JP5375398B2 - Paint degradation diagnosis method and apparatus - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a method and device for diagnosing coating degradation that detect degradation situation of a coating surface of a steel structure, such as, a steel tower and steel bridge, and for determining the degree of degradation. <P>SOLUTION: The method of diagnosing coating degradation comprises a process of building a degradation section extracting filter for converting an input image obtained by photographing the surface of the coated steel structure into a conversion where only degradation section of the coating surface is transmitted; a process of detecting a transmission section in the converted image and creating a degradation section profile; a process of creating a virtual coating surface degradation image simulating occurrence/growth of the coating surface degradation; a process of creating a degradation profile for evaluation from the created virtual coating surface degradation image and degradation reference, accumulating it, and creating a database; and a degradation evaluating process of determining the degradation degree by comparing the degradation profile and the degradation profile for evaluation obtained from the input image whose degradation is diagnosed. <P>COPYRIGHT: (C)2010,JPO&amp;INPIT

Description

  The present invention relates to a coating deterioration diagnosis method and apparatus for detecting a deterioration state of a painted surface of a steel structure such as a steel tower or a steel bridge and determining the degree of progress of the deterioration.

  Steel structures such as steel towers and steel bridges are periodically inspected for deterioration of the coating applied to the steel surface for rust prevention, and repairs such as repainting are performed based on the inspection results. .

  Conventionally, in order to diagnose the deterioration status of the painted surface of steel structures, the operator inspects the structure to be diagnosed on-site by visual inspection, palpation, percussion, etc., and compares it with the degradation diagnosis criteria for each customer. (See, for example, Non-Patent Documents 1 to 3.)

  As another conventional paint deterioration diagnosis method, a diagnostic object is photographed as grayscale image information, and a deformed portion indicating rust or peeling is detected from the grayscale image information and the number of deformed portions, each area, and The ratio of the entire area of the deformed portion to the entire image is obtained as deformed portion data, and the gray image information is divided into small areas at predetermined intervals, and consists of the average value and variance of the gray values included in the small area. A method has been proposed in which statistics are obtained and the degree of progress of deterioration of a diagnosis target is determined from these statistics and the deformed portion data (see, for example, Patent Document 1).

JP-A-6-116914

Railway Technical Research Institute, “Paint Degradation State and Keren Degree Sample Book”, Railway Technical Research Institute, Showa 64 Nippon Telegraph and Telephone Public Corporation, Architecture Department, “Decoration Photo Sample of Iron Painting”, Communications Architecture Laboratory, January 1984 Edited by Japan Road Association, “Standard Photo Book of Degree of Film Degradation”, Japan Road Association, June 1990

  The method of diagnosing paint deterioration by visual inspection, palpation, and percussion on site is dependent on the experience of the operator, and there is a problem that the results of diagnosis vary due to individual differences. In addition, when there are many structures to be diagnosed or when the scale of the structures is large, there is a problem that the amount of diagnosis work becomes enormous.

  In addition, in the method of judging the degree of deterioration progression from grayscale image information, deterioration diagnosis is performed by image processing techniques, so there is no variation in diagnosis results due to individual differences, and deterioration diagnosis work by on-site visual inspection, palpation, and percussion Can save labor and time, but there are many non-degraded parts on the painted surface of the steel material that are rust-colored due to adhesion of rust juice and are judged as degraded parts. There is a problem of end up. The rust juice is a substance produced by corrosion of steel and metal fittings attached to the steel, carried by rainwater, etc., and fixed to the painted surface, and has the same brown or brown color as the corroded part. The adhesion of rust juice significantly stains the painted surface and is not good on the landscape, but the rust juice itself does not deteriorate the performance of the structure.

  In order to advantageously solve the above-described problems, the paint deterioration diagnosis method of the present invention uses a full-color bitmap image obtained by photographing the surface of a steel structure such as a steel tower or steel bridge that has been painted as an input image. A step of constructing a deteriorated part extraction filter that converts only the deteriorated part of the painted surface in the input image into a transparent image, and transmission indicating the deteriorated part in the converted image obtained by applying the deteriorated part extraction filter. The process of generating a degradation profile by calculating the number of degraded parts and the area of each degraded part and generating the degradation profile, and expressing the amount of degraded parts generated by the number of circles approximating the degraded part. The amount is expressed by the total area ratio of the circle approximated to the deteriorated part and the area increase rate of the total area ratio for each growth stage, and the generation and growth are ordered according to the preset generation amount and growth amount of the deteriorated part. Repeatedly, the appearance of paint surface deterioration・ Simulate growth, create various virtual painted surface degradation images, calculate the number of degraded parts and the area of each degraded part for the created virtual painted surface degraded images, and Create a virtual painted surface degradation image for evaluation that approximates the film degradation surface as much as possible, create a degradation profile for evaluation from the virtual painted surface degradation image for evaluation and the customer's degradation diagnostic criteria, and evaluate the degradation profile for evaluation and A process of accumulating virtual painted surface degradation image data to form a database, a degradation profile for evaluation corresponding to an input image to be diagnosed from the database is extracted from the database, and a degradation profile obtained from the input image to be degraded diagnosis The degree of deterioration of the painted surface is diagnosed by a deterioration degree evaluation process for determining the degree of deterioration by comparison.

  The process of constructing the deteriorated part extraction filter is composed of a three-stage process including a prototype identification process, an initial filter generation process, and an interactive identification process.

  In the prototype identification process, the input image group is classified into a plurality of categories based on the shooting conditions such as the shooting target and shooting situation, and only a plurality of typical input images and degraded portions of the input image are clearly shown for each category. Creates training data consisting of image pairs with the ideal output image and generates a sub-optimized spatial filter to extract only the degraded part from the input image using the genetic algorithm method for each category. The generated spatial filter group is held as a prototype group of the deteriorated part extraction filter.

  In the initial filter generation process, a plurality of deteriorated part extraction filter prototypes of each category are selected from the deteriorated part extraction filter prototype group, and an input image is applied to the selected deteriorated part extraction filter prototype. A plurality of converted images are selected from the image group according to the result determined by the diagnostician to be relatively appropriate, and the deteriorated portion extraction filter from which the selected image is derived is held as the initial filter group.

  The interactive identification process includes a first step of selecting a converted image according to a result determined by a diagnostician to be relatively appropriate from a converted image group obtained by applying an input image to an initial filter group; A second step of generating a next generation degraded part extraction filter group having a degraded part extraction filter group derived from the selected image using a genetic algorithm technique, and a next generation degraded part extraction filter group It consists of three steps, the third step of selecting a converted image according to the result judged by the diagnostician to be relatively appropriate from the converted image group obtained by applying the input image. A sub-optimal deteriorated part extraction filter is created by repeating the second to third steps by the method.

  The evaluation deterioration profile is calculated from the number of deteriorated portions and the area of each deteriorated portion, the total area ratio of the deteriorated portions relative to the determination area, and a value obtained by making the number of deteriorated portions dimensionless by the number of pixels of the determination area. A three-dimensional evaluation reference space with three parameters as the axis and the maximum value of the area of each deteriorated part or a value representing the variation in the size of the deteriorated part in the same image as a parameter It is.

  In addition, the paint deterioration diagnosis apparatus of the present invention receives, as an input, an input image of a full color bitmap image obtained by photographing the surface of a steel structure such as a steel tower or steel bridge to which paint has been applied, and paint in the input image Deterioration part extraction filter construction means for constructing a deterioration part extraction filter that converts and outputs an image that transmits only the deterioration part of the surface, and a transmission part that indicates the deterioration part in the converted image obtained by applying the deterioration part extraction filter A deterioration profile generation means for generating a deterioration profile by calculating the number of deterioration portions and the area of each deterioration portion, and the generation amount of the deterioration portion is represented by the number of occurrences of a circle that approximates the deterioration portion. Is expressed by the total area ratio of the circle approximated to the deteriorated part and the area increase rate of the total area ratio for each growth stage. systematically Again, a virtual painted surface deterioration image creation means for simulating the occurrence / growth of painted surface degradation and creating various virtual painted surface degradation images, and a degradation part for the created virtual painted surface degradation image group The virtual painted surface deterioration image for evaluation that approximates the actual coated film deteriorated surface as much as possible is created and the evaluation virtual painted surface deteriorated image and the customer's deterioration diagnosis standard are calculated. Corresponding to the degradation image creation means for creating the degradation profile for evaluation, the database formed by accumulating the degradation profile for evaluation and the virtual painted surface degradation image for evaluation, and the input image for degradation diagnosis from the database Degradation rating for determining the degree of degradation by taking out the degradation profile for evaluation and comparing it with the degradation profile obtained from the input image of the degradation diagnosis target It is intended and means.

