Estimation device

Abstract

An estimation device (10) according to the present disclosure is provided with: an extraction unit (12) for extracting a corrosion image, which is an image of a corrosion portion, from a captured image obtained by imaging the inside of a metal pipe line to be inspected; an estimation model training unit (13) for generating an estimation model that is trained using a data set including a pair of a training image obtained by imaging a corrosion portion inside the metal pipe line and an actual measurement of the thickness reduction amount of the metal pipe line in the corrosion portion so as to estimate the thickness reduction amount in a corrosion portion inside a metal pipe line to be inspected from the captured image obtained by imaging the inside of the metal pipe line to be inspected; and an estimation unit (14) for inputting the corrosion image into the estimation model and estimating the thickness reduction amount in the corrosion portion corresponding to the corrosion image.

Description

estimation device

[0001] The present disclosure relates to an estimation device.

[0002] Metallic pipelines, such as steel pipelines, experience a decrease in the thickness of the metal material as corrosion progresses in the thickness direction of the pipeline, resulting in a decrease in structural strength. For the maintenance and management of metallic pipelines, it is desirable to assess the structural strength of the metal pipeline and plan repairs based on an estimate of future structural strength. However, currently, only the occurrence of corrosion is confirmed from images of the inside of the metal pipeline, and repairs are carried out according to the degree of corrosion. Economical and reliable maintenance and management based on a structural strength assessment requires an understanding of the wall thickness of the entire metal pipeline.

[0003] Non-Patent Document 1 describes a method for measuring the amount of wall thinning of a metal pipeline due to corrosion using an ultrasonic measuring instrument, etc. In the method described in Non-Patent Document 1, a transmitting and receiving probe for transmitting and receiving ultrasonic waves is brought into contact with the inside of the metal pipeline, and the wall thickness of the metal pipeline is measured based on the propagation time from when the ultrasonic waves are output from the transmitting and receiving probe to when the reflected waves are reflected by the outer surface of the metal pipeline and received by the transmitting and receiving probe.

[0004] Jun Murakoshi et al., "Study on the applicability of various measurement techniques to measuring the remaining thickness of corroded steel members", Journal of Structural Engineering, Vol. 59A, p711-724

[0005] The method described in Non-Patent Document 1 requires that a transmitting and receiving probe be brought into contact with the inspection location. In order to inspect a metal pipeline buried underground, the method described in Non-Patent Document 1 requires that a transmitting and receiving probe be inserted into the metal pipeline through a manhole to which the pipeline is connected and that the transmitting and receiving probe be brought into contact with the inspection location. Therefore, the method described in Non-Patent Document 1 is only capable of inspecting the metal pipeline near the manhole, and it is difficult to inspect the entire metal pipeline.

[0006] The present disclosure has been made in consideration of the above-described problems, and aims to provide an estimation device that can estimate the amount of wall thinning of the entire metal pipeline.

[0007] In order to solve the above-mentioned problems, the estimation device according to the present disclosure is an estimation device that estimates the amount of wall thinning of a metal pipeline, and includes: an extraction unit that extracts corrosion images that are images of corroded portions from images taken of the inside of the metal pipeline to be inspected; an estimation model learning unit that generates an estimation model that estimates the amount of wall thinning in corroded portions inside the metal pipeline to be inspected from the images taken of the inside of the metal pipeline to be inspected by learning a dataset that pairs training images of corroded portions inside the metal pipeline with actual measured values ​​of the amount of wall thinning of the metal pipeline in the corroded portions; and an estimation unit that inputs the corrosion images into the estimation model and estimates the amount of wall thinning in the corroded portions that correspond to the corrosion images.

[0008] According to the estimation device of the present disclosure, it is possible to estimate the amount of thinning of the entire metal pipeline.

