JP5248213B2 - Corrosion degree evaluation method, apparatus and program - Google Patents

Description

  The present invention relates to a corrosion degree evaluation method, apparatus, and program. More specifically, the present invention relates to a technique suitable for use in evaluating the degree of corrosion based on image data obtained by photographing hardware.

  In this specification, the degree of corrosion is used to mean the degree of corrosion due to the occurrence of rust in hardware. In addition, parts that are in-service equipment, etc., and that are actually subject to judgment of the degree of corrosion during inspections, etc. are called judgment implementations. Judgment implementation such as parts of equipment of the same type as the judgment implementation in addition to the same kind of judgment implementation as the part corresponding to the judgment implementation of the same type of equipment An object that is judged to be a reference when determining the degree of corrosion of a judgment object using image data because the situation is similar to the above and the situation of corrosion is similar.

  As a conventional method of evaluating the degree of corrosion based on image data obtained by photographing hardware that is a judgment implementation object, a label indicating which corrosion degree can be discarded or reused on an image obtained by photographing the hardware of the evaluation object Next, the image of the object to be evaluated is represented on the feature space, and an image on which it is difficult to discriminate the corrosion degree label on the feature space is selected, and the degree of corrosion on the feature space is selected from the selected image. The discriminant surface of the label is generated, and the image obtained by photographing the judgment object is expressed on the same feature space, and the corrosiveness that can be discarded or reused is judged based on the discriminant surface ( Patent Document 1).

  Here, in the method of Patent Document 1, a discrimination surface is generated based on an image obtained by photographing an evaluation object. That is, if the photographed image appropriately represents the corrosion degree of the hardware, the discrimination surface generated in the feature space corresponds to the corrosion degree of the hardware appropriately.

  However, even if the subject to be photographed is the same, the photographed image will be different if the illumination at the time of photographing is different. For example, if it is photographed in the evening, it becomes a reddish image, and if it is cloudy, it becomes a bluish image. As described above, when images having different illumination conditions are used, a discriminant plane is generated that is influenced by the difference in the images due to the difference in the illumination conditions during photographing that is irrelevant to the actual degree of corrosion.

  On the other hand, in many of the commercially available digital cameras, an automatic exposure setting function that automatically adjusts the brightness of an image in consideration of the brightness of a shooting target in order to cope with a change in lighting conditions at the time of shooting, and It is equipped with an automatic white balance setting function that automatically adjusts the color of an image in consideration of the color of the subject. Then, the automatic exposure setting function and the automatic white balance setting function work, resulting in an image that is arbitrarily converted into a color and brightness different from the actual state of the surface of the evaluation object.

  On the other hand, in the method of Patent Document 1, the discrimination surface is generated by strictly handling the data of the color and brightness recorded in the image obtained by photographing the evaluation object. Therefore, even for hardware with the same degree of corrosion, an image whose color and brightness have been adjusted and converted by the automatic exposure setting function and automatic white balance setting function will be used, and the actual state due to errors in the adjustment and conversion The discriminant plane is generated by reflecting the deviation from. The influence of the adjustment / conversion error on the discriminant plane generation by these automatic setting functions increases as the variation in illumination conditions increases and decreases as the number of images decreases. In other words, the generated discriminant surface may not be appropriate because it does not take into account the adjustment / conversion error caused by the automatic setting function of the camera. The accuracy of the determination is reduced.

  That is, in the method of Patent Document 1, when an image taken outdoors is used, the image is recorded as a different color even if the degree of corrosion is the same because the light source of the shooting environment changes during outdoor photography. The generation of the discriminating surface is affected, and the determination accuracy of the corrosion degree is lowered. For this reason, in the method of Patent Document 1, for example, a digital camera with a hood in which outside light does not enter is used, and an illumination condition is obtained by close-up of the evaluation object and the judgment object using artificial illumination in the hood. It is desirable to devise a method to capture the same image if the image capturing target is the same.

JP 2005-274299 A

  However, in the corrosion degree evaluation method disclosed in Patent Document 1, the evaluation object and the determination object can be close-up and the lighting condition can be kept constant to suppress a decrease in determination accuracy. There is a problem that the determination accuracy is lowered when it is difficult to keep the lighting conditions constant because the determination implementation object and the shooting point are not close enough to perform close-up photography. For this reason, for example, it is impossible to perform close-up photography like a steel material above a steel tower, or it is dangerous to perform close-up photography due to high voltage or high heat, and the evaluation target cannot maintain constant lighting conditions. It is difficult to say that the accuracy of the determination of the degree of corrosion performed by photographing the object and the determination object is high. Furthermore, even if close-up photography can be performed, in order to perform close-up photography, the work must be performed while repeating the movement to the immediate vicinity of the evaluation target part over the entire evaluation object or judgment object, so work efficiency is poor. There's a problem.

  Therefore, the present invention can accurately determine the degree of corrosion even if the image is taken from a point away from an outdoor evaluation object or judgment object and the illumination condition is not kept constant. An object of the present invention is to provide a method, apparatus and program for evaluating the degree of corrosion.

  In order to achieve this object, a corrosion degree evaluation method according to claim 1 includes a step of creating a case image of an evaluation object and data of a corrosion degree label for each case image, and a color temperature and brightness for the case image. Or changing one of them to create processing case image data, and attaching the data on the corrosion degree of the case image that is the basis of the processing case image to the generated processing case image data Generating a corrosion degree discrimination surface by a support vector machine using the case image and the corrosion level label data of the case image and the processing case image and the corrosion level label data of the machining case image A step of creating determination image data of an object, and a step of determining the degree of corrosion of the determination image using the determination image data and the determination surface That.

  According to a fourth aspect of the present invention, there is provided a corrosion degree evaluation apparatus that reads a case image of an evaluation object and label data of a corrosion degree for each case image, and changes color temperature and / or brightness of the case image. The processing example image data is created, and the processing example image data created is attached with data on the label of the corrosion degree of the example image that is the basis of the processing example image, the example image, and the example image Corrosion degree label data, processing example image, means for generating a corrosion degree discrimination surface by a support vector machine using the corrosion degree label data of the processing example image, and reading of judgment image data of a judgment object Means and means for judging the degree of corrosion of the judgment image using the judgment image data and the judgment surface.

  According to a seventh aspect of the present invention, there is provided a corrosion degree evaluation program for reading a case image of an evaluation object and data of a corrosion degree label for each case image from storage means, and a color temperature and / or brightness for the case image. The processing example image data is created by changing the data, and the processing example image data that is created is attached with the data of the corrosion degree label of the case image that is the basis of the processing case image, Corrosion degree label data of the case image and processing example image and corrosion degree label data of the processing case image are used to generate a corrosion degree discrimination surface by a support vector machine, and Let the computer perform the process of reading data from the storage means and the process of determining the corrosion degree of the determination image using the determination image data and the determination surface That.