  The degraded portion extraction filter construction means includes means for performing a three-stage process including a means for performing a prototype identification process, a means for performing an initial filter generation process, and a means for performing an interactive identification process.

  The means for performing the prototype identification process classifies the input image group into a plurality of categories based on shooting conditions such as a shooting target and a shooting situation, and a plurality of typical input images and degraded portions of the input images for each category. In order to extract only the degraded part from the input image by using the training teacher data creation means that creates the training teacher data consisting of the image pair with the ideal output image that clearly indicates only and the genetic algorithm method for each category And a prototype filter construction means for generating a sub-optimized spatial filter and holding the generated spatial filter group as a prototype group of the deteriorated part extraction filter.

  The means for performing the initial filter generation process includes: a deteriorated portion extraction filter selecting means for selecting several deteriorated portion extraction filter prototypes for each category from a prototype group of deteriorated portion extraction filters; and a selected deteriorated portion extraction filter prototype. Receiving means for receiving information of a result determined by the diagnostician to be relatively appropriate from the converted image group obtained by applying the input image to the image; and image selecting means for selecting the converted image according to the received information; And an initial filter generation means for holding a deteriorated portion extraction filter from which the selected image is derived as an initial filter group.

  The means for performing the interactive identification process is a result that the diagnostician determines that the diagnostic image is relatively appropriate from the converted image group obtained by applying the input image to the initial filter group or the next-generation deteriorated part extraction filter group. A reception means for receiving information on the image, an image selection means for selecting a converted image in accordance with the received information, and a next generation deterioration part extraction filter group having a deterioration part extraction filter group derived from the selected image as a parent. And updating means for generating using an algorithm technique.

  According to the paint deterioration diagnosis method and apparatus of the present invention, since deterioration diagnosis is performed by an image processing method, there is almost no variation in diagnosis results due to individual differences, compared with deterioration diagnosis work by visual inspection, palpation, and percussion on site. Labor and time-consuming work can be saved.

  In addition, a deteriorated part extraction filter that transmits only the deteriorated part of the painted surface in the image, a prototype identification process using a genetic algorithm method, an initial filter generation process, and an interactive type using an interactive evolutionary calculation method By generating by a three-stage process including the identification process, it is possible to determine a portion that is not a deteriorated portion while exhibiting a rust color such as rust juice.

It is a figure which shows the flow of the coating deterioration diagnostic method of this invention. It is a figure which shows an example of an input image and a conversion image. It is a figure which shows the flow of a prototype identification process. It is a figure which shows an example of teacher data (a pair of a sample image-ideal output image). It is a figure which shows an example of the data structure of a spatial filter group. It is a figure which shows the flow of an interactive identification process. It is a figure which shows an example of a virtual painted surface degradation image. It is a figure which shows an example of the virtual painting surface degradation image for evaluation. It is a figure which shows an example of the three-dimensional evaluation reference | standard space | space at the time of deleting a minute degradation part. It is a figure which shows an example of the three-dimensional evaluation reference | standard space which considered the dispersion | variation in the magnitude | size of a degradation part. It is a figure which shows the component of the apparatus which performs a paint deterioration diagnosis. It is a figure which shows the component of a degradation part extraction filter construction means. It is a figure which shows an example of the filtering result of the prototype identification process of an Example. It is a figure which shows an example of the filtering result of the interactive identification process of an Example. It is a figure which shows an example of the uniform type virtual coating surface degradation image of an Example. It is a figure which shows an example of the half type | mold and center type virtual paint surface degradation image of an Example. It is a figure which shows the example of an input image of deterioration diagnosis object. It is a figure which shows an example of the result of having plotted the parameter value in the 2nd evaluation reference | standard space. It is a figure which shows the flow of the coating deterioration diagnostic method after database formation.

  Embodiments of the present invention will be described below in detail with reference to the accompanying drawings.

  FIG. 1 is a flowchart of the paint deterioration diagnosis method of the present invention. An image obtained by photographing the surface of a steel structure such as a steel tower or a steel bridge that has been painted is photographed by a digital camera or the like and captured by a computer. The photographed image is used as the input image 1 as a full color bitmap image. The input image 1 is represented by a three-dimensional integer value vector (r, g, b) of RGB tristimulus values, 0 ≦ r, g, b ≦ 255 for each pixel.

  Using the input image 1, construction 2 of the deteriorated part extraction filter is performed. The deteriorated portion extraction filter is an input / output system that converts and outputs an image through which only the deteriorated portion of the painted surface of the input image 1 is transmitted. FIG. 2 shows an example of the input image and the converted image.

The configuration of the spatial filter used in the present invention will be described. The spatial filter converts tristimulus values (r, g, b), (0 ≦ r, g, b ≦ 255, r, g, b are integers) of each pixel of the image I into a one-dimensional transmission value t, ( 0 ≦ t ≦ 255, t is a function t = F ₁ (r, g, b). The deteriorated part extraction filter is a spatial filter that transmits the deteriorated part in the image and does not transmit the other part, and gives a large transmission value to the pixel of the deteriorated part, and a low transmission value to the other part. give.

First, the tristimulus values of (r, g, b) are converted into points (L, u, v) on the L ^* u ^* v ^* color space through Equations 1 and 2. Next, the pixels of the image I are clustered by the k-means method with respect to the distance ((L ² + u ² + v ² ) ^1/2 ) in the L ^* u ^* v ^* color space. Here, b _I (i, j) = (L _b (i, j), u _b (i, j), v _b (i , J)).

Next, the luminance component Y (r, g, b) is separated from the tristimulus value (r, g, b) of each pixel through Equation 3. A grayscale image according to the luminance component and I _Y. In addition, the luminance of the pixel at the coordinates (i, j) of the image I _Y is represented by Y _{(i, j)} , and the spatial second derivative (smoothness of luminance change) D _Y (i, j) of the luminance component is expressed by Equation 4. Give in.

In the spatial filter F _I (r, g, b), first, b _I (i, j) = (L _b (i, j), u _b (i, j), v at all coordinates (i, j). _b (i, j)) is subjected to affine transformation according to Equation 5, and through Equation 6, n reference points (L _k , u _k , v _k ), k = 1,. Find the shortest distance d (i, j). Next, a convex combination v between the shortest distance d (i, j) and the spatial second derivative of the luminance component is obtained by Expression 7. Then, the transmission value t is determined by Expression 8.

In Equation _{5, c i, i = 1} , ···, 9 is a real number becomes -1 ≦ _{c i} ≦ 1, in Equation 7, lambda is 0 ≦ λ ≦ 1 becomes real numbers, the equation 8, [ ] Is a Gaussian symbol, K is a constant for controlling the transmittance, and is a real number satisfying K> 0.

  In the deteriorated part extraction filter construction step 2, identification of the deteriorated part extraction filter includes a three-stage process including a prototype identification process 3, an initial filter generation process 4, and an interactive identification process 5. In the prototype identification process 3 in the first stage, a spatial filter serving as a prototype of the deteriorated part extraction filter is created by using learning teacher data (a set of typical input image-ideal output image pairs) created as teacher data. Identify by genetic algorithm. In the initial stage filter generation process 4 of the second stage, a deteriorated part extraction filter group suitable for the input image 1 of the deterioration diagnosis target is selected from the prototype group of the deteriorated part extraction filter generated in the prototype identification process 3 of the first stage, Create an initial filter group. In the third stage interactive identification process 5, a sub-optimal deteriorated part extraction filter is identified using a method of interactive evolution calculation based on the initial filter group generated in the second stage initial filter generation process 4. I will do it.