[0009] 1 is a diagram illustrating an example of the configuration of an estimation device according to a first embodiment of the present disclosure. FIG. 1 is a diagram illustrating an example of the configuration of an extraction unit shown in FIG. 1. FIG. 2 is a diagram for explaining the operation of a cutout unit and a coordinate storage unit shown in FIG. 2. FIG. 3 is a diagram for explaining an estimation model dataset input to the estimation model learning unit shown in FIG. 1. FIG. 1 is a diagram illustrating an example of the configuration of the estimation model learning unit shown in FIG. 1. FIG. 1 is a diagram illustrating an example of the configuration of an output unit shown in FIG. 1. FIG. 1 is a diagram illustrating an example of output from the output unit shown in FIG. 1. A flowchart illustrating an example of the operation of the estimation device shown in FIG. 1. FIG. 1 is a diagram illustrating an example of the configuration of an estimation device according to a second embodiment of the present disclosure. FIG. 9 is a diagram illustrating an example of the configuration of a development unit shown in FIG. 10. FIG. 11 is a diagram illustrating the operation of an exclusion unit shown in FIG. 10. FIG. 12 is a diagram illustrating the operation of a partition unit shown in FIG. 10. FIG. 13 is a diagram illustrating the operation of a development diagram creation unit shown in FIG. 10. FIG. 9 is a diagram illustrating an example of the configuration of a restoration unit shown in FIG. 9. FIG. 10 is a diagram illustrating an example of the configuration of an estimation device according to a third embodiment of the present disclosure. FIG. 15 is a diagram illustrating an example of the configuration of a classification model learning unit shown in FIG. 15. FIG. 16 is a diagram illustrating the operation of a division unit shown in FIG. 17. FIG. 17 is a diagram illustrating an example of output from a result output unit shown in FIG. 17. FIG. 18 is a diagram illustrating an example of the configuration of a computer functioning as an estimation device according to the present disclosure.

[0010] Hereinafter, embodiments of the present disclosure will be described with reference to the drawings.

[0011] 1 is a diagram illustrating an example of the configuration of an estimation device 10 according to a first embodiment of the present disclosure. The estimation device 10 according to the present disclosure estimates the amount of wall thinning of a metal pipeline such as a steel pipeline.

[0012] As shown in FIG. 1 , an estimation device 10 according to this embodiment includes an input unit 11 , an extraction unit 12 , an estimation model learning unit 13 , an estimation unit 14 , and an output unit 15 .

[0013] An image of the interior of a metal pipeline to be inspected, captured by a pipe camera, is input to the input unit 11. The pipe camera is a camera that can be inserted into the pipeline. By moving the pipe camera axially inside the metal pipeline, it is possible to capture an image of any location inside the metal pipeline. The captured image is an RGB image with a width of w pixels and a height of h pixels (e.g., 1080 pixels x 720 pixels). The input unit 11 outputs the input captured image to the extraction unit 12.

[0014] The extraction unit 12 extracts corrosion images, which are images of corroded areas such as rust, from the captured images of the inside of the metal pipeline to be inspected that are output from the input unit 11. The extraction unit 12 outputs the extracted corrosion images to the estimation unit 14.

[0015] 2 is a diagram showing an example of the configuration of the extraction unit 12. As shown in FIG. 2, the extraction unit 12 includes a cutout unit 121 and a coordinate storage unit 122.

[0016] As shown in Fig. 3, the cutout unit 121 extracts an image of a corroded portion inside the metal pipe to be inspected (corrosion image) of a pre-specified size from a captured image of the inside of the metal pipe to be inspected. The cutout unit 121 detects the corroded portion using, for example, any image recognition technology and extracts the corrosion image. The corrosion image is preferably an image of a small rectangular area that can fit a rust bump within the image.

[0017] By extracting a corrosion image from a captured image, an image can be created in which only specific corrosion (e.g., rust) is extracted. This eliminates the need to learn unnecessary image information other than the specific corrosion (e.g., rust) in the learning of an estimation model by the estimation model learning unit 13 (described later), enabling learning of a highly accurate estimation model. Furthermore, by extracting a corrosion image of a predetermined size, a corrosion image of a fixed size can be extracted regardless of the resolution of the imaging device (pipe camera) or the aspect ratio of the captured image.

