according to the present invention, since the width of a crack that significantly shows the progress of deterioration is used as a characteristic amount for diagnosis of the degree of deterioration, the degree of deterioration can be quantitatively evaluated with high accuracy. Since the crack width is obtained by weighting by distance conversion, it can be obtained from the image with high accuracy. Further, according to the present invention, as the image feature amount used for diagnosis, in addition to the width of the crack, one or both of the area ratio occupied by the crack region and the average width of the crack region are used. Since the diagnosis is performed in combination, the residual quality can be obtained with very high accuracy despite the complete non-destructive inspection that does not change the diagnosis target.

In the method of the present invention, as the feature quantity extracted in the feature quantity extraction process, in addition to the feature quantity related to the size of the crack and the feature quantity related to the width of the crack, the feature quantity obtained from the following local maximum value May be used. The feature amount obtained from the local maximum value is obtained by extracting the local maximum value that is the maximum value of the pixel value in each crack region from the image weighted by the distance value, in order of decreasing distance conversion value or Histograms are sorted in descending order and equally divided into n (n: natural number of 2 or more) levels, and the frequency of each level i (1 ≦ i ≦ n) is used as the feature amount obtained from the local maximum value. .
The local maximum value is a value indicating whether there are many thin cracks or many thick cracks. If there are many thick cracks, it is considered that deterioration has progressed. For this reason, the residual quality can be obtained with higher accuracy by adding the local maximum value as an image feature amount for diagnosis.

In the method of the present invention, as the feature quantity extracted in the feature quantity extraction process, in addition to the feature quantity related to the size of the crack and the feature quantity related to the width of the crack, as a feature quantity by comparing adjacent pixels, For a set of pixels included in a belt-like neighborhood whose distance is i (i is a natural number) pixels, the neighborhood is defined as a continuous crack region only when all pixels are cracked pixels. You may use the value which shows whether the crack area | region is distributed according to the neighborhood distance of the pixel of an area | region, and the value which shows how many continuous crack areas have spread about each pixel of a crack area | region.
The value indicating whether the crack area is distributed according to the proximity distance of the pixels in the crack area and the value indicating how far the continuous crack area is spread for each pixel in the crack area are also prominent in the progress of deterioration. Appear. Therefore, by adding the feature value obtained by comparing these adjacent pixels as an image feature value for deterioration diagnosis, it is possible to make a diagnosis by combining many types of image feature values, and the remaining quality with higher accuracy. Can be requested.

In the present invention, as the image processing / feature amount extraction process, the original image storage is performed together with a region dividing step of dividing the region by the binarization processing and a feature amount extracting step of extracting a feature amount from the binary image divided into the regions. The scalar value of each pixel of the original image stored in the means is regarded as the elevation of the terrain, and the pixel is classified into each divided cluster by dividing the region with the watershed dividing each basin as a boundary. Calculate at least one of the median, standard deviation, minimum, and maximum values of brightness, sort the calculation results in ascending or descending order, divide the values evenly into buckets, A feature amount extraction process for obtaining an average value of data as a feature quantity, and in the residual quality calculation process, the feature obtained through the region division process performed through the binarization process When, both the characteristic quantity obtained through the area division process for dividing the watershed as a boundary, may be used in the analysis of the residual quality.
In addition to the binarization process, a technique using a water divide process is also effective as a technique for segmenting the original image with respect to the cracks. This type of image feature amount can be combined and diagnosed, whereby the remaining quality can be obtained with higher accuracy.

In the present invention, the residual quality calculation process may be performed by analyzing each extracted feature quantity according to a predetermined rule to obtain a residual quality, and as the predetermined rule, image processing / feature quantity extraction process The rule for obtaining the residual quality by regression analysis from the various types of feature values extracted in (1) is preferable.
The accuracy of the deterioration diagnosis is determined by what kind of feature value is extracted and used as the feature amount indicating deterioration and how the extracted feature amount is used. According to the regression analysis, the residual quality can be obtained with higher accuracy using the actual measurement value.

  As the regression analysis, multiple regression analysis is preferable. That is, in the residual quality calculation process, all of the various types of feature quantities extracted in the image processing / feature quantity extraction process, or those selected from the feature quantities according to a predetermined criterion are used as explanatory variables, The residual quality should be obtained by multiple regression analysis using the residual quality as an objective variable. According to the multiple regression analysis, the estimated model can be re-learned according to the situation, the experimental results of the deteriorated waterproof sheet that will increase in the future can be taken in, and a more reliable model can be easily created with more samples, Residual quality can be obtained with high accuracy.

When performing multiple regression analysis, in the residual quality calculation process, create an interaction term by multiplying any two feature quantities of each type of feature quantity extracted in the image processing / feature quantity extraction process, Using each feature quantity and each interaction term as an explanatory variable, an explanatory variable is selected from these explanatory variables according to a predetermined criterion, and a residual quality is obtained by multiple regression analysis using the selected explanatory variable. Is good.
A rich selection of image features and a variable selection method that effectively derives meaningful variables from them, the estimated model can be re-learned according to the situation. A large number of samples can make a reliable model easily. Therefore, the remaining quality can be obtained with higher accuracy.

The waterproof sheet deterioration diagnosis device (1) of the present invention is:
An input processing means (5) for storing, in the original image storage means (6), original image data consisting of a digital image obtained by photographing a part of the surface of the waterproof sheet to be determined;
Image processing / feature amount extraction for processing original image data stored in the original image storage means (6) and extracting a plurality of types of feature amounts of images determined as elements appearing with the deterioration of the waterproof sheet. Means (7);
Analyzing the extracted feature quantities of each type to provide a residual quality calculating means (8) for obtaining a residual quality that is a retention rate of elongation at break of the waterproof sheet.
The image processing / feature amount extraction means (7) includes an area dividing unit (10) that divides each pixel of the original image into a pixel indicating a crack area and a pixel indicating a non-crack area by binarization processing; A feature amount extraction unit (11) for extracting a feature amount from the binary image divided into regions;
The feature quantity extraction means (11) is a feature quantity related to the size of a crack as the plurality of types of feature quantities to be extracted. Of the binary image, an area ratio occupied by a crack area, and an average of the crack areas One or both of the feature values of the width and the feature value related to the width of the crack obtained by weighting using distance conversion are used.
The weighting by the distance conversion is a process of generating an image in which the value of each pixel in the crack region is weighted by the distance value assuming that the shortest distance from each pixel in the crack region to the pixel in the non-crack region is the distance value of the pixel. And at least one of the sum, median, maximum value, and standard deviation of the values of each pixel in the weighted image is used as a feature amount related to the width of the crack region.

  According to the waterproof sheet deterioration diagnosis apparatus (1) of the present invention, the waterproof sheet deterioration diagnosis can be performed by a nondestructive inspection that does not change the property of the diagnosis object, as described above for the waterproof sheet deterioration diagnosis method of the present invention. In addition, it can be carried out in the field, can be performed easily and in a short time, and the degree of deterioration can be quantitatively evaluated with high accuracy.