  Therefore, according to this corrosion degree evaluation method, apparatus, and program, in addition to the image actually obtained by photographing, an image obtained by changing the hue and / or brightness is also used to generate the corrosion degree discrimination surface. By doing so, a discriminant plane based on the image group with some ambiguity added to the white balance and exposure is generated, so that the adjustment / conversion error caused by the automatic white balance setting function and the automatic exposure setting function of the camera can be reduced. A discriminant plane with reduced influence is generated.

  According to a second aspect of the present invention, there is provided a corrosion degree evaluation method comprising the steps of creating a case image of an evaluation object and data of a corrosion degree label for each case image, and the case image and the data of the corrosion degree label of the case image. Using the support vector machine to generate a corrosion degree discrimination surface, to create a judgment image data of a judgment object, and to process the judgment image by changing color temperature and / or brightness The method includes a step of creating determination image data, and a step of determining the corrosion degree of the determination image using the determination image data, the processing determination image data, and the determination surface.

  Further, the corrosion degree evaluation apparatus according to claim 5 uses means for reading the case image of the evaluation object and the label data of the corrosion degree for each case image, and the case image and the data of the corrosion degree label of the case image. A means for generating a corrosion degree discrimination surface by a support vector machine, a means for reading judgment image data of a judgment object, and a processing judgment image by changing the color temperature and / or brightness of the judgment image. And means for determining the degree of corrosion of the determination image using the determination image data, the processing determination image data, and the determination surface.

  Further, the corrosion degree evaluation program according to claim 8 reads the case image of the evaluation object and the data of the corrosion degree label for each case image from the storage means, the case image and the corrosion degree label of the case image. A process for generating a determination surface of the corrosion degree by the support vector machine using the data, a process for reading the determination image data of the determination object from the storage means, and a color temperature and / or brightness for the determination image. The computer is caused to perform processing for changing to create data for the processing determination image and processing for determining the degree of corrosion of the determination image using the determination image data, the processing determination image data, and the determination surface.

  Therefore, according to the corrosion degree evaluation method, apparatus, and program, the degree of corrosion is determined by simultaneously using an image obtained by actually taking a photograph and an image obtained by changing the hue and / or brightness. Because of this, the degree of corrosion is determined based on a group of images with some ambiguity added to the white balance and exposure, so the effects of adjustment and conversion errors caused by the camera's automatic white balance setting function and automatic exposure setting function The degree of corrosion is relaxed.

  According to a third aspect of the present invention, there is provided a corrosion degree evaluation method comprising: creating a case image of an evaluation object and data of a corrosion degree label for each case image; and color temperature and / or brightness for the case image. The processing example image data is created by changing the data, and the corrosiveness level label data of the case image that is the basis of the processing case image is added to the created processing case image data, A step of generating a corrosion degree determination surface by a support vector machine using the corrosion degree label data of the case image and the processing case image and the corrosion degree label data of the processing case image; and A step of creating data, a step of changing the color temperature and / or brightness of the judgment image, and creating the data of the processing judgment image, and the judgment image Data and using the data as well as discrimination surface of the processing determination image is to have a determining corrosion rate determination image.

  The corrosion degree evaluation apparatus according to claim 6 reads the case image of the evaluation object and the label data of the corrosion degree for each case image, and changes the color temperature and / or brightness with respect to the case image. The processing example image data is created, and the processing example image data created is attached with data on the label of the corrosion degree of the example image that is the basis of the processing example image, the example image, and the example image Corrosion degree label data, processing example image, means for generating a corrosion degree discrimination surface by a support vector machine using the corrosion degree label data of the processing example image, and reading of judgment image data of a judgment object Means for changing the color temperature and / or brightness of the judgment image and creating one of the judgment image data, the judgment image data and the judgment image And to have a means for determining the corrosion rate of the determination image by using the chromatography data and discrimination surface.

  The corrosion degree evaluation program according to claim 9 includes a process for reading the case image of the evaluation object and the label data of the corrosion degree for each case image from the storage means, and the color temperature and / or brightness for the case image. The processing example image data is created by changing the data, and the processing example image data that is created is attached with the data of the corrosion degree label of the case image that is the basis of the processing case image, Corrosion degree label data of the case image and processing example image and corrosion degree label data of the processing case image are used to generate a corrosion degree discrimination surface by a support vector machine, and A process of reading data from the storage means, a process of changing the color temperature and / or brightness of the determination image, and creating the processing determination image data, and the determination It is the process of determining corrosion rate determination image by using the data as well as discrimination surface data and processing determination image of the image to be performed by the computer.

  Therefore, according to this corrosion degree evaluation method, apparatus, and program, in addition to the image actually obtained by photographing, an image obtained by changing the hue and / or brightness is also used to generate the corrosion degree discrimination surface. In addition, by determining the degree of corrosion, a discriminant plane based on a group of images with some ambiguity added to the white balance and exposure is generated and the degree of corrosion is determined. A determination surface is generated in which the influence of adjustment / conversion errors caused by the balance setting function and the automatic exposure setting function is mitigated, and the degree of corrosion is determined.

  According to the corrosion degree evaluation method, apparatus and program according to claims 1, 4 and 7, a discriminant plane is generated based on an image group in which some ambiguity is added to white balance and exposure. It is possible to generate a discriminating surface that can reduce the effects of adjustment / conversion errors caused by the automatic white balance setting function and automatic exposure setting function, and is an image taken from a point away from an outdoor evaluation object or judgment object. It is possible to accurately determine the degree of corrosion even for an image whose lighting conditions are not kept constant.

  Further, according to the corrosion degree evaluation method, apparatus, and program according to claims 2, 5 and 8, since the degree of corrosion is determined based on an image group in which some ambiguity is added to white balance and exposure. It is possible to judge the degree of corrosion by reducing the effects of adjustment / conversion errors by the automatic white balance setting function and automatic exposure setting function of the camera, and photographed from a place away from the outdoor evaluation object and judgment object Even if the image is an image and the illumination condition is not kept constant, it is possible to accurately determine the degree of corrosion.