  The identification of the deteriorated part extraction filter is not necessarily guaranteed that a sub-optimal solution is obtained as a result under the given conditions only by the first-stage prototype identification process 3. Therefore, the interactive identification process 5 is performed in the third stage to identify the deteriorated part extraction filter by reflecting the subjective evaluation value of the diagnostician, thereby obtaining a sub-optimal solution.

  FIG. 3 is a flowchart of the prototype identification process 3. In the procedure of the prototype identification process 3 in the first stage, first, the input image group 1 is classified into a plurality of categories based on photographing conditions such as a photographing target and a photographing situation. Next, for each category, a plurality of typical input images 1 are extracted, and using the extracted input images 1 as sample images, an ideal output image in which only the degraded portion is transmitted is created, as shown in FIG. Teacher data (sample image-ideal output image pair) is created. The ideal output image is a binarized image in which the gradation level of the degraded portion is 255 and the gradation level of the non-degraded portion is 0. These teacher data groups (a set of sample image-ideal output image pairs) are used as learning teacher data, and a space that is semi-optimized to extract only the degraded part using the genetic algorithm method for each category. A filter is generated, and the generated spatial filter group is held as a prototype group of the deteriorated part extraction filter.

  The genetic algorithm used in the first stage prototype identification process 3 will be outlined. A genetic algorithm is an algorithm that simulates the evolution process of a living organism and is a method for obtaining a suboptimal solution to a problem. Neurocomputers target organisms as individuals, whereas genetic algorithms differ greatly in that they target multiple organisms for generations.

A general processing procedure of the genetic algorithm will be described.
(1) Initialization Coat the target problem in the form of a gene, generate N individuals with different chromosomes, and make it the initial generation. A gene is a basic component that defines the character of an individual. If a chromosome is a bit string, it is represented by 0 or 1.
(2) Selection The fitness of each individual is calculated, and the individual is selected according to a certain rule based on the fitness. In general, individuals with high fitness grow and individuals with low fitness are culled.
(3) Crossover Crossover is performed according to the set crossing probability and crossover method, and a new individual is generated.
(4) Mutation Mutation is performed according to the set mutation probability and mutation method to generate a new individual.
(5) Termination determination If the termination condition is satisfied, the best individual obtained at that time is set as the suboptimal solution of the problem.
If the end condition is not satisfied, (2) return to selection.

In the present invention, the deteriorated part extraction filter identified by using the genetic algorithm technique includes nine parameters c _i , i = 1, 2,..., 9 and (L _k , u _k , v _k ). , K = 1,..., N and λ and K are combined to form 3n + 11 parameters, and each (3n + 11) parameter is represented by an 8-bit gray code. That is, the chromosome to be coated is composed of 0 and 1 character strings of 8 bits × (3n + 11) parameter = 8 × (3n + 111) bits. In addition, it is assumed that the chromosomes of the N initial individuals are all randomly generated, and there is no increase or decrease in the number of individuals at the time of generation update.

  The fitness of the individual s is determined by the fitness function a (s) of Equation 9. The fitness of each individual is calculated, and individuals are selected according to a certain rule based on the fitness. Individuals with high fitness are selected as individuals that can leave offspring in the next generation, and individuals with low fitness are deceived.

In Equation 9, a sample image is I in the teacher data image group, C is a set of all the pixels that are degraded in the ideal output image, and D is a set of all other pixels. Further, the tristimulus values at each pixel p are assumed to be (r (p), g (p), b (p)). That is, the individual s that gives a larger transmission value F _I (r, g, b) in the degraded portion and a smaller transmission value F _I (r, g, b) in the non-degraded portion has a higher fitness. .

  In the genetic algorithm, individuals of the initial generation are randomly created, genetic operations (selection, crossover, and mutation) are performed on the individuals of the initial generation to create the next generation, and gene manipulation is performed on the individuals of the generated generation. The optimum individual is searched for by repeating the operation. This genetic manipulation is repeated after a predetermined number of generation updates.

  When the degraded part extraction filter is identified using the genetic algorithm method for each category, at the end of the generation update, as shown in the data structure of the spatial filter group in FIG. Several individuals 11 occupy the population. The generated spatial filter group is held as a prototype of the deteriorated part extraction filter.

  Next, the procedure of the initial stage initial filter generation process 4 will be described. First, the input image 1 to be deteriorated is designated. Several prototypes of deteriorated part extraction filters of each category are selected from the prototype group of deteriorated part extraction filters generated in the first stage prototype identification process 3. A group of converted images obtained by applying the input image 1 to the prototype of the selected degraded portion extraction filter is presented together with the input image 1. A plurality of converted images are selected according to a result that the diagnostician determines to be relatively appropriate from the presented converted image group. Next, from the category to which the deteriorated part extraction filter from which the selected image is derived belongs, the deteriorated part extraction filter from which the selected image is derived and the respective variants are held as an initial filter group.

  FIG. 6 is a flow of the interactive identification process 5. In the procedure of the interactive identification process 5 in the third stage, first, as a first step, the input image 1 to be diagnosed for deterioration is applied to the initial filter group held in the initial filter generation process 4 in the second stage. The obtained converted image group is presented to the diagnostician together with the input image 1, and the converted image is selected according to the result that the diagnostician determines to be relatively appropriate from the displayed converted image group. As a second step, a next generation deteriorated part extraction filter group having a deteriorated part extraction filter group derived from the selected image as a parent is generated using a genetic algorithm technique. As a third step, the converted image group obtained by applying the input image 1 to the next-generation degraded portion extraction filter group is presented to the diagnostician together with the input image 1, and the diagnostician selects the converted image group from the presented converted image group. A converted image is selected according to a result determined to be relatively appropriate. A sub-optimal deteriorated part extraction filter is created by repeating the second to third steps by the method of interactive evolutionary calculation.

The update / determination of the deteriorated part extraction filter is performed by the following procedure using the method of interactive evolution calculation with the initial filter group as the initial individual group.
(1) From the group of presented degradation level extraction filters, a degradation level extraction filter that the diagnostician determines to be relatively appropriate is selected.
(2) The selected deterioration degree extraction filter is set to “an individual group that can leave offspring in the next generation”, the generation is updated, and the updated individual group is presented. However, the individuals selected in (1) are left as they are in the next generation, and the remaining individuals (the number of individuals excluding the number of individuals selected from the initial population) are “crossed” in the same way as in the prototype identification process 3. Generate by adding operations. In other words, the generation function is updated by replacing the fitness function in the prototype identification process 3 with the subjective evaluation value of the diagnostician.
(3) If there is a sufficiently satisfactory degradation part extraction filter for the degradation diagnosis target input image 1 in the re-presented population, it is adopted and terminated. Otherwise, return to (1) and repeat the series of steps.

  In the interactive identification process 5, the judgment as to how well the deteriorated portion in the input image is extracted is made visually by the diagnostician, and other processing is performed by the computer. Although it is difficult to make a person's judgment on a computer, it is classified into the same category as the input image 1 to be diagnosed for deterioration by accumulating the judgment results of the diagnostician as a database. It is also possible to automatically perform the interactive identification process 5 on the computer by extracting similar image determination results from the database and determining how well the degraded portion is extracted on the computer. is there.

  In the degradation profile generation step 7, the conversion image 6 obtained by applying the input image 1 to be subjected to degradation diagnosis to the degradation portion extraction filter quasi-optimized in the degradation portion extraction filter construction step 2 is converted into a converted image. The transmission part indicating the deterioration part is detected, the number of deterioration parts and the area of each deterioration part are calculated, and a deterioration profile is generated.

  Next, in order to evaluate the degree of deterioration of the converted image 6 obtained by applying the deteriorated part extraction filter, a virtual painted surface deterioration image is created 8. Since the coating deterioration diagnosis criteria differ from customer to customer, it is not possible to create certain criteria, and it is necessary to have a system that can handle the diagnostic criteria for each customer. As an example of paint deterioration diagnosis criteria, there is a standard of judgment method P of “Coating film deterioration state and kelen degree sample book” edited by Railway Research Institute shown in Table 1, crack, peeling, rusting, etc. Among the deteriorations, there are 6 stages of deterioration depending mainly on the state of rust generation.