  In the device according to the present invention, as the feature quantity extracted by the feature quantity extraction unit (11), in addition to the feature quantity relating to the size of the crack and the feature quantity relating to the width of the crack, Using the obtained feature value, the feature value obtained from this local maximum value is obtained by extracting the local maximum value, which is the maximum value of the pixel value in each crack region, from the image weighted by the distance value. Sorting in order of increasing or decreasing value, creating a histogram equally divided into n (n: natural number of 2 or more) levels, and calculating the frequency of each level i (1 ≦ i ≦ n) as the local maximum value It is good also as the feature-value obtained from.

  In the device according to the present invention, as the feature quantity extracted by the feature quantity extraction unit (11), in addition to the feature quantity related to the size of the crack and the feature quantity related to the width of the crack, a feature obtained by comparing adjacent pixels. As a quantity, for a set of pixels included in a belt-like neighborhood whose distance is i (i is a natural number) pixels, only when the pixel is a crack pixel, the neighborhood is regarded as a continuous crack region. It is also possible to use a value indicating whether the crack region is distributed according to the neighborhood distance of the pixel in the crack region, and a value indicating to what range the continuous crack region is spread for each pixel in the crack region.

  In the present invention apparatus, as the image processing / feature amount extraction means (7), the region dividing unit (10) for dividing the region by the binarization processing and the feature amount extraction for extracting the feature amount from the binary image obtained by the region division In addition to the part (11), a second area that is partitioned by using a watershed dividing each basin as a boundary, considering the scalar value of each pixel of the original image stored in the original image storage means (6) as the altitude of the topography For each divided cluster and each divided cluster, calculate at least one of the median value, standard deviation, minimum value, and maximum value of the lightness of the pixels classified therein, and the calculation results in ascending order Or a second feature quantity extraction unit (13) for rearranging in descending order, dividing the values equally into buckets, and obtaining an average value of data in each bucket as a feature quantity, and providing the remaining quality calculating means In 8), both the feature amount obtained through the region dividing unit (10) performed through the binarization process and the feature amount obtained through the region dividing unit (12) that partitions the watershed as a boundary are retained. It may be used for quality analysis.

In the apparatus of the present invention, the residual quality calculating means (8) may obtain the residual quality by regression analysis from each type of feature quantity extracted by the image processing / feature quantity extraction means (7).
In this case, the residual quality calculating means (7) is configured to determine all of the various types of feature quantities extracted by the image processing / feature quantity extraction means (7) or a predetermined standard among these feature quantities. The selected quality may be used as an explanatory variable, and the residual quality may be obtained by multiple regression analysis using the residual quality as an objective variable.

  In the case of performing multiple regression analysis, the residual quality calculation means (8) calculates any two feature quantities of each type of feature quantity extracted by the image processing / feature quantity extraction means (7). Create a crossed interaction term, use each feature quantity and each interaction term as an explanatory variable, select an explanatory variable from these explanatory variables according to a predetermined criterion, and use the selected explanatory variable Thus, the residual quality may be obtained by multiple regression analysis.

  The waterproof sheet deterioration diagnosis system of the present invention includes the waterproof sheet deterioration diagnosis device of any one of the above configurations of the present invention, and a photographing device for obtaining original image data input to the input processing means in the waterproof sheet deterioration diagnosis device, The photographing device has a cup-shaped lighting cover whose opening end is in contact with the waterproof sheet, and a plurality of lighting devices are provided around the center of the opening in the vicinity of the opening end in the lighting cover and emit illumination light toward the center of the opening. A plurality of light sources composed of light emitting diodes to irradiate, and a microscope with a photographing function for enlarging a part of an image of a waterproof sheet in front of the opening cover and in contact with the opening end to photograph as a digital image It is characterized by that. When a plurality of light-emitting diodes that irradiate illumination light toward the center of the opening are provided in the vicinity of the opening end in the illumination cover as the photographing device, the illumination light is irradiated substantially parallel to the surface of the waterproof sheet. As a result, the crack generated on the surface of the waterproof sheet becomes a shadow. Therefore, an original image clearly showing cracks can be obtained, and the residual quality can be obtained with higher accuracy by the waterproof sheet deterioration diagnosis of the present invention.

The waterproof sheet imaging apparatus of the present invention is an imaging apparatus for obtaining original image data to be input to the input processing means in the waterproof sheet deterioration diagnosis apparatus having any one of the above-described configurations of the present invention, wherein the opening end is brought into contact with the waterproof sheet. A plurality of light sources including a cup-shaped illumination cover, a plurality of light emitting diodes provided around the center of the opening and irradiating illumination light toward the center of the opening in the vicinity of the opening end in the lighting cover; And a microscope with a photographing function for enlarging a part of the image of the waterproof sheet in contact with the opening end and photographing as a digital image.
According to this waterproof sheet photographing apparatus, as described above, an original image clearly showing a crack can be obtained.

  The waterproof sheet deterioration diagnosis method of the present invention includes an input processing step of storing original image data consisting of a digital image obtained by photographing a part of the surface of a waterproof sheet to be determined in an original image storage means, and the original image storage means Image processing / feature amount extraction process that extracts the multiple types of feature values of the image that are defined as elements that appear along with the deterioration of the waterproof sheet, and the extracted image types And a residual quality calculation process for obtaining a residual quality which is a retention rate of the elongation rate at the time of breaking of the waterproof sheet, and the image processing / feature quantity extraction process includes: Each pixel is divided into a pixel indicating a crack region and a pixel indicating a non-crack region by a binarization process, and a special image is obtained from the binary image obtained by the region division. A feature amount extraction process for extracting a quantity, and the plurality of types of feature quantities to be extracted in the feature quantity extraction process are feature quantities related to the size of a crack. Using the feature amount of one or both of the area ratio and the average width of the crack region, and the feature amount related to the width of the crack obtained by weighting using distance conversion, the weighting by the distance conversion is A process of generating an image in which the value of each pixel in the crack region is weighted by the distance value, assuming that the shortest distance from each pixel to the pixel in the non-crack region is the distance value of the pixel, and each of the weighted images At least one of the sum, median, maximum value, and standard deviation of the pixel values is used as a feature amount related to the width of the crack region. Therefore, waterproof sheet deterioration diagnosis can be performed by non-destructive inspection that does not change the properties of the diagnosis target, can be performed on site, can be performed easily and in a short time, and the degree of deterioration can be determined with high accuracy. Can be evaluated.

  In particular, as the feature quantity extracted in the feature quantity extraction process, in addition to the feature quantity related to the size of the crack and the feature quantity related to the width of the crack, the feature quantity obtained from the local maximum value and the adjacent pixel When using the feature value obtained by dividing the region with the watershed as a boundary, the combination of a large number of image feature values in spite of the complete non-destructive inspection that does not change the diagnosis target, Residual quality can be determined with very high accuracy.

  In addition, when using image features that have been abundantly facilitated as described above, and using a variable selection method that efficiently derives meaningful variables from them, the estimated model can be re-learned according to the situation, and will increase in the future. By taking in the experimental results of the coming deteriorated waterproof sheet, a more reliable model can be easily made with a larger number of samples, and the remaining quality can be obtained with higher accuracy.