  In addition, according to the corrosion degree evaluation method, apparatus, and program according to claims 3, 6, and 9, the discrimination surface based on the image group in which some ambiguity is added to the white balance and the exposure is generated and the corrosion is generated. Since the determination of the degree is performed, it is possible to generate a determination surface that mitigates the influence of the adjustment / conversion error by the automatic white balance setting function and the automatic exposure setting function of the camera and to determine the corrosion degree, It is possible to accurately determine the degree of corrosion even if the image is taken from a point far from the outdoor evaluation object or judgment execution object and the illumination condition is not kept constant.

  Hereinafter, the configuration of the present invention will be described in detail based on the best mode shown in the drawings.

  1 and 2 show an example of embodiments of the corrosion degree evaluation method, apparatus, and program of the present invention. This corrosion degree evaluation method includes a step of creating a case image of an evaluation object and data of a corrosion degree label for each case image, and changing the color temperature and / or brightness of the case image to process examples. Creating the image data and attaching the data of the corrosion degree label of the case image that is the basis of the processed case image to the created case image data; and the case image and the corrosion degree of the case image A step of generating a corrosion degree determination surface by a support vector machine using the label data, the processing example image, and the corrosion level label data of the processing example image; and a step of generating determination image data of a determination object , A step of changing the color temperature and / or brightness of the determination image and generating one of the determination image data, and the determination image data and processing So that a step determining the corrosion rate of the determination image by using the data as well as discrimination surface constant image.

  The said corrosion degree evaluation method is implement | achieved as a corrosion degree evaluation apparatus of this invention. The corrosion degree evaluation apparatus of the present invention is a means for reading a case image of an object to be evaluated and a corrosion degree label data for each case image, and processing the case image by changing the color temperature and / or brightness. Means for creating case image data and attaching data of the label of the degree of corrosion of the case image that is the basis of the processed case image to the created case image data, the case image, and the degree of corrosion of the case image Means for generating a determination surface for the corrosion degree by a support vector machine using the data of the label and the processing example image and the corrosion level label data of the processing example image, and means for reading the determination image data of the determination object , Means for changing the color temperature and / or brightness of the determination image to create processing determination image data, determination image data, and processing determination image data Beauty and a means for determining the corrosion rate of the determination image by using a discrimination plane.

  The corrosion degree evaluation method and the corrosion degree evaluation apparatus described above can also be realized by executing the corrosion degree evaluation program of the present invention on a computer. In the present embodiment, a case where a corrosion degree evaluation program is executed on a computer will be described as an example.

  FIG. 2 shows the overall configuration of the corrosion degree evaluation apparatus 10 of the present embodiment for executing the corrosion degree evaluation program 17. The corrosion degree evaluation apparatus 10 includes a control unit 11, a storage unit 12, an input unit 13, a display unit 14, and a memory 15, and is connected to each other by a signal line such as a bus. A data server 16 is connected to the corrosion degree evaluation apparatus 10 through a signal line such as a bus, and signals such as data and control commands are transmitted / received (ie, input / output) to / from each other via the signal line. Is called.

  The control unit 11 performs calculation related to the control of the entire corrosion degree evaluation apparatus 10 and the evaluation of the corrosion degree using image data by the corrosion degree evaluation program 17 stored in the storage unit 12. Processing device). The storage unit 12 is storage means capable of storing at least data and programs, and is, for example, a hard disk. The memory 15 serves as a memory space that is a work area when the control unit 11 executes various controls and operations, and is, for example, a RAM (Random Access Memory). .

  The input unit 13 is an interface for giving at least an operator's command to the control unit 11, and is, for example, a keyboard.

  The display unit 14 performs drawing / display of characters, graphics, and the like under the control of the control unit 11 and is, for example, a display.

  The data server 16 is storage means capable of storing at least data, for example, a server.

  Then, by executing the corrosion degree evaluation program 17, the control unit 11 of the corrosion degree evaluation apparatus 10 reads the case image of the evaluation object and the case data reading unit as means for reading the data of the corrosion degree label for each case image. 11a and a case image that is the basis of the processed case image in the data of the processed case image created by changing the color temperature and / or brightness of the case image and changing one of them. A case image processing unit 11b as means for giving the data of the corrosion degree label, the case image, the data of the corrosion degree label of the case image, and the data of the processing case image and the corrosion degree label of the processing case image. The case image feature quantity extraction unit 11c and the case image expression unit 11d as a means for generating a corrosion degree discrimination surface using a support vector machine and discrimination The generation unit 11e, the determination image data reading unit 11f as means for reading the determination image data of the determination object, and the data of the processing determination image by changing the color temperature and / or brightness of the determination image. A determination image processing unit 11g as a means for creating, a determination image feature amount extraction unit 11h as a means for determining the corrosion degree of the determination image using the data of the determination image, the data of the processing determination image, and the determination surface, and the corrosion degree determination Part 11i.

  Further, as shown in FIG. 1, the corrosion degree evaluation program 17 for realizing the corrosion degree evaluation method of the present invention in the corrosion degree evaluation apparatus 10 largely generates a discriminant surface for determining the corrosion degree. A determination surface generation step (S1) and a determination step (S2) for determining the degree of corrosion of the determination object using the image and the determination surface taken during inspection of the determination object.

  More specifically, as shown in FIG. 1, the corrosion degree evaluation program 17 reads the case image of the evaluation object and the data of the corrosion degree label for each case image (S1-1), and the case image The processing example image data is created by changing the color temperature and / or brightness, and the data on the label of the corrosion degree of the example image that is the basis of the processing example image is created. (S1-2), step (S1-3) of extracting the image feature amount of the case image data and the processed case image data, and the case image data and the processed case image data using the image feature amount. A step of expressing on the feature space (S1-4), and a step of generating a determination surface of the corrosion degree using the case image data and the processed case image data expressed on the feature space (S1-5). A step of reading determination image data of the determination object (S2-1), and a step of creating processing determination image data by changing the color temperature and / or brightness of the determination image (S2-). 2), a step (S2-3) of extracting image feature amounts of the determination image data and the processing determination image data, and a step (S2-4) of determining the corrosion degree of the determination image using the determination surface. .

  Here, in this embodiment, the case where the determination of the degree of corrosion is performed on a steel material of a power transmission tower will be described as an example. Therefore, in this embodiment, the steel material in general of the power transmission tower can be an evaluation object, and the steel material that is a part of the steel tower selected as an inspection object among the in-service towers is a determination execution object.

  First, the discrimination surface generation step (S1) will be described.

  In executing the corrosion degree evaluation method of the present invention, first, the case data reading unit 11a of the control unit 11 reads the case data of the evaluation object (S1-1).

  The case data in the present invention is combination data of case image data and corrosion degree label data for each case image.