[0018] The coordinate storage unit 122 stores the coordinates of the corrosion image in the captured image (in FIG. 3 , the coordinates of four points of the corrosion image indicated by black circles: upper left (x1, y1), upper right (x2, y2), lower left (x3, y3), and lower right (x4, y4)). The coordinate storage unit 122 outputs the corrosion image and the coordinates of the corrosion image to the estimation unit 14.

[0019] 1 again, an estimation model dataset is input to the estimation model training unit 13. As shown in Fig. 4, the estimation model dataset is a set of pairs of training images of corroded portions inside a metal pipeline and actual measured values ​​of the amount of wall thinning of the metal pipeline at the corroded portions. The estimation model training unit 13 learns from the dataset, which pairs of training images of corroded portions inside a metal pipeline and actual measured values ​​of the amount of wall thinning of the metal pipeline at the corroded portions, to generate an estimation model that estimates the amount of wall thinning at the corroded portions inside the metal pipeline to be inspected, from the images of the inside of the metal pipeline to be inspected.

[0020] Fig. 5 is a diagram showing an example of the configuration of the estimation model learning unit 13. As shown in Fig. 5, the estimation model learning unit 13 includes a data set input unit 131, an allocating unit 132, a metal-loss amount learning unit 133, and an evaluation unit 134.

[0021] The dataset input unit 131 receives the estimation model dataset described with reference to Fig. 4. The learning images included in the estimation model dataset are images of corroded portions of a predetermined size cut out from images of the interior of a metal pipeline, similar to the corrosion images described above. The dataset input unit 131 outputs the input estimation model dataset to the sorting unit 132.

[0022] The allocating unit 132 allocates the input estimation model dataset into a training dataset and an evaluation dataset. The allocating unit 132 allocates the estimation model dataset so that the proportion of the training dataset is greater than the proportion of the evaluation dataset (e.g., 80% training dataset, 20% evaluation dataset). The allocating unit 132 outputs the training dataset to the wall-loss learning unit 133 and outputs the evaluation dataset to the evaluating unit 134.

[0023] The wall-thinning amount learning unit 133 receives as input the learning images included in the learning data set output from the sorting unit 132, performs learning using the wall-thinning amounts paired with the learning images as correct answers, and generates an estimation model. Learning is preferably performed using a convolutional neural network (CNN) as a model structure. The wall-thinning amount learning unit 133 outputs the estimation model generated by learning to the evaluation unit 134.

[0024] The evaluation unit 134 evaluates the estimation model generated by the wall-thinning amount learning unit 133 using an evaluation dataset and an arbitrary evaluation function. For example, the evaluation unit 134 inputs a learning image included in the evaluation dataset into the estimation model, and evaluates, using an arbitrary evaluation function, an error between an estimated value of wall-thinning amount by the estimation model for the input and an actual measured value of wall-thinning amount paired with the learning image. For example, if the evaluation unit 134 evaluates that the estimation model has a certain level of accuracy or higher, it outputs the estimation model to the estimation unit 14.

[0025] By dividing the data set for the estimation model into a training data set and an evaluation data set, the independence of training and evaluation is maintained, making it possible to generate a highly accurate estimation model.

[0026] Referring back to FIG. 1 , the estimation unit 14 receives the corrosion image and its coordinates from the extraction unit 12. The estimation unit 14 also receives the estimation model from the estimation model learning unit 13. The estimation unit 14 inputs the corrosion image into the estimation model and estimates the amount of metal loss in the corroded portion corresponding to the corrosion image. The estimation unit 14 stores a set of the corrosion image input into the estimation model and an estimate of the amount of metal loss calculated by the estimation model in response to the input of the corrosion image. The estimation unit 14 outputs the corrosion image, the coordinates of the corrosion image, and the amount of metal loss estimated from the corrosion image to the output unit 15.