  The waterproof sheet deterioration diagnosis apparatus according to the present invention includes an input processing means for storing original image data consisting of a digital image obtained by photographing a part of the surface of a waterproof sheet to be determined in an original image storage means, and the original image storage means Image processing / feature amount extraction means for processing the original image data stored in the image and extracting each of a plurality of types of feature amounts of the image defined as elements appearing with the deterioration of the waterproof sheet, and each of the extracted types And a residual quality calculating means for obtaining a residual quality that is a retention rate of the elongation rate at the time of breaking of the waterproof sheet, wherein the image processing / feature quantity extracting means is an original image Each pixel is divided into a pixel indicating a crack region and a pixel indicating a non-crack region by a binarization process, and a feature is obtained from the binary image divided into regions. A feature amount extracting unit that extracts a crack region of the binary image that is a feature amount related to a crack size as the plurality of types of feature amounts to be extracted. Using the feature amount of one or both of the area ratio and the average width of the crack region and the feature amount related to the width of the crack obtained by weighting using the distance conversion, the weighting by the distance conversion is the crack region In this weighted image, an image is generated by weighting the value of each pixel in the crack region with the distance value, assuming that the shortest distance from each pixel to the pixel in the non-crack region is the distance value of the pixel. At least one of the total value, median value, maximum value, and standard deviation of each pixel is used as a feature value related to the width of the crack region. Therefore, the deterioration diagnosis of waterproof sheets can be performed by non-destructive inspection that does not change the properties of the diagnosis target, can be performed on site, can be performed easily and in a short time, and the degree of deterioration can be determined with high accuracy. Can be evaluated.

  The waterproof sheet imaging apparatus of the present invention is an imaging apparatus for obtaining original image data to be input to the input processing means in the waterproof sheet deterioration diagnosis apparatus of any one of the above configurations of the present invention. , A cup-shaped lighting cover whose opening end is brought into contact with the waterproof sheet, and a light emitting diode which is provided in the vicinity of the opening end in the lighting cover around the central portion of the opening and emits illumination light toward the central portion of the opening A waterproof sheet having a plurality of light sources, and a micro-coop with a photographing function for enlarging a part of the image of the waterproof sheet that is positioned in the lighting cover and in contact with the opening end to capture a digital image It is possible to obtain an original image that clearly shows the cracks.

It is a flowchart which shows each process of the waterproof sheet degradation diagnostic method which concerns on one Embodiment of this invention. It is a block diagram which shows the conceptual structure of the waterproof sheet degradation diagnostic apparatus which concerns on one Embodiment of this invention. It is a partially omitted sectional view of a waterproof sheet imaging apparatus that captures an original image used in the waterproof sheet deterioration diagnosis method. (A) is a bottom view of the waterproof sheet photographing apparatus, and (B) is a bottom view showing the waterproof sheet photographing apparatus in a state where a fixed base is removed. It is an external appearance perspective view of the waterproof sheet imaging device. It is a photograph which shows the example of a waterproof sheet imaging | photography with the waterproof sheet imaging device. It is operation | movement explanatory drawing of the waterproof sheet imaging device. (A) is a figure which shows the typical example of deterioration of a waterproof sheet, and the exceptional example of deterioration, (B) is a figure which shows the change of the waterproof sheet surface accompanying deterioration. It is a figure showing distribution of the residual quality and deterioration degree of a sample. It is a figure showing the image sample of a waterproof sheet. (A) is a figure showing the image before trimming, (B) is a figure showing the image after trimming. (A) is a figure showing the brightness value before trimming, (B) is a figure showing the brightness value after trimming. It is a figure which shows the example of a binarized image. It is a figure which shows the example of a 3x3 median filter. It is a figure which shows the example of the water separation process in a regular hexagonal grid. It is a figure which shows the example of the area | region division by a watershed process. (A) is a diagram showing an original image, (B) is a diagram showing a binary image, and (C) is a diagram showing a thinned image. It is a figure which shows the example of distance conversion. It is a figure which shows the example of the weighting by distance conversion. It is the figure which plotted the value obtained by distance conversion three-dimensionally. It is a figure which shows the vicinity with respect to an attention pixel. It is a scatter diagram (the 1) of each feature-value (horizontal axis) and residual quality (vertical axis). It is a scatter diagram (the 2) of each feature-value (horizontal axis) and residual quality (vertical axis). It is a scatter diagram (the 3) of each feature-value (horizontal axis) and residual quality (vertical axis). It is a scatter diagram (the 4) of each feature-value (horizontal axis) and residual quality (vertical axis). It is a scatter diagram of interaction term (horizontal axis) and residual strength (vertical axis). It is a figure showing the algorithm of conversion selection. It is a conceptual diagram of a cross-validation method. It is a figure which shows the precision of the cross-validation of the residual strength estimation model which does not contain an interaction term. It is a figure which shows the precision of the cross-validation of the residual strength estimation model containing an interaction term.

  A waterproof sheet deterioration diagnosis method, a diagnostic apparatus, a diagnostic system, and a waterproof sheet photographing apparatus according to an embodiment of the present invention will be described with reference to FIGS. The waterproof sheet to be diagnosed is, for example, a PVC-based waterproof sheet, in particular, a fully-adhesive PVC-based waterproof sheet, and is used on a flat roof of a building, a floor of a balcony, or the like. In brief, this waterproof sheet deterioration diagnosis method is a method for estimating the residual quality from a digital image of the surface of the waterproof sheet taken by a microscope or the like using an image analysis technique. That is, a feature amount that appears with deterioration is extracted from an image obtained by photographing the deteriorated waterproof sheet surface, and is used as an explanatory variable. Then, regression analysis is performed using the obtained explanatory variables, and the residual quality is output as an objective variable.

  FIG. 1 shows a flow chart of this waterproof sheet deterioration diagnosis method. In this method, the processing process is roughly divided into an image processing / feature extraction process (S1) for extracting image feature amounts using image analysis / image recognition technology, and a residual quality is accurately determined by data analysis from these feature amounts. The remaining quality calculation process (S2) is divided. In the image processing / feature extraction process (S1), attention is paid to cracks, which are elements that appear with deterioration in the waterproof sheet, and form features are extracted as image feature amounts. In the residual quality calculation process (S2), after creating interaction terms and selecting explanatory variables, a residual strength prediction model is constructed using multiple regression analysis, which is a typical data analysis technique, to determine the residual quality. In addition, an input processing step (S0) is included.

The input processing step (S0) includes a step (R1) of inputting original image data. The input original image data is a color digital image of the surface of the waterproof sheet, and is an image data in which pixels are arranged vertically and horizontally.
The image processing / feature extraction process (S1) includes two processes: a preprocessing process (R1) for trimming an original image, a gray scale conversion process (R2) for an original image composed of a color digital image, and the like. This is a region division / feature amount extraction process (S1a, S1b). The first region segmentation / feature amount extraction step (S1a) includes a region segmentation step (R4) for segmenting into a crack region and a non-crack region by binarization, and a process for extracting image feature amounts from the binary image. (R5). The second region segmentation / feature amount extraction process (S1b) includes a region segmentation process R6 by watershed processing (Watershed process) and a feature amount extraction step (R7) for extracting image feature amounts from the region segmentation contents. . These two series of region segmentation / feature amount extraction processes (S1a, S1b) may be performed in parallel, or one of them may be performed first. After both S1a and S1b) are performed, the process proceeds to the remaining quality calculation process (S2).