  Case image data (hereinafter referred to as case image data) is image data obtained by photographing an evaluation object with a digital camera. Note that the image used in the present invention may be an image taken using an automatic exposure setting function and an automatic white balance setting function provided in the digital camera. Furthermore, the image used in the present invention may be one in which the illumination conditions change every time the image is taken, or may be one taken using a telephoto lens at a position away from the evaluation object.

  Here, in the image (hereinafter referred to as a captured image) of the evaluation object that has been captured by the camera, the size of the steel material in the captured image varies, and distortion occurs depending on the captured angle. In addition, images other than the steel material to be evaluated, such as the sky, buildings, fields, and other steel materials, are also photographed together with the image captured by the telephoto lens. Therefore, if such a photographed image is used as it is to determine the corrosion degree, a determination result depending on the background and the size and shape in the image is irrelevant to the corrosion degree. For this reason, as the case image data used in the present invention, data obtained by cutting an evaluation object portion (a steel material portion in the present embodiment) in an actual captured image into a square is used. In other words, the image data cut into squares from the actual photographed image is stored in the case database 18 as case image data.

  Note that, in the present embodiment, it is desirable that the case image data is cut out from the captured image so as to include the entire steel material in the width direction. This is because rust may be generated from the edge of the steel material, and the edge information may be important. In addition, when the steel material is imaged diagonally, it is desirable that the steel material portion is cut out after the photographed image is rotated so that the steel material is located in the horizontal or vertical direction. On the other hand, when the steel material is photographed while being tilted in the depth direction, the width of the steel material is shown to be narrower toward the back side of the photographed image, so that the entire width direction cannot be cut into a square by simple processing such as rotation. In addition, since the straight steel material is curved and photographed due to the spherical distortion of the camera lens, the entire width direction cannot be cut into a square by simple processing. In such a case, for example, it is desirable that the image is cut after being transformed into a square by performing enlargement or distortion correction on the back side using a perspective cut-out function of commercially available image processing software.

  The corrosion degree label indicates the classification (also called class) of the degree of progress of corrosion with the lapse of time of the evaluation object. The type (contents) and number of labels of the degree of corrosion are not limited to specific ones, and are appropriately set by the operator in consideration of the type of equipment to be evaluated and the purpose of evaluation.

  Specifically, for example, when carrying out the corrosion degree determination using a steel material that is a part of an in-service steel tower as a determination object as in this embodiment, the deterioration of hot dip galvanizing that is a rust-proof coating of the steel material Depending on the degree, early label (class 1) where hot dip galvanization remains sufficiently, suitable label (class 2) where galvanization degradation partially reaches the alloy layer 3) Use a label with three corrosion degrees, a late label (class 3), where the alloy layer is exposed on the entire surface and rust is seen from some parts of the iron substrate and parts close to the iron substrate. Can be considered. In the present embodiment, the description will be made on the assumption that the labels of the three corrosion degrees are used.

  The corrosion degree label is assigned to each case image by an expert who determines the degree of corrosion of the evaluation object by looking at the entire original captured image from which the case image is cut. When the corrosion degree label is combined with the case image data and stored in the case database 18, it is given as numerical data (also called a flag) such as 1, 2, 3, etc., for example.

  In the present embodiment, the case data relating to the evaluation object is accumulated as a case database 18 in the data server 16 which is a storage means. The case database 18 may be stored in the storage unit 12 that is a storage unit.

  The case data reading unit 11 a reads the combination data of the case image data of the evaluation object and the corrosion level label data assigned to each case image from the case database 18 stored in the data server 16. Then, the case data reading unit 11 a stores the read case data in the memory 15.

  Next, the case image processing unit 11b of the control unit 11 performs color processing on the case image of the evaluation object (S1-2).

  Specifically, the case image processing unit 11b first reads from the memory 15 case image data stored in the memory 15 in the process of S1-1 (referred to as an original case image in the process of S1-2).

  Then, the case image processing unit 11b creates an image with an increased color temperature and an image with a reduced color temperature with respect to the original case image. When the color temperature is increased, redness is increased with respect to the original case image, and when the color temperature is decreased, blueness is increased with respect to the original case image.

  The degree of change in color temperature as the color processing of the image is not limited to a specific value, but the overall color of the evaluation object, the lighting conditions at the time of shooting the case image, and the color of the actual original case image group Based on the tendency, the color temperature of the original case image group is assumed, and the width of the change is appropriately determined by the operator in consideration of the assumed color temperature and the overall tendency of the hue.

  Specifically, for example, when using a rust image of steel as in the present embodiment, it is considered that the color temperature is low due to redness of rust in the color temperature of sunlight 5000 to 6000K, and the original case image Is assumed to be 5000K, and an image in which the color temperature is converted to 4500K, 0.9 times the assumed color temperature of 5000K, and an image in which the color temperature is converted to 5500K, which is 1.1 times, are created. Can be considered.

  The case image processing unit 11b further creates an image that is brighter and a darker image than the original case image. The degree of change in brightness as color processing of the image is not limited to a specific value, but considers the overall tendency of the brightness of the actual original case image group and the lighting conditions when shooting the case image The operator decides accordingly.

  Specifically, for example, it is conceivable to create an image obtained by converting the original case image by a factor of 0.9 to make it slightly darker and an image obtained by changing the factor by a factor of 1.1 to make it slightly brighter. .

  Also, the number of images created by changing the color temperature and brightness (hereinafter referred to as a processed image) is not limited to a specific number, and is an operation for generating a discrimination surface in the processing of S1-5. The operator appropriately determines in consideration of processing load and the like.

  It is also conceivable to determine the degree of color processing (specifically, the degree of change in color temperature and the degree of change in brightness) by the case image processing unit 11b and the number of processed images to be created by a cross-validation method. Specifically, when applying the cross-validation method (also referred to as n-division cross-validation method, where n is the number of divisions and is a natural number of 2 or more) to the present invention, first, all of the given image data is processed. Randomly divide into n so that the number is almost the same (the divided groups are C1, C2,..., Cn respectively). Next, a discriminant plane is created using image data belonging to (n−1) divided groups (for example, C2 to Cn) of the n divided groups. Then, the remaining one divided group (that is, C1) is used as a test determination group, and the degree of corrosion of image data belonging to the group is determined. Similarly, the test determination group is changed from C2 to Cn, the test determination is performed a total of n times, and the number of determination correct answers is totaled. In addition, the process from random division to test determination is performed while changing the degree of color processing and the number of processed images to be created multiple times, and the degree of color processing and the number of processed images to be created when the number of correct answers is the largest. It is also possible to adopt.