[0027] The output unit 15 receives the corrosion image, the coordinates of the corrosion image, and the amount of metal loss estimated from the corrosion image from the estimation unit 14. The output unit 15 outputs the result of the estimation of the amount of metal loss by the estimation unit 14. For example, the output unit 15 outputs the estimated amount of metal loss by superimposing it on the captured image.

[0028] 6 is a diagram showing an example of the configuration of the output unit 15. As shown in FIG. 6, the output unit 15 includes a combining unit 151 and a superimposing unit 152.

[0029] The combining unit 151 combines the corrosion image with the photographed image from which it was extracted, based on the coordinates of the corrosion image used to estimate the amount of metal loss, to restore the original photographed image.

[0030] The superimposing unit 152 superimposes the estimated amount of metal loss on the photographed image restored by the combining unit 151 and outputs the result. For example, as shown in Fig. 7 , the superimposing unit 152 superimposes the position of the corrosion image and the amount of metal loss estimated from the corrosion image on the photographed image and outputs the result.

[0031] Next, the operation of the estimation device 10 according to this embodiment will be described. Fig. 8 is a flowchart showing an example of the operation of the estimation device 10 according to this embodiment, and is a diagram for explaining the estimation method executed by the estimation device 10 according to this embodiment.

[0032] The extraction unit 12 extracts a corrosion image, which is an image of a corroded portion, from a captured image of the inside of a metal pipeline to be inspected (step S11).

[0033] The estimation model learning unit 13 generates an estimation model that estimates the amount of thinning in the corroded parts inside the metal pipeline to be inspected from the images taken inside the metal pipeline to be inspected by learning a data set that pairs training images of the corroded parts inside the metal pipeline with actual measured values ​​of the amount of thinning in the metal pipeline in the corroded parts (step S12).

[0034] The estimation unit 14 inputs the corrosion image into an estimation model and estimates the amount of thinning in the corroded portion corresponding to the corrosion image (step S13).

[0035] As described above, the estimation device 10 according to this embodiment includes an extraction unit 12, an estimation model learning unit 13, and an estimation unit 14. The extraction unit 12 extracts corrosion images, which are images of corroded portions, from captured images of the interior of a metal pipeline to be inspected. The estimation model learning unit 13 generates an estimation model that estimates the amount of wall thinning in a corroded portion inside the metal pipeline to be inspected from the captured images of the interior of the metal pipeline to be inspected by learning a data set that pairs training images of corroded portions inside the metal pipeline with actual measured values ​​of the amount of wall thinning of the metal pipeline in the corroded portions. The estimation unit 14 inputs the corrosion images into the estimation model and estimates the amount of wall thinning in the corroded portion corresponding to the corrosion image.

[0036] The estimation device 10 according to this embodiment can estimate the amount of thinning of a corroded portion from an image of the interior of a metal pipeline to be inspected. Images of any position inside the metal pipeline can be captured using, for example, a pipe camera. Therefore, the estimation device 10 according to this embodiment can estimate the amount of thinning of the entire metal pipeline.

[0037] Second Embodiment Fig. 9 is a diagram illustrating an example of the configuration of an estimation device 10A according to a second embodiment of the present disclosure. In Fig. 9, the same components as those in Fig. 1 are denoted by the same reference numerals, and description thereof will be omitted.

[0038] 9, the estimation device 10A according to this embodiment includes an input unit 11, an extraction unit 12, an estimation model learning unit 13, an estimation unit 14, an output unit 15, an expansion unit 16, and a restoration unit 17. The estimation device 10A according to this embodiment differs from the estimation device 10 shown in FIG. 1 in that the expansion unit 16 and the restoration unit 17 are added.

[0039] The expansion unit 16 is provided between the input unit 11 and the extraction unit 12. The expansion unit 16 divides the captured image input to the input unit 11 into a plurality of transformation regions, and converts each of the plurality of transformation regions into a rectangular region with the same number of pixels by projective transformation, thereby generating a development of the captured image.

[0040] Fig. 10 is a diagram showing an example of the configuration of the development unit 16. As shown in Fig. 10, the development unit 16 includes a compression unit 161, an exclusion unit 162, a division unit 163, and a development drawing creation unit 164.