The residual quality calculation process (S2) includes a process of creating an interaction term (R8), a process of selecting explanatory variables (R9), a data analysis process (R10), and a predicted value that is a diagnosis result of the residual quality. And outputting (R11).
Each of the processes R1 to R11 will be described in detail later.

  In the image processing / feature extraction process (S1) and the remaining quality calculation process (S2), the waterproof sheet deterioration diagnosis apparatus 1 shown in FIG. 2 using a computer is used. Device 2 is used.

  The waterproof sheet deterioration diagnosis apparatus 1 of FIG. 2 includes a computer such as a personal computer, an operation system thereof, and a waterproof sheet deterioration diagnosis program that is an application program executed on the computer. Function achievement means are configured. The computer is provided with or connected to an input device 3 such as a keyboard, a mouse, a communication line connection interface, and a storage medium driver, and an output device 4 such as a liquid crystal display device or a printer for displaying an image. Is done.

  The waterproof sheet deterioration diagnosis apparatus 1 includes an input processing unit 5, an original image storage unit 6, an image processing / feature amount extraction unit 7, a remaining quality calculation unit 8, and an output processing unit 18 as the function achievement unit. The input processing means 5 is means for accepting input of original image data composed of a digital image obtained by photographing a part of the surface of the waterproof sheet to be determined and storing it in the original image storage means 6. The input of the original image data may be received from the waterproof sheet photographing apparatus 2 via a wiring or a communication network, or the original image data photographed by the waterproof sheet photographing apparatus 2 and stored in the storage medium. May be read from the storage medium. The original image storage means 6 is composed of a memory or other storage medium provided in the computer.

  The image processing / feature amount extraction unit 7 processes the original image data stored in the original image storage unit 6 and extracts a plurality of types of feature amounts of the image determined as elements appearing with the deterioration of the waterproof sheet. It is means to do. The image processing / feature quantity extraction means 7 includes a preprocessing section 9, a first area dividing section 10, a first feature quantity extracting section 11, a second area dividing section 12, and a second feature. A quantity extraction unit 13. The pre-processing section 9 is means for executing the pre-processing process (R2) and the gray scale conversion process (R3) in FIG.

  The first region dividing unit 10 is a unit that divides each pixel of the original image into a pixel indicating a crack region and a pixel indicating a non-crack region by binarization processing. The first feature amount extraction unit 11 is a unit that extracts a feature amount from the binary image divided into regions. The first area dividing unit 10 and the first feature quantity extracting means 11 are means for executing the first area dividing process (R4) and the first feature quantity extracting process (R5) in FIG. 1, respectively.

  In FIG. 2, the second area dividing unit 12 is means for dividing an area by a watershed process. The second feature quantity extraction unit 13 is a means for extracting the feature quantity of the image from the area division contents. The second region dividing unit 12 and the second feature amount extracting unit 13 are means for executing the second region dividing step (R6) and the first feature amount extracting step (R7) in FIG.

In FIG. 2, the remaining quality calculation means 8 analyzes each type of feature quantity extracted by the image processing / feature quantity extraction means 7 according to a predetermined rule, and uses the retention rate of the elongation rate when the waterproof sheet breaks. It is a means for obtaining a certain residual quality. This residual quality is also called residual strength. The residual quality calculation means 8 includes an interaction term creation unit 14, an explanatory variable selection unit 15, a data analysis unit 16, and a measured value storage unit 17.
The interaction term creating unit 14 is a means for creating an interaction term that is obtained by multiplying any two feature quantities among the various types of feature quantities extracted by the image processing / feature quantity extraction means 7. The explanatory variable selection unit 15 is a unit that selects an explanatory variable to be used for analysis from among each type of characteristic amount extracted by the image processing / feature amount extraction unit 7 according to a predetermined criterion. The data analysis unit 16 is a means for obtaining the residual quality by multiple regression analysis using the explanatory variables selected by the explanatory variable selection unit 15. The measured value storage unit 17 is means for storing data used for the multiple regression analysis by the data analysis unit 16, and the stored contents are updated sequentially.
The interaction term creation unit 14, the explanatory variable selection unit 15, and the data analysis unit 16 execute the interaction term creation process (R8), the explanatory variable selection process (R9), and the data analysis process (R10) in FIG. It is means to do.

  The output processing means 18 in FIG. 2 is means for outputting the residual quality obtained by the residual quality calculating means 8 to the output device 4.

  The specific configuration of each means constituting the waterproof sheet deterioration diagnosis device 1 of FIG. 2 and the contents of the processing performed in each process in the flowchart of FIG. 1 will be described together with the following description of each examination content.

(Target materials, image shooting, pre-processing)
-Target waterproof sheet In this embodiment and each test, a vinyl-based waterproof sheet (full adhesion type) is targeted.
・ Evaluation Method and Degree of Degradation Criteria A conventional waterproof sheet is diagnosed for its deterioration state by using a tensile tester and evaluated by the retention rate of the elongation rate when the waterproof sheet breaks.

As shown in Table 1, the criteria for determining the degree of deterioration of a waterproof sheet performed using a tensile tester are determined from the degree of deterioration I to III based on the elongation retention rate. Deterioration degree I maintains 60% or more of the initial value ratio, Degradation degree II is 30% or more and less than 60% of the initial value ratio, and Degradation degree III is 30% or less of the initial value ratio. Partial repair, large scale or partial repair, large A scale improvement is required.

・ Deterioration process PVC-based tarpaulins are often used on flat roofs, and they are used with the tarpaulins fully exposed outdoors. In such a case, the waterproof sheet is considered to deteriorate due to the influence of ultraviolet rays, heat, moisture and the like. According to Sera et al. (Non-Patent Document 3), polyvinyl chloride resin is the most deteriorated by ultraviolet rays, and ultraviolet irradiation by sunlight leads to a decrease in molecular weight from the surface side, and this phenomenon appears in appearance changes such as generation of surface cracks. It is going to be. In this way, the element that appears with deterioration in the waterproof sheet is a crack, and the research proceeds on the assumption that the shape of the crack changes with the deterioration. The left side of FIG. 8 (A) is such a typical example, and the right side of FIG. 8 (A) is an exceptional case where the surface is worn due to human walking. Further, from the image sample in FIG. 8B, the remaining strength is 100%, 62%, 28%, 4%, and the crack width becomes wider and the crack area is expanded as the deterioration progresses. Recognize.

-Waterproof sheet sample About 46 samples which showed the typical deterioration pattern of the waterproof sheet sample which performed the tension test, residual quality has distribution like FIG.
In this embodiment, the surface of the waterproof sheet is photographed by the devised microscope-type waterproof sheet photographing apparatus 2 (FIGS. 3 to 5).
We take a square size and take it with the highest possible resolution in consideration of the meaning of the document, and save it as a 24-bit color jpg (high quality). However, since such high resolution is not necessary for image analysis, a photographed image reduced to 640 × 640 pixels is used. FIG. 10 shows a sample image of the waterproof sheet. The shooting range is about 2mm in length and width.