  In the present embodiment, a description will be given below on the assumption that a total of eight types of images (referred to as processed case images) are created in addition to the original case image by combining three types of hue and three types of brightness.

  The case image processing unit 11b creates eight types of processing case images for each of the case image data (that is, the original case images) stored in the memory 15 in the process of S1-1. Further, the case image processing unit 11b reads the value of the corrosion label for each original case image from the memory 15, and processes the same label as the label of the corrosion degree given to the original original case image of the created processing case image. It is given to the case image.

  Then, the case image processing unit 11b causes the memory 15 to store the processing case image data and the corrosion level label data provided for each processing case image as combination data.

  Next, the case image feature amount extraction unit 11c of the control unit 11 extracts the image feature amount of the case image data of the evaluation object and the processed case image data (S1-3).

  The number of images that can be used for the generation of the discriminating surface in the processing of S1-5 is considered to be about several hundred per corrosiveness label considering the effort of photographing and corrosiveness labeling. Further, the data of the case image cut out from the photographed image and the data of the processed case image are, for example, RGB data (that is, three-dimensional) of several hundreds of pixels in the vertical and horizontal directions, and the total data amount reaches several hundred thousand.

  As described above, it is known that when the amount of data per image is too large with respect to the total number of images, the reliability of identification is lowered in image identification using machine learning. Further, in machine learning, the larger the amount of data, the longer it takes to calculate. For this reason, it is necessary to convert the image into small data (referred to as an image feature amount) expressing the feature of the image.

  In the present invention, it is conceivable to use, for example, a histogram or scaling as a method for converting original image data into image feature values.

  The histogram is a feature amount obtained by summing up the frequency of colors of all pixels in the image data, and represents color information of the entire image. Therefore, in general, it is suitable for the feature amount when the appearance frequency of the overall color of the image is meaningful regardless of the position in the image.

  Scaling is a feature amount that divides an image into a grid pattern and calculates a representative value of a pixel group in each area, and represents rough arrangement information of colors in the image. Therefore, in general, it is suitable for a feature amount when the arrangement of colors in an image is meaningful. For example, an average value or an intermediate value is used as the representative value.

  In addition, as a basic color system of a digital image, an RGB color system that expresses the color of each pixel with three primary colors of R (red), G (green), and B (blue), and a color of each pixel that is H ( There is an HSV color system expressed by three variables of hue), S (saturation), and V (lightness).

  Therefore, in the present invention, it is conceivable to use an RGB histogram, RGB scaling, HSV histogram, and HSV scaling that combine these as the image feature amount.

  For example, in RGB scaling, an image is divided into a grid pattern to calculate a representative value of RGB values of all pixels for each grid, and the calculated value is set as a representative value of the grid. Thereby, the RGB scaling represents a rough arrangement of colors such as “the center is bluish and the corner is reddish”. In general, normalization is performed so that each component of the feature value has an average of 0 and a variance of 1.

  The HSV histogram is calculated as the frequency of the entire image after converting RGB of each pixel in the image data into HSV. Accordingly, the HSV histogram represents the appearance frequency of the overall color of the image such as “many red portions”. In general, normalization is performed so that each component of the feature value has an average of 0 and a variance of 1. In addition, when it is confirmed by preliminary examination or the like that the accuracy of image discrimination is improved by placing emphasis on one of H (hue), S (saturation), and V (lightness). It is effective to adjust the ratio of the dimensionality of H, S, and V. For example, when it is confirmed that the accuracy is improved by placing emphasis on H (hue), the ratio of the number of dimensions may be H: S: V = 2: 1: 1.

  The case image feature amount extraction unit 11c reads from the memory 15 the case image data stored in the memory 15 in the process of S1-1 and the processing case image data stored in the memory 15 in the process of S1-2. Image feature values are calculated for each data.

  Then, the case image feature amount extraction unit 11c stores the image feature amount for each image data in the memory 15 in association with the image data that is the basis of the image feature amount calculation.

  Next, the case image expression unit 11d of the control unit 11 expresses the case image data and the processed case image data of the evaluation object on the feature space (S1-4).

  The feature space in the present invention is a space that represents image data by vectorization (that is, coordinated) by parameters (image feature amounts in the present invention) that constitute image data, or data that is vectorized by parameters. A space that represents image data by nonlinear mapping in a dimensional space.

  The case image expression unit 11d reads, from the memory 15, the image feature amount for each of the case image data and the processed case image data stored in the memory 15 in the process of S1-3. Then, the case image representation unit 11d represents the case image data and the processed case image data on the feature space based on the read image feature amount.

  Next, the discriminant surface generation unit 11e of the control unit 11 generates the discriminant surface of the corrosion degree using the case image data and the processing case image data of the evaluation object expressed in the feature space (S1-5). .

  In the present embodiment, a support vector machine (hereinafter referred to as SVM) is used as a pattern identification method for image data. The SVM is a method for finding a case where the class is most difficult to discriminate from a given group of cases and generating a discriminant plane (for example, a hyperplane) based on the found case.

  FIG. 3 shows a conceptual diagram of discriminant plane generation when the SVM is applied to the present invention. FIG. 3 is a conceptual diagram of discriminant plane generation for discriminating between class 2 and class 3. When SVM is applied to the present invention, as shown in FIG. 3, a case image and a processed case image data group 2 can be expressed on the feature space 1 and a class can be discriminated from the image group 2. Difficult image data 2 ′ is extracted, and a discrimination surface 3 is generated based on the image data 2 ′. Since SVM itself is a well-known technique, details are omitted here (for example, Takashi Onoda: Support Vector Machine, Ohm Corporation, 2007).

  In the present embodiment, the discriminant plane generation unit 11e discriminates on the feature space an image that is difficult to discriminate whether it is an early label or an appropriate label, and whether it is an appropriate label or a late label. An image that is difficult to detect and an image that is difficult to determine whether it is a late label or an early label are extracted.

  Then, the discriminant plane generating unit 11e generates an early-time-appropriate discriminant plane in the middle between an image that is difficult to discriminate as an early label and an image that is difficult to discriminate as an appropriate label on the feature space, It is possible to generate a time-delay discriminating surface in the middle between an image that is difficult to discriminate as a timely label and an image that is difficult to discriminate as a late label, and further discriminate that it is a late label. A late-early discrimination plane is generated in the middle between a difficult image and an image that is difficult to discriminate as an early label.

  Then, the discriminant plane generation unit 11e stores parameters representing the generated three discriminant planes in the memory 15.