[0041] The compression unit 161 resizes the photographed image input from the input unit 11 to an image of a predetermined square size (for example, 480 pixels × 480 pixels) to facilitate the creation of a development diagram, which will be described later. The compression unit 161 outputs the resized photographed image to the exclusion unit 162.

[0042] The exclusion unit 162 excludes portions (exclusion portions) that are not suitable for estimating the amount of wall thinning from the resized captured image output from the compression unit 161. The exclusion portions are, for example, hollow portions and portions with low pixel resolution near the center of the captured image, as shown in Fig. 11. The exclusion unit 162 outputs the captured image after the exclusion portions have been excluded to the separator unit 163.

[0043] By excluding parts that are not suitable for estimating the amount of metal thinning from the captured image, it is possible to limit the area from which corrosion images are extracted, which makes it possible to more efficiently train the estimation model and estimate the amount of metal thinning using the estimation model.

[0044] The separator 163 separates the captured image after removing the exclusion portion output from the exclusion unit 162 into multiple transformation regions, as shown in Fig. 12. Fig. 12 shows an example in which the captured image is divided into eight transformation regions counterclockwise, centered on the exclusion portion near the center of the captured image. The separator 163 saves the coordinates of the vertices of each of the multiple transformation regions. The separator 163 outputs the image of each of the multiple transformation regions and the coordinates of the vertices of the transformation region to the development drawing creation unit 164.

[0045] The development drawing creation unit 164 performs projective transformation on each of the images of the multiple conversion regions output from the delimiter 163 into rectangular images with the same number of pixels. Based on the coordinates of the vertices of each of the multiple conversion regions, the development drawing creation unit 164 then creates a single development drawing by joining together the images of the multiple conversion regions after projective transformation, as shown in FIG. 13 . The development drawing creation unit 164 outputs the created development drawing to the extraction unit 12. The extraction unit 12 extracts a corrosion image from the created development drawing. In this embodiment, images of the corroded portion in the development drawing of the captured image are also used as learning images included in the estimation model dataset input to the estimation model learning unit 13.

[0046] When extracting a rectangular corrosion image from an image captured by a pipe camera, the actual size of the metal pipeline wall surface shown in the corrosion image varies depending on the position of the extracted corrosion image due to differences in pixel resolution. Specifically, even if the size on the captured image is the same, the actual size of the metal pipeline wall surface shown in the captured image increases the closer it is to the front. By performing projective transformation on each of the multiple transformation regions to create a development, the area of ​​the corrosion image is corrected, making it possible to display the interior of the metal pipeline at a scale close to the actual corrosion area.

[0047] Referring again to FIG. 9 , the restoration unit 17 is provided between the estimation unit 14 and the output unit 15. The restoration unit 17 receives the corrosion image, the coordinates of the corrosion image, and the amount of metal loss estimated from the corrosion image from the estimation unit 14. The restoration unit 17 inversely transforms each of the multiple transformed regions after the projective transformation by the development unit 16 back to their original shapes, thereby restoring the photographed image from the development. Specifically, the restoration unit 17 restores the original development from the development from which the corrosion image was extracted and the corrosion image extracted from the development by the extraction unit 12. The restoration unit 17 then inversely transforms each of the multiple transformed regions constituting the restored development back to their original shapes, thereby restoring the photographed image from the development.

[0048] Fig. 14 is a diagram showing an example of the configuration of the restoration unit 17. As shown in Fig. 14, the restoration unit 17 includes a development restoration unit 171 and an original image restoration unit 172.

[0049] The development restoration unit 171 restores the original development from the development from which the corrosion image has been extracted and the extracted corrosion image. Specifically, the development restoration unit 171 restores the development by returning the corrosion image extracted by the extraction unit 12 to a position corresponding to the corrosion image in the development from which the corrosion image was extracted. The development restoration unit 171 outputs the restored development to the original image restoration unit 172.