  Degradation processes that can be perceived visually can be roughly classified into two types: generation and growth. The degree of progress of deterioration also changes depending on how the generation and growth rates change and how they relate to each other. Therefore, a virtual painted surface deterioration image that simulates the deterioration process focusing on the generation / growth process of the number and area of the deteriorated parts is created, and the criterion is quantified.

Since the deterioration process can be roughly divided into two types, generation and growth, the generation amount of the deterioration portion is represented by the number of occurrences of circles (n) approximating the deterioration portion, and the growth amount of the deterioration portion is expressed as follows: It is represented by the total area ratio (AAR _i ) of the circle approximated to the deteriorated part and the area increase rate (β) of the total area ratio at each growth stage, and the generation and growth are repeated regularly every time it passes through the stage (L). Then, a virtual painted surface deterioration image simulating the deterioration process is created. The virtual painted surface deterioration image is created on the assumption that the deteriorated portion grows according to Equation 10 representing the total area ratio of the deteriorated portion at the i-th stage.

In addition, the virtual painted surface deterioration image shows the surface when the distribution of the occurrence position of the deteriorated part is uniformly distributed (uniform type), when one side is biased (half type), Three types are created when concentrated in the center (center type). FIG. 7 shows an example of the virtual painted surface deterioration image created. The virtual painted surface degradation image shown in the figure is the number of occurrences n = {5, 25, 200} (pieces), the total area ratio AAR _i = 0.1%, and the area increase rate β = 2. Among the images created with an initial value of 0, a virtual painted surface deteriorated image at the growth stage i = fourth stage (AAR _i = 0.8%) and i = 8th stage (AAR _i = 12.8%) is shown. ing.

  Since the input image 1 to be diagnosed for deterioration has different shooting conditions such as the shooting situation, shooting site, and shooting range for each image, the conditions for occurrence and growth of deterioration are changed so as to correspond to the shooting conditions, and the virtual painted surface A plurality of degradation images are created, and a virtual painted surface degradation image that matches the conditions of the degradation diagnosis target image is created.

  In step 9 for creating the degradation profile for evaluation, first, for the virtual painted surface degradation image group created in the virtual painted surface degradation image creating step 8, the number of degraded portions and the area of each degraded portion are calculated and evaluated. Create a virtual painted surface degradation image. The virtual painted surface deterioration image for evaluation is an image obtained by approximating the virtual painted surface deteriorated image as much as possible to the actual coated film deteriorated surface. For example, as shown in FIG. 8, an image of a steel structure that has been painted. An image in which a virtual painted surface deterioration image is arranged is used. Next, the virtual painted surface deterioration image for evaluation is classified according to the degree of deterioration based on the customer's deterioration diagnosis criteria, and the total area ratio (AAR) of the deteriorated portion relative to the determination area (HA) and the number of the deteriorated portions are determined. A non-dimensional value (N / HA) of the number of pixels, a maximum area value (AMR) of each deteriorated portion, and a value representing variation in the size of the deteriorated portion in the same image, A degradation profile for evaluation composed of a three-dimensional evaluation reference space with three parameters as axes is created. The created degradation profile for evaluation is stored together with the virtual painted surface degradation image for evaluation to form a database.

  When a diagnostician visually evaluates the degree of deterioration, a minute deterioration portion with respect to the determination area is ignored. In order to reflect this in the degradation profile for evaluation, AAR, N / HA, and AMR are obtained after erasing the degraded portion displayed by one pixel, which is the minimum unit in the image. Parameters after erasing the minute deteriorated part are represented by FAAR, FN / HA, and FAMR, respectively. An example of the created three-dimensional evaluation reference space is shown in FIG.

In addition, when there is almost no large deteriorated part and mainly the minute deteriorated part spreads over the entire surface, it is necessary to consider the minute deteriorated part as an evaluation criterion. In order to reflect this in the degradation profile for evaluation, the size of each degraded portion is multiplied by a weighting function, and a value representing the variation in the size of the degraded portion in the same image is set as a parameter. A value (Σ {log (SA _i )} ² ) is set by squaring the size of each deteriorated portion using a value representing the variation in size of the deteriorated portion as a parameter. SA _i (i = 1,..., N) indicates the area of each deteriorated portion. An example of the created three-dimensional evaluation reference space is shown in FIG.

The procedure of the degradation degree evaluation step 10 will be described. From the deterioration profile obtained from the input image 1, the total area ratio (AAR, FAAR) of the deteriorated portion relative to the determination area (HA) and a value obtained by making the number of deteriorated portions dimensionless by the number of pixels of the determination area (N / HA) , FN / HA) and a maximum value (AMR, FAMR) of the area of each deteriorated portion or a value (Σ {log (SA _i )} ² ) representing variation in size of the deteriorated portion in the same image. Then, the degradation profile for evaluation obtained in the step 9 for creating the degradation profile for evaluation is plotted in the three-dimensional evaluation standard space. The plotted point is represented by one point in the evaluation reference space, and a determination result corresponding to the point is output as a result of the deterioration degree.

  As a method for extracting the deterioration profile for evaluation corresponding to the input image 1 to be subjected to the deterioration diagnosis from the database storing the deterioration image for evaluation painted surface and the deterioration profile for evaluation, the virtual painted surface deterioration image is obtained by taking the imaging state, Since multiple images are created so as to correspond to the imaging conditions such as the imaging region and imaging range, it is selected by selecting the virtual painted surface degradation image for evaluation corresponding to the imaging condition (category classification) of the degradation diagnosis target image. The evaluation deterioration profile corresponding to the evaluation virtual painted surface deterioration image can be taken out.

  As a component of an apparatus for performing such paint deterioration diagnosis, as equipment for accumulating a deterioration profile, as shown in FIG. 11, an image capturing means 13 and the captured image group are images of a full color bitmap image. Image conversion means 14 for converting to data, degradation part extraction filter construction means 15 for constructing a degradation part extraction filter for converting and outputting an image through which only the degradation part of the painted surface of the input image 1 is transmitted, and construction of a degradation part extraction filter The transmission part which shows the degradation part of 6 in the conversion image obtained by applying the degradation part extraction filter quasi-optimized by the means 15 is detected, and the degradation profile is calculated by calculating the number of degradation parts and the area of each degradation part. A degradation profile generation means 16 for generating, and a storage means 17 for accumulating degradation profile and degradation portion extraction filter data are provided. The computer system used.

  As shown in FIG. 12, the deteriorated part extraction filter construction means 15 is a three-stage process comprising means 20 for performing a prototype identification process, means 25 for performing an initial filter generation process, and means 29 for performing an interactive identification process. There is a means to perform.

  The means 20 for performing the prototype identification process classifies the input image group 1 into a plurality of categories based on photographing conditions such as a photographing object and a photographing situation, and for each category, a pair of a plurality of sample images and ideal output images. Using a learning teacher data creation means 18 for creating a teacher data group and a genetic algorithm method for each category, to generate a sub-optimized spatial filter for extracting only the degraded portion from the input image 1 And a prototype filter construction means 19 for holding the generated spatial filter group as a prototype group of the deteriorated part extraction filter.

  The means 25 for performing the initial filter generation process includes a deteriorated part extraction filter selecting means 21 for selecting several prototypes of deteriorated part extraction filters for each category from a prototype group of deteriorated part extraction filters, and a selected deteriorated part extraction filter. Receiving means 22 for receiving information of a result determined by the diagnostician to be relatively appropriate from the converted image group obtained by applying the input image 1 to the prototype of the image, and selecting the converted image according to the received information Initial filter generation that retains the image selection means 23 and the deteriorated portion extraction filter from which the selected image is derived and the respective subspecies as an initial filter group from the category to which the deteriorated portion extraction filter from which the selected image is derived belongs Means 24.