-Trimming of the image The photographed image sample is bright in the center and dark in the periphery due to the problem of the amount of light around the camera lens. In addition, when focusing on the center, the focus on the periphery tends to be slightly sweet. Although the sample used this time was taken on a horizontal plane, the waterproof sheet may be used in a curved place, so the above problem appears to be more prominent. Since these optical corrections are difficult, a 240 × 240 pixel central portion of the image, which is considered to be less affected by uneven illumination and out-of-focus, is extracted and analyzed (FIG. 11). In the preprocessing step (R2) of FIG. 1 and the preprocessing and the like processing unit 9 of FIG.

  For reference, the images before and after trimming are decomposed into the Munsell color system HSV color system, and graphs of the lightness values (V) for each pixel are shown in FIGS. 12A and 12B, respectively. The brightness value, which was low in the peripheral area before trimming, is in a nearly uniform state after trimming.

(Feature extraction using image analysis)
・ Outline of feature extraction Focusing on cracks appearing on the surface of the waterproof sheet from the consideration of changes in the waterproof sheet due to deterioration, the feature amount is extracted.

Gray scale conversion n gradation gray scale conversion is performed as a pre-processing of area division from a photographed 24-bit color image. This gray scale conversion is performed in the gray scale conversion process (R3) in FIG. 1 and the preprocessing section 9 in FIG. In this embodiment, for binarization processing, the image is converted into a gray scale image of 256 gradations that is generally used. On the other hand, in the watershed process (Wartershed process), if the gradation of the pixel is too fine, noise is easily picked up and appropriate division is not performed. Therefore, a slightly rough gray scale conversion of 64 gradations is performed.

・ Area division Try to divide the cracked part and the non-cracked part. There are many area division methods, but here, a method of binarizing the original image and a method called a watershed method are used. The first area dividing process (R1) in FIG. 1 and the first area dividing unit 10 in FIG. 2 perform area division by the following binarization process. The second region dividing process (R6) in FIG. 1 and the second region dividing unit 12 in FIG. 2 perform region division by the following watershed method (Watershed).

-Area division by binarization processing By binarization processing, an image is classified into a crack area and a non-crack area from a sample image. In the following, binarization processing is performed with the crack to be taken out as black (0) and the other crack as white (1).
As a binarization processing method, a typical p-tile method and a discriminant analysis method were examined. The p-tile method is a method of binarization by designating the ratio of the area to be binarized to the entire image area. On the other hand, the discriminant analysis method is a method in which, when binarized, a threshold value is automatically determined so that the variance ratio of intra-class variance and inter-class variance for each region is maximized. Any of these methods may be used, but here, a discriminant analysis method with high division accuracy was adopted as a result of experiments using samples. Further, a median filter is applied to the image binarized by the discriminant analysis method to perform fine noise removal.
FIG. 13 shows binary images obtained from the 62%, 28%, and 4% image samples shown in FIG.

The outline of discriminant analysis and median filter is shown below.
a) Discriminant Analysis Method When the density value distribution in the density value histogram of an image is divided into two classes (t or more and less than t) with a threshold value t, t is determined so that the two classes are best separated. Is a threshold selection method based on discriminant analysis. As a measure of separability, the ratio (discriminant ratio) of the variance of the average values of two classes (interclass variance) and the variance of each class (intraclass variance) is used, so that the discriminant ratio is maximized. Is selected. The method for obtaining t is shown below.

  Assume that the given image has L level density values (1, 2,..., L). Here, the threshold value is divided into two classes, ie, a pixel having a density value equal to or greater than t and a pixel having a smaller value than the threshold value, which are class 1 and class 2, respectively. If the number of pixels of class i (i = 1, 2) is Wi, the variance is σi2, and the average density value of all pixels is Mr, the intra-class variance is

となる。

It becomes.

Since the total variance σT ² is a constant unrelated to t, it can be seen that σB ² should be maximized in order to maximize the discrimination ratio. That is, the value of t that maximizes σB ² by changing t may be obtained.

b) Median filter The median filter removes noise by arranging the density values in the n × n local area in ascending order and setting the median as the output density of the pixel at the center of the area. This is particularly effective in the case of random white noise that is unrelated to the contents. When considering a median filter in a local region of n = 3, when the original image is FIG. 14, when density values in the vicinity of 8 of the central pixel of interest are arranged in ascending order, {1, 1, 2, 2, 2 , 2, 3, 3, 8}, and replace 8 of the pixel of interest with this median of 2. By performing such processing on all pixels of the image, sesame salt-like noise can be removed.

-Area division by watershed processing (Watershed processing) Watershed processing (Watershed processing) is a region where the scalar value of each pixel of the image is regarded as the elevation of the terrain and the watershed separating each basin is the boundary of the region This is a division method. As shown in FIG. 15, when water flows from a high position to a low position under the condition that there are two local minimum elevation values, a portion that serves as a boundary for flowing into one of the two is divided. This is called a ridge (Watershed), and the area is divided at the ridge. In this embodiment, the 256-level gray-scale brightness value of the image is used as the altitude, and the region is divided using a diversion ridge formed by a square grid. An example is shown in FIG.

-Feature extraction Next, a feature amount indicating the feature of each region divided is extracted. In the binary image, various feature amounts relating to the divided crack regions are calculated. The feature amount relating to the crack region is calculated by the first feature amount extraction process (R5) in FIG. 1 and the first feature amount extraction unit 11 in FIG. In the watershed image, a feature amount is calculated regarding the brightness distribution of each divided area. The calculation of the feature quantity relating to the brightness distribution is performed by the second feature quantity extraction process (R7) in FIG. 2 and the second feature quantity extraction unit 13 in FIG.

-Feature extraction from binary images a) Crack size Since it is considered that the width of the surface crack increases and the crack area increases with the progress of deterioration, the area ratio (CrackRatio) occupied by the crack area, crack area The average width (CrackWidth) is calculated by the following formula.

  Here, CA is the number of pixels in the crack area, TA is the number of pixels in the image (= 240 × 240), CL is the length of the crack, and the unit of measurement is 1 pixel. Note that the length of the crack here is the number of pixels remaining after thinning processing (FIG. 17C) that leaves only the center line segment of the crack region in the binarized image (FIG. 17B). To do.

b) Distance conversion It is considered that the width of the surface crack increases with the progress of deterioration. Therefore, weighting using distance conversion is performed in order to extract the shape feature of the crack accompanying this deterioration.
In the distance conversion, the shortest distance from each pixel of the figure to the non-crack pixel is the distance value in that pixel, and the value of each pixel is converted so that the value increases as it goes to the center of the figure.
For a binary image composed of 0 and 1 by binarization processing, a distance between a pixel having a value of 0 and a pixel having a value of 1 nearest to the pixel is calculated, and an image based on the distance is created. In this embodiment, for two pixels x = (i, j) and y = (k, l)

と表されるユークリッド距離を用いて距離を測定することとする。図１８に３×３ピクセルで構成される２値画像の距離変換の例を示す。

The distance is measured using the Euclidean distance expressed as follows. FIG. 18 shows an example of distance conversion of a binary image composed of 3 × 3 pixels.