  The process of the discriminant plane generation step (S1) is performed only once for an evaluation object group similar to the determination implementation object. For example, in the case of the present embodiment, when the same hot dip galvanizing is applied to the steel material similar to the steel material of the transmission tower that is the judgment implementation object, it is assumed that it is the same evaluation object, and any of the power transmission towers Regardless of the location, the generation of the discriminating surface for determining the degree of corrosion of the steel material is performed only once for a steel material group similar to the steel material as the judgment object.

  Then, the following determination step (S2) is performed using the determination surface generated by the determination surface generation step (S1). When the discrimination surface generated by the processing of the discrimination surface generation step (S1) is used in the continuous inspection work of the determination object belonging to the class of the evaluation object of the case image data that is the source of the determination surface generation Alternatively, the parameter representing the discrimination surface may be stored in the storage unit 12 or the data server 16 as storage means.

  Next, the determination step (S2) will be described. The process of this determination step (S2) is a process used in the inspection work for determining the necessity of repair or replacement of steel parts of equipment in service, and is generated in the process of the determination surface generation step (S1). The determined degree of corrosion is used to determine the degree of corrosion of the steel parts of the equipment in service as a determination object.

  As a determination step (S2), first, the determination image data reading unit 11f of the control unit 11 reads the image data of the determination implementation (S2-1).

  The image data of the determination implementation (hereinafter referred to as determination image data) is data of an image obtained by photographing the determination implementation with a digital camera when the determination implementation is inspected. The determination image data may be captured using an automatic exposure setting function and an automatic white balance setting function provided in the digital camera, as in the case image data. Further, like the case image data, the determination image data may be one in which the illumination condition changes each time photographing is performed, or may be photographed using a telephoto lens at a position away from the evaluation target.

  As for the determination image data, as in the case image data, a determination execution part (a steel material part in the present embodiment) in a captured actual image is cut into a square. In other words, the image data cut out in a square shape from the photographed actual image is accumulated in the determination image database 19 as the determination image data. Further, the process of cutting out image data and processing such as transformation when necessary are the same as in the case of case image data.

  In the present embodiment, the determination image data related to the determination object is stored as a determination image database 19 in the data server 16 serving as a storage unit. The determination image database 19 may be stored in the storage unit 12 that is a storage unit.

  The determination image data reading unit 11 f reads the determination image data of the determination object from the determination image database 19 stored in the data server 16. Then, the determination image data reading unit 11 f stores the read determination image data in the memory 15.

  Next, the determination image processing unit 11g of the control unit 11 performs color processing on the determination image of the determination implementation (S2-2).

  Color processing (specifically, change in color temperature and change in brightness) for the image is the same as the content of the process in S1-2. That is, the determination image processing unit 11g first reads determination image data (referred to as an original determination image in the process of S2-2) stored in the memory 15 from the memory 15 in the process of S2-1.

  Then, the determination image processing unit 11g generates an image with an increased color temperature and a decreased image with respect to the original determination image, and also generates an image with a lighter brightness and a darkened image with respect to the original determination image. To do.

  In the present embodiment, a total of eight types of images (referred to as processing determination images) are created in addition to the original determination image by combining three types of hue and three types of brightness as in the process of S1-2. This will be described below.

  The determination image processing unit 11g creates eight types of processing determination images for each of the determination image data (that is, the original determination image) stored in the memory 15 in the process of S2-1.

  Then, the determination image processing unit 11g stores the processing determination image data in the memory 15 in association with the original original determination image, for example, by assigning the same ID number.

  Next, the determination image feature amount extraction unit 11h of the control unit 11 extracts the image feature amounts of the determination image data of the determination implementation and the processing determination image data (S2-3).

  Extraction of the image feature amount of the determination image data and the processing determination image data is performed using the same image feature amount as in the processing of S1-3.

  The determination image feature amount extraction unit 11h reads the determination image data stored in the memory 15 in the process of S2-1 and the processing determination image data stored in the memory 15 in the process of S2-2 from the memory 15, and reads the read image. Image feature values are calculated for each data.

  Then, the determination image feature amount extraction unit 11h stores the image feature amount for each image data in the memory 15 in association with the image data that is the basis of the image feature amount calculation.

  Next, the corrosion degree determination part 11i of the control part 11 determines the corrosion degree of the determination implementation object (S2-4).

  Specifically, the corrosion degree determination unit 11i first reads parameters representing the three determination surfaces stored in the memory 15 from the memory 15 in the process of S1-5.

  Next, the corrosion degree determination unit 11i reads the image feature amount of the determination image data of the determination implementation stored in the memory 15 from the memory 15 in the process of S2-3.

  Here, the SVM is basically a technique for discriminating two classes, and it is necessary to devise in order to discriminate three classes (that is, an early label, an appropriate label, and a late label) as in this embodiment. It is. In the present embodiment, in the process of S1-5, three discriminant surfaces are generated: an early-appropriate time discriminant plane, an appropriate-term-late discriminant plane, and a late-early discriminant plane. Then, in the process of S2-4, these three discriminating surfaces are sequentially applied to the image of the determination object, and the majority of the class of the corrosion degree to which the image belongs is determined. Note that when all the discrimination results by the three discrimination surfaces are different, the determination is made based on the proximity to each class (that is, the proximity to the support vector constituting each class). (For details on the above processing, see, for example, Michiko Yamana et al .: Development of a method for judging the reuse of armbands using rust images, research report R05001, Central Research Institute of Electric Power Industry, 2005, and Michiko Yamana et al .: (Refer to the method for judging the reuse of metallurgy independent of the position of rust in the image, Research Report R05004, Central Research Institute of Electric Power Industry, 2006.)

  That is, the corrosion degree determination unit 11i expresses the determination image data read from the memory 15 on the feature space, class determination when the early-time determination surface is applied, and application of the time-delay determination surface Class determination and class determination when the late-early determination surface is applied are sequentially performed to determine a class of corrosion degree of the determination image.

  Further, the corrosion degree determination unit 11i is associated with the determination image data (that is, the original determination image) of the determination execution object among the image feature amounts for each processing determination image data stored in the memory 15 in the process of S2-3. The image feature amounts of the processed determination image data are sequentially read from the memory 15.

  Then, the corrosion degree determination unit 11i sequentially expresses the processing determination image data read from the memory 15 on the feature space and determines the corrosion degree class of the processing determination image by applying the three determination surfaces.

  Subsequently, the corrosion degree determination unit 11i uses the majority of the corrosion degree classes (in this embodiment, nine in total) for the determination image and the processing determination image related to one determination object to determine the corrosion degree of the determination object. Determine the class. When the distribution of the nine classes of corrosion degree is the same, the class of corrosion degree in the original judgment image is set as the class of corrosion degree of the judgment execution object.