  The means 29 for performing the interactive identification process accepts information on a result determined by the diagnostician to be relatively appropriate from the converted image group obtained by applying the input image 1 to the deteriorated part extraction filter group. Means 26, an image selecting means 27 for selecting a converted image according to the received information, and a next generation degraded portion extraction filter group having a degraded portion extraction filter group from which the selected image is derived as a genetic algorithm. The update means 28 which produces | generates using is provided.

  A procedure for accumulating a deterioration profile using a computer system having such a function will be described. An image obtained by photographing the surface of a steel structure such as a steel tower or steel bridge that has been painted is captured by the image capturing means 13 and the captured image group is converted into a full-color bitmap image by the image converting means 14. And used as the input image 1.

  The deteriorated part extraction filter construction means 15 constructs a deteriorated part extraction filter that converts and outputs an image that transmits only the deteriorated part of the painted surface of the input image 1. First, the means 20 for performing the prototype identification process categorizes the input image group 1 into a plurality of categories by the learning teacher data creation means 18 based on the photographing conditions such as the photographing object and the photographing situation, and a plurality of them are classified for each category. A teacher data image group composed of a pair of the sample image and the ideal output image is created. Next, the prototype filter construction means 19 generates a sub-optimized spatial filter for extracting only the deteriorated portion from the input image 1 using a genetic algorithm method for each category, and the generated spatial filter The group is held as a prototype group of the deteriorated part extraction filter.

  Next, the means 25 for performing the initial filter generation process selects several prototypes of the deteriorated portion extraction filters for each category by the deteriorated portion extraction filter selecting means 21. Information on the result determined by the diagnostician to be relatively appropriate from the converted image group obtained by applying the input image 1 to the prototype of the selected degraded portion extraction filter is received by the receiving means 22 and received. A converted image is selected by the image selection means 23 according to the information. From the category to which the deteriorated part extraction filter from which the selected image is derived belongs, the deteriorated part extraction filter from which the selected image is derived and the respective variants are held as an initial filter group by the initial filter generation means 24.

  Next, in the means 29 for performing the interactive identification process, first, as a first step, the diagnostician is relatively appropriate from the converted image group obtained by applying the input image 1 to the initial filter group. Information of the determined result is received by the receiving unit 26, and a converted image is selected by the image selecting unit 27 according to the received information. As a second step, a next-generation degraded part extraction filter group having a degraded part extraction filter group from which the selected image is derived as a parent is generated by the update means 28 using a genetic algorithm technique. As a third step, the reception unit 26 receives information on a result determined by the diagnostician to be relatively appropriate from the converted image group obtained by applying the input image 1 to the next-generation deteriorated part extraction filter group. The converted image is selected by the image selection means 27 in accordance with the received information. A sub-optimal deteriorated part extraction filter is created by repeating the second to third steps by the method of interactive evolutionary calculation.

  For the converted image 6 obtained by applying the deteriorated part extraction filter quasi-optimized by the deteriorated part extraction filter construction means 15, the deterioration profile generating means 16 detects a transmission part indicating the deteriorated part in the converted image. The number of deteriorated parts and the area of each deteriorated part are calculated to generate a deterioration profile, and the deterioration profile and the data of the deteriorated part extraction filter are stored in the storage means 17.

  Note that in the means 29 for performing the interactive identification process, it is necessary for the diagnostician to visually determine how well the deteriorated portion in the input image is extracted. The judgment result of the diagnostician is accumulated in the storage means 17 together with the data of the above, and the database is formed, so that the judgment result of the similar image classified in the same category as the input image 1 to be deteriorated is extracted from the database. Then, by determining on the computer how well the deteriorated portion is extracted, the means 29 for performing the interactive identification process can be automatically performed on the computer.

  That is, when the database is formed, as shown in FIG. 19, in the case of generating a degradation profile of the newly captured input image 1, the captured input image 1 is based on the imaging conditions such as the imaging target and the imaging situation. Degradation part extraction filters of similar images that are classified into a plurality of categories and are classified in the same category as the input image 1 subject to deterioration diagnosis are extracted from the database, and the deterioration part extraction filter extracted to the input image 1 subject to deterioration diagnosis is extracted. By applying this, the converted image 6 transmitted through the deteriorated part is output, and the deterioration profile can be generated by calculating the number of deteriorated parts and the area of each deteriorated part.

  Moreover, as a component of the apparatus for performing the paint deterioration diagnosis, as equipment for accumulating the deterioration profile for evaluation, as shown in FIG. 11, the amount of generation of the deteriorated portion is represented by the number of generated circles approximated to the deteriorated portion. The growth amount of the deteriorated part is represented by the total area ratio of the circle approximated to the deteriorated part and the area increase rate of the total area ratio for each growth stage, and according to the preset generation amount and growth amount of the deteriorated part A virtual painted surface deterioration image creating means 30 for simulating the occurrence and growth of paint surface deterioration by repeating generation and growth regularly and creating various virtual paint surface deterioration images, and the created virtual paint surface deterioration image For each group, the number of deteriorated parts and the area of each deteriorated part are calculated, a virtual painted surface deterioration image for evaluation is created, and the evaluation deterioration is determined from the evaluation virtual painted surface deteriorated image and the customer's deterioration diagnosis standard. Review profile creation And use the degradation profile creating unit 31, a computer system that includes a storage unit 32 for storing the data of the evaluation degradation profile and virtual paintwork deteriorated image for evaluation used.

  A procedure for accumulating the degradation profile for evaluation using a computer system having such a function will be described. First, the degradation model creation means 30 simulates the occurrence / growth of painted surface degradation and creates various virtual painted surface degradation images. Next, the evaluation deterioration profile creating means 31 calculates the number of deteriorated portions and the area of each deteriorated portion for the created virtual painted surface deteriorated image group, and also creates an evaluation virtual painted surface deteriorated image. . The virtual painted surface deterioration image for evaluation is classified according to the degree of deterioration based on the customer deterioration diagnosis criteria, and the total area ratio (AAR) of the deteriorated portion with respect to the determination area (HA) and the number of the deteriorated portions are pixels of the determination area. A value (N / HA) made dimensionless by number, a maximum value (AMR) of the area of each deteriorated portion, and a value representing variation in size of the deteriorated portion in the same image are calculated, A deterioration profile for evaluation composed of a three-dimensional evaluation reference space with a parameter as an axis is created. The created degradation profile for evaluation is stored in the storage means 32 together with the virtual painted surface degradation image for evaluation to form a database.

  Furthermore, as a component of the apparatus for performing the paint deterioration diagnosis, as equipment for performing the deterioration degree evaluation, as shown in FIG. 11, a database storing evaluation virtual paint surface deterioration images and evaluation deterioration profiles is stored. An evaluation deterioration profile corresponding to the input image 1 of the deterioration diagnosis target is extracted from the inside, and a parameter calculated from the deterioration profile of the input image 1 of the deterioration diagnosis target is compared with the deterioration profile for evaluation, and the input to be the deterioration diagnosis target A computer system provided with deterioration degree evaluation means 33 for determining the degree of deterioration of image 1 is used.

A procedure for performing the deterioration degree evaluation using a computer system having such a function will be described. First, the deterioration degree evaluation means 33 determines the total area ratio (AAR, FAAR) of the deteriorated portion relative to the determination area (HA) and the number of the deteriorated portions from the deterioration profile obtained from the input image 1 to be deteriorated. Values (N / HA, FN / HA) made dimensionless with the number of pixels and the maximum value (AMR, FAMR) of the area of each deteriorated portion or a value representing variation in the size of the deteriorated portion in the same image (Σ {log (SA _i )} ² ) is calculated. Next, the degradation profile for evaluation corresponding to the input image 1 to be diagnosed for degradation is extracted from the database storing the degradation image for evaluation painted surface and the degradation profile for evaluation, and parameters calculated from the degradation profile are obtained. By comparing, the degree of deterioration of the deterioration diagnosis target image is determined.

  Next, an example explains the present invention. An image obtained by photographing the surface of a steel structure such as a steel tower or a steel bridge that has been painted is photographed by a digital camera or the like and captured by a computer. The photographed image is used as the input image 1 as a full color bitmap image.