  When weighting is performed by distance conversion, a difference between an image composed of thin cracks and an image composed of thick cracks can be expressed as shown in FIG. Both of these two examples have the same crack area, but the sum of the values obtained by converting the distance between them is 15 for a thin crack and 20 for a thick crack. In the case of an image composed of thick cracks considered to be deteriorated in this way, even if the area ratio of cracks is about the same as that composed of thin cracks, a larger value can be obtained. In this way, from the image obtained by distance-converting the binary image, the sum is extracted as a feature amount by Dsum, the median value is Dmedian, the maximum value is Dmax, and the standard deviation is Dstd.

c) Local maximum value FIG. 20 is a three-dimensional plot of the values obtained by the distance conversion described above. This peak portion corresponds to a crack portion in the binary image, and it means that the higher the peak is, the wider the crack width is due to the nature of distance conversion. Each peak of this mountain is the widest part of the crack, that is, the center of the crack, and the feature amount is extracted from this local maximum value.
First, only a local maximum value is extracted from an image obtained by distance conversion, and a histogram is created by sorting in order of increasing distance conversion value. As a result of the examination, the local maximum value is equally divided into 10 levels to create a histogram, and the frequency of each level i (1 ≦ i ≦ 10) is set as Maxima as the feature amount. This index represents whether there are many small peaks (thin cracks) or many large peaks (thick cracks) in the entire image. That is, if the histogram has a large frequency in the Maxima 1 direction, there are many thin cracks. Conversely, if the frequency in the Maxima 10 direction is large, there are many thick cracks.

とし、そこから距離がｉピクセルだけ離れた帯状の近傍（図２１）に含まれるピクセルの集合を

とする。 d) Adjacent pixel For the pixel x of the binary image, the value is represented as v (x) = {0, 1}. Now, let v (x) = 0, that is, the set of pixels classified as cracks.

And a set of pixels included in a belt-like neighborhood (FIG. 21) that is i pixels away from it.

And

-Feature extraction from the watershed image Next, for each cluster divided by the watershed processing, the average value, median, standard deviation, minimum value of the brightness of the pixels classified there, and Calculate the maximum value. These are rearranged in ascending order, and the values are equally divided into I and converted into buckets, and the average values of the data in the bucket of level i are extracted as feature values by means of meani, mediani, stdi, mini, and maxi, respectively. As a result of the examination, here, I = 9 divisions.

Next, the remaining quality calculation process (S2) (FIG. 1) will be described. In the interaction term creation process (R8) in FIG. 1 and the interaction term creation output 14 in the residual quality calculation means 8 in FIG. 2, an interaction term is created as follows.
(Correlation analysis)
-Each feature-value and residual quality The correlation coefficient of each feature-value and residual quality is shown in Table 2, and a scatter diagram is shown in FIGS. Only a part of the scatter diagram is shown, and the others are not shown. Apart from the image features, the years of use when the sample strength experiment was performed is added as Year.

-Interaction term and residual strength The explanation of the interaction term is made in 5.1, but in brief, it is an operation of creating a new attribute by multiplying single attributes. In the case of this embodiment, 3403 interaction terms are newly created from 83 attributes. Since it is complicated to show the relationship with the residual strength for all of these, only the interaction term used in the model shown in 5.6.2 is correlated with the residual strength in Table 3, and a scatter diagram is shown. 26. In addition, the symbol of the interaction term in the figure is an abbreviation excluding the average value.

Construction of Residual Strength Prediction Model In this embodiment, regression analysis of residual quality (elongation retention rate) is performed by multivariate analysis including nonlinearity including interaction terms. This regression analysis is performed by the data analysis process (R10) of FIG. 1 and the data analysis unit 16 of FIG. 2, and the selection of explanatory variables at that time is performed by the explanatory variable selection process (R9) of FIG. This is performed by the explanatory variable selection unit 15.
New methods such as support vector machine regression, which has been attracting attention in recent years, may be used for the regression analysis model, but in this embodiment, orthodox multiple regression analysis is adopted in terms of prediction accuracy and calculation speed. . The model construction procedure and verification results are shown below.

-Creation of interaction term In addition to the feature quantity extracted by the method described above, an interaction term is created. An interaction term is defined as the product of two different explanatory variables. However, in the creation of the interaction term, the variables are centered in order to avoid an increase in the correlation between the main effect variable and the interaction term variable. The variable centering is an operation of subtracting the average value of the corresponding variable from the value of the original variable. A variable generated by the interaction is represented by connecting two variable names with an underscore, such as “variable 1_variable 2”. In summary, the interaction term is calculated as follows:
Interaction term between variable 1 and variable 2 = (variable 1 − average of variable 1) × (variable 2 − average of variable 2)

・ Multiple regression analysis

The actual measurement value Y ⁱ used for the multiple regression analysis is stored in the actual measurement value storage unit 17 of FIG. The input processing means shown in the figure has a processing function for storing measured values Y ⁱ in addition to the processing for storing the original image. The data analysis unit 16 stores an unknown coefficient a and an intercept b obtained in the process of multiple regression analysis.

-Variable selection In multiple regression analysis, the problem of multiple collinearity is likely to occur when dealing with a large number of explanatory variables. In order to improve the generalization of the model, it is necessary to narrow down the explanatory variables as much as possible. There are a number of variable selection methods. Here, variables are selected by a method called stepwise method in which variables are added or deleted sequentially using the significance (P value) of explanatory variables of the model as a criterion. The specific procedure is shown in FIG.

  Note that the interaction term is introduced only when all three variables, ie, the variable before the combination and the interaction term are significant. In this embodiment, Pi = Po = 0.25.

・ Accuracy evaluation method

-Cross-validation method In this embodiment, a model is constructed using 46 images. However, applying a multiple regression model to all data, and constructing a model with high accuracy, increases the explanatory variables. It ’s not difficult. However, if the explanatory variables are increased, the multicollinearity and generalization will decrease, and the prediction accuracy for unknown data will decrease.
In order to approximately grasp the prediction accuracy for unknown data, accuracy evaluation is performed by a method called a cross-validation method. As shown in FIG. 28, this is done by dividing the data into v, using v-1 set of data for learning, and using the remaining 1 set of data for testing, and changing this combination v times. An attempt is made to obtain approximate prediction accuracy for unknown data. In this method, the accuracy evaluation index (a, b) of 6.4 is the average value for each test data. In addition, v is generally about 10 times, and v = 10 times in this embodiment.

Construction of residual strength estimation model a) Model that does not include interaction terms First, the model was constructed by selecting variables using the 83 feature quantities shown in Table 2 as explanatory variables without including interaction terms. . As a result, the model shown in Table 4 was constructed. In addition, FIG. 29 shows the prediction accuracy when the cross-validation is performed 10 times with this combination of explanatory variables.
Except median2, it is a significant variable with a significance probability of 5%, and there is no multicollinearity. The accuracy of cross-validation is relatively good, and even this is a model that does not have any significant problems in practice.

b) Model including interaction terms Next, the interaction terms up to the second order are created from 83 explanatory variables, and the variables are selected from the primary (83) secondary (3,403) explanatory variables. The residual strength estimation model shown in Fig. 1 was obtained. FIG. 30 shows the result of obtaining the prediction accuracy by performing 10 cross-validations with this combination of variables.
The P value of each variable of the model is very small and the learning accuracy is very high. Furthermore, the accuracy of cross-validation is very high, and good estimation results can be expected for unknown sheets.