  And the corrosion degree determination part 11i displays the label of the corrosion degree for every judgment implementation thing on the display part 14, and also memorize | stores the combination data of the judgment image and corrosion degree label for every judgment thing as a judgment result file 12 is stored.

  When a plurality of pieces of determination image data related to similar determination implementations to which the determination surface generated in the determination surface generation step (S1) can be applied are stored in the determination image database 19, the control unit 11 determines The processing from S2-1 to S2-4 is repeated for each image data.

  And the control part 11 complete | finishes the process of the corrosion degree evaluation program 17 (END).

  According to the corrosion degree evaluation method, apparatus, and program of the present invention having the above-described configuration, a determination surface based on an image group in which some ambiguity is added to white balance and exposure is generated and the degree of corrosion is determined. Therefore, it is possible to generate a discriminating surface that mitigates the effects of adjustment / conversion errors caused by the camera's automatic white balance setting function and automatic exposure setting function, and to determine the degree of corrosion. It is possible to accurately determine the degree of corrosion even if the image is taken from a point away from the object or the determination object and the illumination condition is not kept constant.

  In addition, although the above-mentioned form is an example of the suitable form of this invention, it is not limited to this, A various deformation | transformation implementation is possible in the range which does not deviate from the summary of this invention. For example, in this embodiment, in the color processing in the processing of S1-2 and S2-2, a processed image is generated by changing the color temperature and changing the brightness with respect to the original image. However, the present invention is not limited to this, and a processed image obtained by converting only the color temperature of the original image may be created, or a processed image converted by changing only the brightness of the original image. You may make it create.

  Further, in the present embodiment, the degree of color processing when creating a machining example image in the processing of S1-2 for use in generating a discrimination surface, the number of machining example images created, and the determination of the degree of corrosion of the judgment object However, the present invention is not limited to this, and the color processing in the processing of S1-2 is the same as the degree of color processing and the number of processing determination images to be created in the processing of S2-2. And the number of processed example images created may not coincide with the degree of color processing and the number of processed determination images created in step S2-2.

  In the present embodiment, a plurality of processing example images are created in the processing of S1-2, and a plurality of processing determination images are created in the processing of S2-2. However, the present invention is not limited to this. In the process of S1-2, while creating a plurality of processing example images for generating the discriminant surface, only the original determination image may be used for determining the corrosion degree without performing the process of S2-2. Alternatively, only the original case image is used to generate the discrimination surface without performing the process of S1-2, while a plurality of processing determination images are created for the determination of the corrosion degree in the process of S2-2. good.

  In the present embodiment, the operator appropriately determines the degree of color processing in S1-2 and S2-2 and the number of processed images to be created, or uses a cross-validation method. However, if there is a hardware sample with a known degree of corrosion, not limited to this, the sample may be used under a variety of possible lighting conditions (for example, direct light and backlight, clear and cloudy, day and evening). The degree of color processing and the number of processed images to be created may be determined in advance by checking the degree of white balance and exposure with a photographed image and taking the degree of the fluctuation into consideration.

  Regarding the determination of the degree of color processing and the number of processed images to be created in the processing of S1-2 and S2-2, if a location where corrosion has not occurred in the image can be specified, the color value of the location is used. The degree of color processing and the number of processed images to be created may be determined. Specifically, color processing is performed to change the color so that the white balance (ratio of R, G, B) is uniform at the locations where corrosion has not occurred. In addition, when it is possible to specify a plurality of locations where corrosion has not occurred, a plurality of processed images can be created by using each as a reference.

  An embodiment in which the corrosion degree evaluation method, apparatus and program of the present invention are applied to the determination of the corrosion degree of the steel material of an actual power transmission tower will be described with reference to FIG. In this embodiment, color processing (S1-2) is performed on the case image in the determination surface generation step (S1), and color processing is performed on the determination image in the determination step (S2). (S2-2) was not performed.

  In this example, 1645 steel tower rust images photographed using a digital camera were used as photographed images. In this embodiment, an image photographed using the automatic exposure setting function and the automatic white balance setting function provided in the digital camera is used. Furthermore, the image used in the present embodiment includes an image in which the illumination condition changes every time shooting is performed.

  And, considering that all the images have sufficient resolution to extract the image feature amount and it is not necessary to enlarge the smallest photographed steel tower steel material, it is necessary to enlarge the steel tower steel material from the photographed image. A portion of 320 vertical pixels × horizontal 320 pixels was cut out and used as example image data.

  Moreover, in this example, as a label for the degree of corrosion on the surface of the galvanized steel material, an early label in which hot dip galvanization remains sufficiently, in a state where the deterioration of galvanization partially reaches the alloy layer. Three labels were used: a label at a certain time, an alloy layer exposed on the entire surface, and a part of the iron base and a part of the rust from the part close to the iron base. And the expert gave the label of the corrosion degree to each of the 1645 rust images.

  In the present embodiment, in the color processing in the process of S1-2, the color temperature of the original case image is assumed to be 5000K, and the color temperature is converted to 4500K, 0.9 times the assumed color temperature of 5000K. An image and an image obtained by converting the color temperature to 1.1 times 5500K were created. If the RGB value at 5000K is (r, g, b), the value at 4500K is (1.149r, g, 0.940b), and the value at 5500K is (0.897r, g, 1. 043b).

  Furthermore, an image obtained by converting the original case image and the image obtained by converting the color temperature to a slightly darker brightness by 0.9 times and an image converted to a slightly brighter image by 1.1 times were created.

  That is, in the present embodiment, as shown in FIG. 4, a total of eight processing examples are obtained for one original case image by combining three types of hue (that is, color temperature) and three types of brightness (that is, exposure). Created an image.

  In the present embodiment, when extracting the image feature amount in the processing of S1-3, it is possible to generate a discrimination plane that is a feature amount that handles fine hues and that is sensitive to changes in white balance and exposure. Considering what can be considered, an HSV histogram was used as an image feature amount.

  In this embodiment, when the HSV histogram is used, the frequency distribution is calculated by dividing H into 96 stages, and the frequency distribution is calculated by dividing S and V into 48 stages. That is, 96 + 48 + 48 = 192-dimensional data. In addition, normalization was performed so that each component of the feature amount had an average of 0 and a variance of 1.

  In the present embodiment, first, a quarter of the whole is randomly extracted from the case image group of each label of the corrosion degree and used as determination image data, and all the remaining images are used as case image data. After generating the discriminant surface using the data, the label of the corrosion degree of the judgment image data was judged using the discriminant surface. In this example, 100 trials of the evaluation process of “random extraction → discrimination surface generation → determination” were performed.