  First, a deteriorated part extraction filter is constructed using the input image 1. In the deteriorated part extraction filter construction step 2, identification of the deteriorated part extraction filter includes a three-stage process including a prototype identification process 3, an initial filter generation process 4, and an interactive identification process 5.

  The first stage prototype identification process 3 is performed. First, the input image group 1 is classified into a plurality of categories based on shooting conditions such as shooting targets and shooting conditions. Next, for each category, a plurality of typical input images 1 are extracted, and using the extracted input images 1 as sample images, an ideal output image in which only the degraded portion is transmitted is created, as shown in FIG. Teacher data (sample image-ideal output image pair) is created. The ideal output image is a binarized image in which the gradation level of the degraded portion is 255 and the gradation level of the non-degraded portion is 0. These teacher data groups (a set of sample image-ideal output image pairs) are used as learning teacher data, and a space that is semi-optimized to extract only the degraded portion using a genetic algorithm method for each category. A filter is generated, and the generated spatial filter group is held as a prototype group of the deteriorated part extraction filter.

In this embodiment, the deteriorated part extraction filter identified by using the genetic algorithm method is composed of nine parameters c _i , i = 1, 2,..., 9 and (L _k , u _k , v _k ). , K = 1, 2, and 3 and λ and K are combined into 20 parameters. The total 20 parameters are each expressed by an 8-bit gray code. That is, the chromosome to be coated is composed of 0 and 1 character strings of 8 bits × 20 parameters = 160 bits. In this embodiment, it is assumed that the initial number of individuals is 100, and the number of individuals does not increase or decrease when the generation is updated. All 160-bit chromosomes of each initial individual are randomly generated.

The fitness of the individual s is determined by the fitness function a (s) of Equation 9. In Equation 9, a sample image is I in the teacher data image group, C is a set of all the pixels that are degraded in the ideal output image, and D is a set of all other pixels. Further, the tristimulus values at each pixel p are assumed to be (r (p), g (p), b (p)). That is, the individual s that gives a larger transmission value F _I (r, g, b) in the degraded portion and a smaller transmission value F _I (r, g, b) in the non-degraded portion has a higher fitness. .

  In this example, genetic operations were selected and crossed according to the following rules. “Select” selects an individual having a fitness degree of the top 25% in an individual group as an individual capable of leaving offspring in the next generation. “Crossover” is a multipoint crossover in which two individuals are selected randomly (allowing duplication) from the selected individual group, and two crossover points (that is, 160/8 × 2 = 40 points) are provided every 8 bits. , Generate two new individuals for the next generation. However, at the time of crossover, each gene locus (bit) was replaced with an allele at the establishment of 0.01 (mutation). The number of generation updates was set to 300.

  As a result of identifying the deteriorated part extraction filter by using the above-described genetic algorithm method for each category, as shown in the data structure of the spatial filter group in FIG. Excluding, several individuals 11 occupied the population. An example of the filtering result by the identified five typical deteriorated part extraction filters is shown in FIG. In FIG. 13, the image (1) indicates the input image 1, and the images (2) to (6) indicate the filtering results by the five deteriorated part extraction filters. These generated degraded part extraction filter groups are held as prototypes of the degraded part extraction filters.

  Next, a second stage initial filter generation process 4 is performed. First, the input image 1 to be deteriorated is designated. Several prototypes of deteriorated part extraction filters of each category are selected from the prototype group of deteriorated part extraction filters generated in the first stage prototype identification process 3. A group of converted images obtained by applying the input image 1 to the prototype of the selected degraded portion extraction filter is presented together with the input image 1. A plurality of converted images are selected according to a result that the diagnostician determines to be relatively appropriate from the presented converted image group. Next, from the category to which the deteriorated part extraction filter from which the selected image is derived belongs, the deteriorated part extraction filter from which the selected image is derived and the respective variants are held as an initial filter group.

  Next, a third stage interactive identification process 5 is performed. First, as a first step, the converted image group obtained by applying the input image 1 to be diagnosed for degradation to the initial filter group held in the initial stage filter generation process 4 in the second stage is sent to the diagnostician together with the input image 1. The converted image is selected according to a result determined by the diagnostician to be relatively appropriate from the displayed converted image group. As a second step, a next generation deteriorated part extraction filter group having a deteriorated part extraction filter group derived from the selected image as a parent is generated using a genetic algorithm technique. As a third step, the converted image group obtained by applying the input image 1 to the next-generation degraded portion extraction filter group is presented to the diagnostician together with the input image 1, and the diagnostician selects the converted image group from the presented converted image group. A converted image is selected according to a result determined to be relatively appropriate. A sub-optimal deteriorated part extraction filter is created by repeating the second to third steps by the method of interactive evolutionary calculation.

The update / determination of the deteriorated part extraction filter is performed by the following procedure using the method of interactive evolution calculation with the initial filter group as the initial individual group.
(1) From the group of presented degradation level extraction filters, a degradation level extraction filter that the diagnostician determines to be relatively appropriate is selected.
(2) The selected deterioration degree extraction filter is set to “an individual group that can leave offspring in the next generation”, the generation is updated, and the updated individual group is presented. However, the individuals selected in (1) are left as they are in the next generation, and the remaining individuals (number of individuals excluding the number of selected individuals from the initial population) are subjected to “crossover” similar to the prototype identification process 3. Generate by adding operations. “Crossover” is a multipoint crossover in which two individuals are selected randomly (allowing duplication) from the selected individual group, and two crossover points (that is, 160/8 × 2 = 40 points) are provided every 8 bits. , Generate two new individuals for the next generation. However, at the time of crossover, each gene locus (bit) was replaced with an allele at the establishment of 0.01 (mutation). In other words, the generation function is updated by replacing the fitness function in the prototype identification process 3 with the subjective evaluation value of the diagnostician.
(3) If there is a sufficiently satisfactory degradation part extraction filter for the degradation diagnosis target input image 1 in the re-presented population, it is adopted and terminated. Otherwise, return to (1) and repeat the series of steps.

  An execution example of the interactive identification process 5 is shown in FIG. The example shown in FIG. 14 shows an image obtained by applying nine spatial filters, the image (1) shows the input image 1, and the images (2) to (10) show nine degraded parts. The filtering result by the extraction filter is shown. As shown in FIG. 14, there are many portions that are not degraded portions while exhibiting a rust color due to adhesion of rust juice on the coated surface of the steel material, but these degraded portion extraction filters obtained through the above procedure are also good. Judgment.

  A degradation profile is generated for a converted image 6 obtained by applying the degradation diagnosis target input image 1 to the degradation part extraction filter quasi-optimized in the degradation part extraction filter construction step 2. In the degradation profile generation step 7, a transmission part indicating a degradation part in the converted image is detected, the number of degradation parts and the area of each degradation part are calculated, and a degradation profile is created.

Next, in order to evaluate the degree of deterioration of the converted image 6 obtained by applying the deteriorated part extraction filter, a virtual painted surface deterioration image is created 8. The degradation process that can be perceived visually can be broadly classified into two types: generation and growth. Therefore, the generation amount of the degradation part is represented by the number of occurrences of circles (n) approximating the degradation part. Is expressed by the total area ratio (AAR _i ) of the circle approximated to the deteriorated part and the area increase rate (β) of the total area ratio for each growth stage, and the generation and growth are progressed through each stage (L). It repeats regularly and creates a virtual painted surface degradation image that mimics the degradation process. The virtual painted surface deterioration image is created on the assumption that the deteriorated portion grows according to Equation 10 representing the total area ratio of the deteriorated portion at the i-th stage.

In addition, the virtual painted surface deterioration image shows the surface when the distribution of the occurrence position of the deteriorated part is uniformly distributed (uniform type), when one side is biased (half type), Three types are created (central type) when concentrated in the center of FIG. 15 shows an example of the uniform type virtual painted surface deterioration image created, and FIG. 16 shows an example of the half type and center type virtual painted surface deteriorated images created. The virtual painted surface degradation image is the number of occurrences n = {5, 15, 25, 50, 100, 150, 200} (pieces), the total area ratio AAR _i = 0.1%, the area of the total area ratio for each growth stage It was created with an initial value of an increase rate β = 2.0, and the determination area (HA) was 500 pixels × 500 pixels.