[0050] The original image restoration unit 172 receives the restored development from the development restoration unit 171. The original image restoration unit 172 restores the original captured image from the input development. Specifically, the original image restoration unit 172 inversely transforms each of the multiple transformed regions that make up the development to its original shape, and restores the original captured image by combining the transformed regions after the inverse transformation based on the coordinates of the vertices of each of the multiple transformed regions. The original image restoration unit 172 outputs the restored original captured image to the output unit 15.

[0051] The output unit 15 superimposes the amount of metal loss estimated by the estimation unit 14 on the restored photographed image and outputs it. Here, in the first embodiment, the output unit 15 was equipped with a combining unit 151 that combines the extracted corrosion image with the photographed image from which the corrosion image was extracted based on the coordinates of the corrosion image to restore the original photographed image. In this embodiment, the restoration unit 17 (original image restoration unit 172) restores the original photographed image from the development view and inputs it to the output unit 15. Therefore, in this embodiment, the output unit 15 does not need to be equipped with the combining unit 151.

[0052] As described above, in this embodiment, the estimation device 10A includes an expansion unit 16 and a restoration unit 17. The expansion unit 16 divides the captured image into multiple transformation regions and converts each of the multiple transformation regions into rectangular regions with the same number of pixels using projective transformation to generate a development of the captured image. The extraction unit 12 extracts the corrosion image from the development. The restoration unit 17 restores the original development from the development from which the corrosion image has been extracted and the extracted corrosion image, and then restores the captured image from the development by inversely transforming each of the multiple transformation regions constituting the restored development to their original shapes. The output unit 15 superimposes the estimated amount of metal loss on the restored captured image and outputs it.

[0053] By dividing the captured image into multiple transformation regions and then performing projective transformation on the multiple transformation regions to create a development, the pixel resolution becomes uniform, making it possible to extract images of uniform corrosion on the metal pipe wall regardless of position. As a result, the accuracy of the learning of the estimation model and the estimation of the amount of wall thinning can be improved.

[0054] 15 is a diagram illustrating an example of the configuration of an estimation device 10B according to a third embodiment of the present disclosure. In Fig. 15, the same components as those in Figs. 1 and 9 are denoted by the same reference numerals, and description thereof will be omitted.

[0055] 15, the estimation device 10B according to this embodiment includes an input unit 11, an extraction unit 12, an estimation model learning unit 13, an estimation unit 14, an output unit 15, an expansion unit 16, a restoration unit 17, a classification model learning unit 18, and a detection unit 19. The estimation device 10B according to this embodiment differs from the estimation device 10A shown in FIG. 9 in that the classification model learning unit 18 and the detection unit 19 are added.

[0056] A classification model dataset is input to the classification model learning unit 18. The classification model dataset is a dataset that classifies images of corroded areas of metal pipelines and images of non-corroded areas of metal pipelines. A corroded area is an area where corrosion of a metal pipeline is visible, such as an area where rust fluid is visible, an area where minute granular rust is visible, and an area where rust with unevenness is visible. A non-corroded area is an area where corrosion is not visible due to dirt on the surface of a metal pipeline or an area where no corrosion exists. Examples of areas where corrosion is not visible include flooded areas and areas with muddy stains. Images included in the classification model dataset are assigned a class name that indicates whether the image is an image of a corroded area or a non-corroded area.

[0057] The classification model learning unit 18 generates a classification model that classifies each specified area of ​​a photographed image of the inside of a metal pipeline to be inspected as either a corroded area or a non-corroded area by learning from the input classification model dataset (a dataset that classifies images of corroded areas and images of non-corroded areas).

[0058] Fig. 16 is a diagram showing an example of the configuration of the classification model learning unit 18. As shown in Fig. 16 , the classification model learning unit 18 includes a dataset input unit 181, an allocating unit 182, a corrosion region learning unit 183, and an evaluation unit 184.

[0059] The above-described classification model dataset is input to the dataset input unit 181. The dataset input unit 181 outputs the input classification model dataset to the sorting unit 182.