  Here, the ratio (namely, m / n) of the number m of the images for which the corrosion degree evaluation method of the present invention has correctly judged the corrosion degree is determined with respect to the number n of the determination images of one quarter of the whole taken out first. The average of the determination accuracy in 100 trials was defined as the average determination accuracy.

  The evaluation process was performed for 100 trials, and the average determination accuracy was 83.8%. From this result, the present invention is a corrosion using the image data photographed using the automatic exposure setting function and the automatic white balance setting function provided in the digital camera and the image data in which the illumination condition is changed every time photographing is performed. It was confirmed that it was useful for determining the degree.

  In addition, it was confirmed that by using an HSV histogram as an image feature amount, a discrimination surface that takes into account variations in hue and exposure can be generated, which contributes to an improvement in the accuracy of determination of corrosion degree based on image data.

It is a flowchart explaining an example of embodiment of the corrosion degree evaluation method of this invention. It is a functional block diagram of the corrosion degree evaluation apparatus in the case of implementing the corrosion degree evaluation method of this embodiment using a program. It is a conceptual diagram explaining the production | generation of a discriminant surface. It is a figure explaining the relationship between the original example image in an Example, and a process example image.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Feature space 2 Case image and process case image data group 2 'Image data with which class distinction is difficult 3 Discrimination surface

Claims (9)

  Creating a case image of the evaluation object and label data of the corrosion degree for each case image, and creating data of a processing case image by changing the color temperature and / or brightness of the case image Adding the data of the corrosion degree label of the case image that is the basis of the machining case image to the data of the machining case image that is created and the data of the label of the corrosion degree of the case image and the case image And a step of generating a determination surface of the corrosion degree by a support vector machine using data of the processing example image and the corrosion degree label of the processing example image, and a step of creating determination image data of a determination object, And a step of determining the degree of corrosion of the determination image using the data of the determination image and the determination surface.   A step of creating a case image of the evaluation object and a label of the corrosion degree for each of the case images; and a surface for determining the degree of corrosion by a support vector machine using the data of the case image and the corrosion degree label of the case image A step of generating determination image data of a determination object, a step of generating data of a processing determination image by changing a color temperature and / or brightness of the determination image, and And determining the corrosion degree of the determination image using the determination image data, the processing determination image data, and the determination surface.   Creating a case image of the evaluation object and label data of the corrosion degree for each case image, and creating data of a processing case image by changing the color temperature and / or brightness of the case image Adding the data of the corrosion degree label of the case image that is the basis of the machining case image to the data of the machining case image that is created and the data of the label of the corrosion degree of the case image and the case image And a step of generating a determination surface of the corrosion degree by a support vector machine using data of the processing example image and the corrosion degree label of the processing example image, and a step of creating determination image data of a determination object, Changing the color temperature and / or brightness of the determination image to create processing determination image data; and the determination image data; and Corrosion evaluation method characterized by having a determining corrosion rate of the determination image by using the data as well as the determination surface of serial processing determination image.   Means for reading the case image of the evaluation object and the label of the corrosion degree for each case image, and creating the processing case image data by changing the color temperature and / or brightness of the case image. Means for attaching the data of the corrosion degree label of the case image that is the basis of the processing case image to the data of the processing case image created together with the data of the label of the corrosion degree of the case image and the case image, and Means for generating a corrosion degree determination surface by a support vector machine using data of the processing example image and the corrosion degree label of the processing example image; means for reading determination image data of a determination implementation; and the determination image And a means for judging the degree of corrosion of the judgment image using the data and the discrimination surface.   Means for reading the case image of the evaluation object and the corrosion degree label data for each case image, and using the support image machine to determine the corrosion degree discrimination surface using the case image and the corrosion degree label data of the case image Means for generating, means for reading determination image data of the determination object, means for changing the color temperature and / or brightness of the determination image to create processing determination image data, and the determination A corrosion degree evaluation apparatus comprising: image data, processing determination image data, and means for determining the corrosion degree of the determination image using the determination surface.   Means for reading the case image of the evaluation object and the label of the corrosion degree for each case image, and creating the processing case image data by changing the color temperature and / or brightness of the case image. Means for attaching the data of the corrosion degree label of the case image that is the basis of the processing case image to the data of the processing case image created together with the data of the label of the corrosion degree of the case image and the case image, and Means for generating a corrosion degree determination surface by a support vector machine using data of the processing example image and the corrosion degree label of the processing example image; means for reading determination image data of a determination implementation; and the determination image Means for changing the color temperature and / or brightness of the image to create processing determination image data, and determining the data of the determination image and the processing determination image. Data and corrosion evaluation device, characterized in that it comprises a means for determining the corrosion rate of the determination image by using the discrimination surface.   Processing of reading the case image of the evaluation object and the label of the corrosion degree for each case image from the storage means, and processing case image data by changing the color temperature and / or brightness of the case image , And the processing example image data that is the basis of the processed case image and the label of the corrosion level of the case image and the case image The processing example image and the processing example image and the corrosion degree label data of the processing example image are used to generate a corrosion degree discrimination surface by a support vector machine, and the judgment image data of the judgment object is read from the storage means. A process for causing a computer to perform a process and a process for determining the degree of corrosion of the determination image using the data of the determination image and the determination surface. Degree evaluation program.   The processing of reading the case image of the evaluation object and the corrosion degree label data for each case image from the storage means, and using the case image and the corrosion degree label data of the case image, the support vector machine Processing for generating a discrimination surface, processing for reading judgment image data of a judgment object from storage means, and creating data of a processing judgment image by changing color temperature and / or brightness of the judgment image A corrosion degree evaluation program that causes a computer to perform processing for determining the degree of corrosion of the determination image using the determination image data, the processing determination image data, and the determination surface.   Processing of reading the case image of the evaluation object and the label of the corrosion degree for each case image from the storage means, and processing case image data by changing the color temperature and / or brightness of the case image , And the processing example image data that is the basis of the processed case image and the label of the corrosion level of the case image and the case image The processing example image and the processing example image and the corrosion degree label data of the processing example image are used to generate a corrosion degree discrimination surface by a support vector machine, and the judgment image data of the judgment object is read from the storage means. Processing, processing for changing the color temperature and / or brightness of the determination image, or creating processing determination image data, and data of the determination image Fine the processing determination image data and corrosion evaluation program characterized by causing the process of determining corrosion rate of the determination image on a computer by using the discrimination surface of.