In the uniform virtual painted surface degradation image of FIG. 15, the vertical direction represents the total area ratio (AAR _i ) of the deteriorated portion, and the horizontal direction represents the number of occurrences of rust (n) in the deteriorated portion. In the half-type and center-type virtual painted surface deterioration images of FIG. 16, the vertical direction represents the number of occurrences of rust in the deteriorated portion (n), and the horizontal direction represents the total area ratio (AAR _i ) of the deteriorated portion. FIG. 15 shows all created virtual painted surface degradation images, but FIG. 16 shows representative growth stages i = fourth stage (AAR _i = 0.8%) and i = 8th stage (AAR). _i = 12.8%) shows a virtual painted surface deterioration image.

  Next, in the evaluation degradation profile creation step 9, first, the number of degradation portions and the area of each degradation portion are calculated for the virtual painting surface degradation image group created in the virtual painting surface degradation image creation step 8. At the same time, a virtual painted surface degradation image for evaluation is created. The virtual painted surface deterioration image for evaluation is an image obtained by approximating the virtual painted surface deteriorated image as close as possible to the actual coated film deteriorated surface, and is virtually displayed on the image of the steel structure that has been painted as shown in FIG. An image with a painted surface degradation image is used. The determination area (HA) of the virtual painted surface deterioration image for evaluation in FIG. 8 was 580 pixels × 730 pixels.

  Next, the virtual painted surface deterioration image for evaluation is classified according to the degree of deterioration based on the customer's deterioration diagnosis criteria, and the total area ratio (AAR) of the deteriorated portion relative to the determination area (HA) and the number of the deteriorated portions are determined. A non-dimensional value (N / HA) of the number of pixels, a maximum area value (AMR) of each deteriorated portion, and a value representing variation in the size of the deteriorated portion in the same image, A degradation profile for evaluation composed of a three-dimensional evaluation reference space with three parameters as axes is created. The created degradation profile for evaluation is stored together with the virtual painted surface degradation image for evaluation to form a database.

  The virtual painted surface degradation image shown in FIG. 15 and FIG. 16 is taken as an example when judged by the criteria of the judgment method P in “Coating film degradation state and Keren degree sample book” edited by the Railway Technical Research Institute shown in Table 1. explain. According to the diagnostic criteria in Table 1, the degree of deterioration progresses in six stages of AA, A1, A2, B, C, and S in descending order of deterioration severity. If AA, A1, and A2, it is a target for repair and replacement. . Therefore, in this embodiment, the deteriorated state of A2 or higher, which is the object of repair and replacement, is consolidated into one of evaluation A, and deteriorated according to the three-stage evaluation criteria of A, B, and C excluding no abnormality S. The creation of a deterioration profile for evaluation when the degree of evaluation is performed will be described.

  FIG. 9 shows an example of the first evaluation standard as the deterioration profile for evaluation. When a diagnostician visually evaluates the degree of deterioration, a minute deterioration portion with respect to the determination area is ignored. In order to reflect this in the degradation profile for evaluation, AAR, N / HA, and AMR are obtained after erasing the degraded portion displayed by one pixel, which is the minimum unit in the image. Parameters after erasing the minute deteriorated part are represented by FAAR, FN / HA, and FAMR, respectively. The evaluation reference space is obtained as a three-dimensional space with these three parameters as axes, with these three parameters obtained from each evaluation virtual painted surface degradation image. The curved surface shown in the figure is a boundary indicating that a model does not physically exist from the relationship between parameters.

An example of the second evaluation criterion is shown in FIG. In the first evaluation standard, the parameter was set based on the value obtained by erasing the minute deteriorated portion. However, when there is almost no large deteriorated portion and the minute deteriorated portion is mainly spread over the entire surface, It is necessary to consider a minute deteriorated part as an evaluation standard. In order to reflect this in the degradation profile for evaluation, the size of each degraded portion is multiplied by a weighting function, and a value representing the variation in the size of the degraded portion in the same image is set as a parameter. A value (Σ {log (SA _i )} ² ) is set by squaring the size of each deteriorated portion using a value representing the variation in size of the deteriorated portion as a parameter. That is, the second evaluation criteria are the total area ratio (AAR) of the deteriorated portion relative to the determination area (HA), a value (N / HA) obtained by making the number of deteriorated portions dimensionless by the number of pixels of the determination area, A value (Σ {log (SA _i )} ² ) representing a variation in the size of the deteriorated portion in one image is used as a parameter, and these three parameters are obtained from each virtual painted surface deterioration image for evaluation. It consists of a three-dimensional evaluation reference space with parameters as axes.

  As shown in FIG. 10, the distribution of the model in the second evaluation reference space is distributed according to a curved surface. A point 34 deviated from the curved surface in the figure is one in which the distribution state of the deteriorated portion is not normally observed, and an image having a larger distance from the curved surface represents a distribution state of deterioration that cannot actually occur. In order to reflect this in the determination result, the coefficient Cd is calculated by Equation 11 from the distance d from the curved surface, and the result obtained by multiplying the determination result in the evaluation reference space is set as the second evaluation reference.

  Two evaluation criteria constituted by a three-dimensional evaluation reference space with these three parameters as axes are accumulated as an evaluation deterioration profile together with an evaluation virtual painted surface deterioration image to form a database.

Next, in the degradation degree evaluation step 10, first, from the degradation profile obtained from the input image 1, the total area ratio (AAR, FAAR) of the degraded part with respect to the judgment area (HA) and the number of degraded parts are judged area. Values (N / HA, FN / HA) made dimensionless with the number of pixels and the maximum value (AMR, FAMR) of the area of each deteriorated portion or a value representing variation in the size of the deteriorated portion in the same image (Σ {log (SA _i )} ² ) is calculated. Next, an evaluation deterioration profile corresponding to the input image 1 to be subjected to deterioration diagnosis is extracted from the database storing the evaluation virtual painted surface deterioration image and the evaluation deterioration profile, and the three-dimensional evaluation deterioration profile is extracted. The parameters of the input image 1 to be deteriorated are plotted in the evaluation reference space. The plotted point is represented by one point in the evaluation reference space, and a determination result corresponding to the point is output as a result of the deterioration degree.

The determination results are not limited to those that can be reliably determined as A determination, B determination, and C determination depending on the object. There are areas where the determination is ambiguous, such as those in which A determination and B determination are indistinguishable from B determination and C determination. Therefore, as a method of expressing the determination results, the determination results of the three-level evaluation criteria A, B, and C are converted into membership functions A (X, Y, Z), B (X, Y, Z), and C (X, Y, Z). Here, X, Y, and Z are parameter values that affect the determination. In the evaluation reference space of the first evaluation reference, X = FN / HA, Y = log (FAMR), Z = log (FAAR). ) In the evaluation criterion space of the second evaluation criterion, X = N / HA, Y = Σ {log (SA _i )} ² , Z = log (AAR). In this embodiment, the evaluation reference space is divided into an appropriate number of grids, and 0 ≦ A (X, Y, Z), B (X, Y, Z), C (X, Y, Z) ≦ 1 for each grid. Gives the value of

  FIG. 17 shows an example of the degradation diagnosis target input image 1, Table 2 shows parameter values obtained from the input image 1 of FIG. 17, and FIG. 18 shows parameter values plotted in the evaluation reference space. Each determination result is obtained from the values in Table 2. FIG. 18 plots the parameter values in the second evaluation reference space and associates them with each determination result function. In the case of the second evaluation criterion, the determination result calculated from the second evaluation criterion is obtained by multiplying the result of FIG. The determination result of the input image 1 of FIG. 17 by the second evaluation criterion of the deterioration profile for evaluation in the above procedure is A = 0.74, B = 0.49, C = 0.00, and the degree of deterioration is The determination result is A determination.