The waterproof sheet deterioration diagnosis method and diagnosis apparatus 1 according to this embodiment have the following features in order to perform each process as described above.
-Despite complete non-destructive inspection that does not change the target, the residual intensity can be estimated with very high accuracy by combining multiple image feature quantities.
・ By using abundantly prepared image features and variable selection methods that efficiently derive significant variables from them, it is possible to re-learn the estimation model according to the situation and incorporate the results of experiments on the deteriorated waterproof sheet that will increase in the future. With a larger number of samples, a reliable model can be easily created.

  A specific example of the waterproof sheet photographing apparatus 2 will be described with reference to FIGS. The waterproof sheet photographing apparatus 2 is an apparatus for obtaining an original image used in the waterproof sheet deterioration diagnosis method and apparatus, and is attached to the microscope 21 and the upper end of the microscope 21 as shown in FIG. And a jig 22 which is a handle or the like held by the operator, and a lighting cover 23. As shown in FIG. 3, the illumination cover 23 has a cup shape that is upside down with the opening end in contact with the waterproof sheet, and has a cylindrical shape with a ceiling surface portion. An elastic sealing material 24 such as an O-ring that presses against the waterproof sheet and closes the gap is provided on the periphery that is the opening end of the illumination cover 23.

  In the vicinity of the opening end in the illumination cover 23, there are a plurality of light sources 25 (four in the illustrated example) that irradiate illumination light around the center of the opening toward the center of the opening as shown in FIG. ) Equally arranged. The light source 25 is made of a light emitting diode, is attached to the fixed base 26, and the irradiation surface 25a faces the center of the opening. The fixed base 26 has a disk shape having a central opening and a notch 26 a in a part in the circumferential direction, and is attached to the lower end of the frame 27 in the lighting cover 23. The frame 27 is provided with a battery case portion 27a and accommodates a battery 28 made of AAA batteries or the like. A battery case lid 29 is provided in the cutout portion 26a of the fixed base 26 so as to be openable and closable. The frame 27 is provided with a light source driving power supply base 30 for driving the light source 25 made of a light emitting diode. An illumination on / off switch 31 is provided on the upper surface of the illumination cover 23.

  The microscope 21 incorporates an image enlarging optical system and an image sensor (not shown) such as a solid-state image sensor, and is fixed to the illumination cover 23 so that the tip is positioned in the illumination cover 23. And a function of enlarging a part of the waterproof sheet in contact with the opening edge of the illumination cover 23 and photographing it as a color digital image. The microscope 21 has a focus adjustment knob 32 and incorporates a storage medium (not shown) for recording captured images.

  According to the waterproof sheet photographing apparatus 2 of this configuration, since a plurality of light sources 25 that irradiate illumination light toward the center of the opening are provided in the vicinity of the opening end in the illumination cover 23, as shown in FIG. The illumination light is applied to the surface of the waterproof sheet W substantially in parallel, and the crack Wa generated on the surface of the waterproof sheet W becomes a shadow. Therefore, an original image clearly showing the crack Wa can be obtained, and the remaining quality can be obtained with higher accuracy by the waterproof sheet deterioration diagnosis. FIG. 6 shows an example of an image of the surface of the waterproof sheet taken by the waterproof sheet photographing apparatus 2.

DESCRIPTION OF SYMBOLS 1 ... Waterproof sheet degradation diagnostic apparatus 2 ... Waterproof sheet imaging device 5 ... Input processing means 6 ... Original image storage means 7 ... Image processing / feature amount extraction means 8 ... Residual quality calculation means 9 ... Pre-processing etc. processing part 10 ... 1st Region dividing unit 11 ... first feature amount extracting unit 12 ... second region dividing unit 13 ... second feature amount extracting unit 14 ... interaction term creating unit 15 ... explanatory variable selecting unit 16 ... data analyzing unit 17 ... Measured value storage unit 21 ... Microscope 22 ... Jig 23 ... Lighting cover 25 ... Light source

Claims (16)

An input processing step of storing in the original image storage means original image data consisting of a digital image obtained by photographing a part of the surface of the waterproof sheet to be determined;
Image processing / feature amount extraction process for processing original image data stored in the original image storage means and extracting each of a plurality of types of feature amounts of the image defined as elements appearing with the deterioration of the waterproof sheet;
Analyzing the extracted feature quantities of each type according to a predetermined rule, and including a residual quality calculation process for obtaining a residual quality which is a retention rate of elongation at break of the waterproof sheet,
In the image processing / feature amount extraction process, each pixel of the original image is divided into a pixel indicating a crack region and a pixel indicating a non-crack region by a binarization process, and the region divided 2 A feature amount extraction process for extracting feature amounts from a value image,
As the plurality of types of feature amounts extracted in the feature amount extraction process, any one of an area ratio occupied by a crack region and an average width of the crack region in the binary image, which is a feature amount related to the size of a crack Using one or both feature values and the feature value related to the width of the crack obtained by weighting using distance conversion,
The weighting by the distance conversion is a process of generating an image in which the value of each pixel in the crack region is weighted by the distance value assuming that the shortest distance from each pixel in the crack region to the pixel in the non-crack region is the distance value of the pixel. And using at least one of the sum, median, maximum, and standard deviation of the values of each pixel in the weighted image as a feature amount related to the width of the crack region,
A method for diagnosing deterioration of a waterproof sheet, characterized in that: The feature amount obtained from the following local maximum value as the feature amount extracted in the feature amount extraction process according to claim 1, in addition to the feature amount related to the size of the crack and the feature amount related to the width of the crack. Use
The feature amount obtained from the local maximum value is obtained by extracting the local maximum value that is the maximum value of the pixel value in each crack region from the image weighted by the distance value, in order of decreasing distance conversion value or Histograms are sorted in descending order and equally divided into n (n: natural number of 2 or more) levels, and the frequency of each level i (1 ≦ i ≦ n) is used as the feature amount obtained from the local maximum value. Waterproof sheet deterioration diagnosis method.   3. The method according to claim 1, wherein the feature amount extracted in the feature amount extraction process is based on comparing adjacent pixels in addition to the feature amount related to the size of the crack and the feature amount related to the width of the crack. As a feature amount, a crack region in which the neighborhood of a set of pixels included in a belt-like neighborhood separated by i (i is a natural number) pixels is a crack pixel only when all of the pixels are crack pixels. Waterproof sheet deterioration diagnosis using a value indicating whether the crack area is distributed according to the neighborhood distance of the pixel in the crack area and a value indicating how far the continuous crack area is spread for each pixel in the crack area Method.   4. The image processing / feature amount extraction process according to claim 1, wherein an area division process for dividing an area by the binarization process and a feature quantity are extracted from the binarized image obtained by the area division. Along with the feature extraction process, the scalar value of each pixel of the original image stored in the original image storage means is regarded as the altitude of the terrain, and the region dividing process is performed by dividing the watershed dividing each basin as a boundary. For each cluster, calculate at least one of the median value, standard deviation, minimum value, and maximum value of the lightness of the pixels that are classified, and sort the calculation results in ascending or descending order. And dividing into buckets, and using a feature quantity extraction process for obtaining an average value of data in each bucket as a feature quantity. In the residual quality calculation process, A feature quantity obtained through the frequency division process, both the characteristic quantity obtained through the area division process for dividing the watershed as a boundary, tarpaulins degradation diagnostic method used for analysis of the residual quality.   5. The waterproof sheet deterioration diagnosis according to claim 1, wherein in the residual quality calculation process, the residual quality is obtained by regression analysis from each type of feature quantity extracted in the image processing / feature quantity extraction process. Method.   6. The residual quality calculation process according to claim 5, wherein, in the residual quality calculation process, one of the types of feature quantities extracted in the image processing / feature quantity extraction process is selected as an explanatory variable, and the residual quality is an objective variable. As a waterproof sheet deterioration diagnosis method for obtaining residual quality by multiple regression analysis.   7. The residual quality calculation process according to claim 6, wherein in the residual quality calculation process, an interaction term is created by multiplying any two feature quantities among each kind of feature quantities extracted in the image processing / feature quantity extraction process, Waterproof sheet deterioration diagnosis that uses quantity and each interaction term as an explanatory variable, selects an explanatory variable from these explanatory variables according to a predetermined criterion, and obtains residual quality by multiple regression analysis using the selected explanatory variable Method. Input processing means for storing original image data consisting of a digital image obtained by photographing a part of the surface of the waterproof sheet to be determined in the original image storage means;
Image processing / feature amount extraction means for processing original image data stored in the original image storage means and extracting each of a plurality of types of feature amounts of the image defined as elements appearing with the deterioration of the waterproof sheet;
Analyzing each extracted feature amount according to a predetermined rule, and comprising a residual quality calculating means for obtaining a residual quality that is a retention rate of elongation at break of the waterproof sheet,
The image processing / feature amount extraction means includes a region dividing unit that divides each pixel of the original image into a pixel indicating a crack region and a pixel indicating a non-crack region by a binarization process, and the region-divided 2 A feature amount extraction unit that extracts a feature amount from a value image;
The feature quantity extraction unit is a feature quantity related to the size of a crack as the plurality of types of feature quantities to be extracted. Of the binary image, any one of an area ratio occupied by a crack area and an average width of the crack area Using one or both of the feature amounts and the feature amount relating to the width of the crack obtained by weighting using distance conversion,
The weighting by the distance conversion is a process of generating an image in which the value of each pixel in the crack region is weighted by the distance value assuming that the shortest distance from each pixel in the crack region to the pixel in the non-crack region is the distance value of the pixel. And using at least one of the sum, median, maximum, and standard deviation of the values of each pixel in the weighted image as a feature amount related to the width of the crack region,
A waterproof sheet deterioration diagnosis device characterized by that. 9. The feature quantity obtained from the following local maximum value as the feature quantity extracted by the feature quantity extraction unit according to claim 8, in addition to the feature quantity related to the size of the crack and the feature quantity related to the width of the crack. Use
The feature amount obtained from the local maximum value is obtained by extracting the local maximum value that is the maximum value of the pixel value in each crack region from the image weighted by the distance value, in order of decreasing distance conversion value or Histograms are sorted in descending order and equally divided into n (n: natural number of 2 or more) levels, and the frequency of each level i (1 ≦ i ≦ n) is used as the feature amount obtained from the local maximum value. Waterproof sheet deterioration diagnosis device.   The feature quantity extracted by the feature quantity extraction unit according to claim 8 or 9, by comparing adjacent pixels in addition to the feature quantity relating to the size of the crack and the feature quantity relating to the width of the crack. As a feature amount, a crack region in which the neighborhood of a set of pixels included in a belt-like neighborhood separated by i (i is a natural number) pixels is a crack pixel only when all of the pixels are crack pixels. Waterproof sheet deterioration diagnosis using a value indicating whether the crack area is distributed according to the neighborhood distance of the pixel in the crack area and a value indicating how far the continuous crack area is spread for each pixel in the crack area apparatus.   11. The image processing / feature amount extraction unit according to claim 8, wherein the image processing / feature amount extraction unit extracts a feature amount from the region dividing unit that divides the region by the binarization processing and the region-divided binary image. In addition to the feature quantity extraction unit, a scalar value of each pixel of the original image stored in the original image storage means is regarded as the altitude of the terrain, and a second region dividing unit that partitions the basin as a boundary For each of these divided clusters, calculate at least one of the median, standard deviation, minimum, and maximum values of the brightness of the pixels that are classified there, and sort the calculation results in ascending or descending order, A second feature amount extraction unit that equally divides the value into a plurality of buckets and obtains an average value of data in each bucket as a feature amount, and the residual quality calculation means includes the binarization process. A feature amount obtained through the area dividing section for performing through, both the characteristic quantity obtained through the area division unit for dividing the watershed as a boundary, tarpaulins deterioration diagnostic device for use in the analysis of the residual quality.   12. The waterproof sheet deterioration diagnosis according to claim 8, wherein the residual quality calculation means obtains the residual quality by regression analysis from each type of feature quantity extracted by the image processing / feature quantity extraction means. apparatus.   13. The residual quality calculating means according to claim 12, wherein the remaining quality calculation means uses, as an explanatory variable, a feature quantity selected by a predetermined criterion from among each type of feature quantity extracted by the image processing / feature quantity extraction means, and uses the remaining quality as an objective variable. As a waterproof sheet deterioration diagnosis device that calculates the residual quality by multiple regression analysis.   14. The residual quality calculation unit according to claim 13, wherein the remaining quality calculation unit creates an interaction term by multiplying any two feature amounts among the respective types of feature amounts extracted by the image processing / feature amount extraction unit. Waterproof sheet deterioration diagnosis that uses quantity and each interaction term as an explanatory variable, selects an explanatory variable from these explanatory variables according to a predetermined criterion, and obtains residual quality by multiple regression analysis using the selected explanatory variable apparatus. A waterproof sheet deterioration diagnostic apparatus according to any one of claims 8 to 14, and a photographing apparatus for obtaining original image data to be input to input processing means in the waterproof sheet deterioration diagnostic apparatus,
The photographing device has a cup-shaped lighting cover whose opening end is in contact with the waterproof sheet, and a plurality of lighting devices are provided around the center of the opening in the vicinity of the opening end in the lighting cover and emit illumination light toward the center of the opening. A plurality of light sources composed of light emitting diodes to irradiate, and a microscope with a photographing function for enlarging a part of an image of a waterproof sheet in front of the illumination cover and in contact with the opening end to photograph as a digital image A waterproof sheet deterioration diagnosis system characterized by that. An imaging device for obtaining data of an original image to be input to an input processing means in the waterproof sheet deterioration diagnostic apparatus according to any one of claims 8 to 14,
From a cup-shaped lighting cover that makes the opening end contact the waterproof sheet, and a light emitting diode that is provided in the vicinity of the opening end in the lighting cover around the center of the opening and emits illumination light toward the center of the opening A plurality of light sources, and a microscope with a photographing function for enlarging a part of the image of the waterproof sheet that is located in the lighting cover and in contact with the opening end to photograph as a digital image. Waterproof sheet imaging device.

Images