DESCRIPTION OF SYMBOLS 1 Input image 2 Deterioration part extraction filter construction process 3 Prototype identification process 4 Initial filter generation process 5 Interactive identification process 6 Conversion image 7 Degradation profile generation process 8 Virtual painting surface degradation image creation process 9 Degradation profile creation process 10 for evaluation Deterioration degree Evaluation Step 11 Spatial Filter 12 Spatial Filter Variant 13 Image Capture Unit 14 Image Conversion Unit 15 Degraded Part Extraction Filter Construction Unit 16 Degradation Profile Generation Unit 17 Storage Unit 18 Educational Teacher Data Creation Unit 19 Prototype Filter Construction Unit 20 Prototype Identification Process Means for performing 21 Degraded part extraction filter selection means 22 Acceptance means 23 Image selection means 24 Initial filter generation means 25 Means for performing initial filter generation process 26 Acceptance means 27 Image selection means 28 Of points that have been excluded from the means 29 interactively identify perform process means 30 virtual paintwork degraded image creating unit 31 for evaluation degradation profile creation means 32 storage means 33 deterioration degree evaluating means 34 curved

Claims (5)

In a method of evaluating the degree of deterioration of the painted surface using a full color bitmap image obtained by photographing the surface of a steel structure such as a steel tower or steel bridge that has been painted as an input image,
A process of constructing a deteriorated part extraction filter for converting into an image in which only the deteriorated part of the painted surface in the input image is transmitted;
Detecting a transmission part indicating a deterioration part in a converted image obtained by applying the deterioration part extraction filter, calculating the number of deterioration parts and the area of each deterioration part, and generating a deterioration part profile;
The amount of generation of the deteriorated part is represented by the number of circles that approximate the deteriorated part, and the amount of growth of the deteriorated part is expressed by the total area ratio of the circle that approximates the deteriorated part and the area increase rate of the total area ratio for each growth stage. The process of creating various virtual painted surface degradation images by simulating the occurrence and growth of painted surface degradation by regularly repeating the generation and growth according to the amount and growth amount of the deteriorated part set in advance When,
For the created virtual paint surface degradation image group, calculate the number of degradation parts and the area of each degradation part and create a virtual paint surface degradation image for evaluation that approximates the actual paint film degradation surface as much as possible, Creating an evaluation deterioration profile from the evaluation virtual painted surface deterioration image and the customer's deterioration diagnosis standard, accumulating the evaluation deterioration profile and the evaluation virtual painted surface deterioration image data, and forming a database;
A deterioration degree evaluation step for extracting a deterioration profile for evaluation corresponding to an input image to be deteriorated from the database and comparing the deterioration profile obtained from the input image to be deteriorated and determining a deterioration degree;
A paint deterioration diagnosis method characterized by diagnosing the degree of deterioration of the paint surface by means of the above. The step of constructing the deteriorated part extraction filter includes:
The input image group is classified into multiple categories based on the shooting conditions such as shooting target and shooting conditions, and for each category, multiple typical input images and ideal output images that clearly indicate only the degraded part of the input image. Create training data consisting of pairs of images, generate a sub-optimized spatial filter to extract only the degraded part from the input image using the genetic algorithm method for each category, and generate the generated space Prototype identification process that retains the filter group as a prototype group of the deteriorated part extraction filter,
Select several prototypes of degraded part extraction filters for each category from the prototype group of degraded part extraction filters, and diagnose from the converted image group obtained by applying the input image to the selected prototype of the degraded part extraction filter An initial filter generation process in which a plurality of converted images are selected according to a result determined by the person to be relatively appropriate, and a deteriorated portion extraction filter derived from the selected images is retained as an initial filter group;
A first step of selecting a converted image according to a result determined by the diagnostician to be relatively appropriate from the converted image group obtained by applying the input image to the initial filter group, and the selected image is derived A second step of generating a next-generation degraded part extraction filter group having a degraded part extraction filter group as a parent using a genetic algorithm technique, and applying an input image to the next-generation degraded part extraction filter group And a third step of selecting a converted image according to a result determined by the diagnostician to be relatively appropriate from the group of converted images, and from the second to the second by an interactive evolutionary calculation method. An interactive identification process for creating a sub-optimal degraded part extraction filter by repeating step 3;
The paint deterioration diagnosis method according to claim 1, comprising: a three-stage process.   The evaluation deterioration profile is calculated from the number of deteriorated portions and the area of each deteriorated portion, the total area ratio of the deteriorated portions relative to the determination area, and a value obtained by making the number of deteriorated portions dimensionless by the number of pixels of the determination area. And a three-dimensional evaluation reference space with three parameters as axes, with the maximum value of the area of each deteriorated portion or the value representing the variation in size of the deteriorated portion in the same image as parameters. The paint deterioration diagnosis method according to claim 1, wherein In a device that evaluates the degree of deterioration of the painted surface using a full-color bitmap image obtained by photographing the surface of a steel structure such as a steel tower or steel bridge that has been painted as an input image,
A deteriorated part extraction filter construction means for constructing a deteriorated part extraction filter that converts and outputs an image through which only the deteriorated part of the painted surface in the input image is transmitted;
A deterioration profile generating means for detecting a transmission portion indicating a deterioration portion in a converted image obtained by applying the deterioration portion extraction filter, calculating the number of deterioration portions and the area of each deterioration portion, and generating a deterioration profile;
The amount of generation of the deteriorated part is represented by the number of circles that approximate the deteriorated part, and the amount of growth of the deteriorated part is expressed by the total area ratio of the circle that approximates the deteriorated part and the area increase rate of the total area ratio for each growth stage. This is a virtual model that simulates the occurrence and growth of paint surface deterioration by regularly repeating the generation and growth according to the amount of generation and growth amount of the deteriorated part set in advance, and creates various virtual paint surface deterioration images. Painting surface degradation image creation means,
For the created virtual painted surface deterioration image, the number of deteriorated parts and the area of each deteriorated part are calculated and a virtual painted surface deteriorated image for evaluation that is approximated as much as possible to the actual coated film deteriorated surface is created. A degradation profile creation means for evaluation that creates a degradation profile for evaluation from a virtual painted surface degradation image for a customer and a customer degradation diagnosis standard;
A database formed by accumulating data of the degradation profile for evaluation and the virtual painted surface degradation image for evaluation,
A deterioration degree evaluation means for taking out a deterioration profile for evaluation corresponding to an input image to be deteriorated from the database and comparing the deterioration profile obtained from the input image to be deteriorated and determining a deterioration degree;
A paint deterioration diagnosis device characterized by comprising: The deteriorated part extraction filter construction means includes:
An input image group is classified into multiple categories based on shooting conditions such as shooting target and shooting conditions, and multiple typical input images and ideal output images that clearly indicate the degraded part of the input image for each category Using a method of creating educational data for learning to create pairs of educational data and a genetic algorithm method for each category, generating a sub-optimized spatial filter to extract only the degraded part from the input image Means for performing a prototype identification process comprising a prototype filter construction means for holding the generated spatial filter group as a prototype group of the deteriorated part extraction filter;
Degraded part extraction filter selecting means for selecting several degraded part extraction filter prototypes for each category from the group of degraded part extraction filter prototypes, and obtained by applying an input image to the selected degraded part extraction filter prototype A receiving unit that receives information on a result determined by the diagnostician to be relatively appropriate from the group of converted images, an image selecting unit that selects a converted image according to the received information, and a degradation derived from the selected image Means for performing an initial filter generation process comprising an initial filter generation means for retaining a partial extraction filter as an initial filter group;
Accepting means for accepting information of a result determined by the diagnostician to be relatively appropriate from the converted image group obtained by applying the input image to the deteriorated portion extraction filter group, and selecting the converted image according to the accepted information IDENTIFICATION PROCESS USING IMAGE SELECTION UNIT AND UPDATE UNIT FOR GENERATING NEXT GENERATION DERIVED EXTENSION FILTER GROUPS USED BY GENERATED ALGORITHM METHOD Means for
5. The paint deterioration diagnosis apparatus according to claim 4, further comprising means for performing the three-stage process.

Images