[0060] The allocating unit 182 allocates the input classification model dataset into a training dataset and an evaluation dataset. The allocating unit 182 allocates the input classification model dataset so that the proportion of the training dataset is greater than the proportion of the evaluation dataset (e.g., 80% training dataset, 20% evaluation dataset). The allocating unit 182 outputs the training dataset to the corrosion region learning unit 183 and outputs the evaluation dataset to the evaluation unit 184.

[0061] The corrosion region learning unit 183 receives as input the images included in the learning dataset received from the sorting unit 182, performs learning using the labels assigned to the images as correct answers, and generates a classification model. It is desirable to perform the learning using a CNN as the model structure. The corrosion region learning unit 183 outputs the classification model generated by learning to the evaluation unit 184.

[0062] The evaluation unit 184 evaluates the classification model generated by the corrosion region learning unit 183 using an evaluation dataset and an arbitrary evaluation function. Specifically, the evaluation unit 184 inputs images included in the evaluation dataset into the classification model, and evaluates, using an arbitrary evaluation function, whether there is a match or mismatch between the classification result of the classification model for the input and the label assigned to the input image. If the evaluation unit 184 evaluates that the classification model has an accuracy equal to or higher than a certain level, for example, it outputs the classification model to the detection unit 19.

[0063] By dividing the classification model dataset into a training dataset and an evaluation dataset, the independence of training and evaluation is maintained, making it possible to generate a highly accurate classification model.

[0064] Referring again to FIG. 15 , the detection unit 19 is provided between the input unit 11 and the expansion unit 16. The detection unit 19 receives a captured image of the interior of the metal pipeline to be inspected from the input unit 11. The detection unit 19 also receives a classification model from the classification model learning unit 18. The detection unit 19 divides the captured image into a plurality of divided regions and inputs them into the classification model, and detects for each divided region whether the divided region is a corroded region or a non-corroded region. The detection unit 19 outputs the detection result to the expansion unit 16.

[0065] 17 is a diagram showing an example of the configuration of the detection unit 19. As shown in FIG. 17, the detection unit 19 includes a division unit 191, a classification unit 192, and a result output unit 193.

[0066] The dividing unit 191 receives a captured image from the input unit 11. As shown in Fig. 18 , the dividing unit 191 divides the input captured image into a plurality of divided regions of the same area (9 divisions in the example shown in Fig. 18 ).

[0067] The classification unit 192 inputs the image of the divided region into a classification model for each divided region divided by the division unit 191, and divides the input image into an image of a corroded region and an image of a non-corrosive region. The classification unit 192 outputs the classification result to the result output unit 193.

[0068] 19 , the result output unit 193 superimposes the classification results of each of the multiple divided regions by the classification unit 192, indicating whether the region is a corroded region or a non-corroded region, on the original captured image and outputs the results to the unfolding unit 16. As described above, the unfolding unit 16 unfolds the captured image into a development and outputs the development to the extraction unit 12. The extraction unit 12 extracts corrosion images from the corroded regions in the development. That is, the extraction unit 12 extracts corrosion images from regions in the captured image that have been detected as corroded regions.

[0069] In the present embodiment, the estimation device 10B has been described using an example in which the estimation device 10A according to the second embodiment has been configured by adding the classification model learning unit 18 and the detection unit 19, but this is not limiting. The estimation device 10B may have a configuration in which the classification model learning unit 18 and the detection unit 19 are added to the estimation device 10 according to the first embodiment. In this case, the detection unit 19 is provided between the input unit 11 and the extraction unit 12.

[0070] As described above, the estimation device 10B according to this embodiment includes a classification model training unit 18 and a detection unit 19. The classification model training unit 18 generates a classification model that classifies each predetermined region of a photographed image of the interior of a metal pipeline to be inspected as a corroded region or a non-corroded region by training a data set that classifies images of corroded regions of the metal pipeline and images of non-corroded regions of the metal pipeline. The detection unit 19 divides the photographed image into a plurality of divided regions and inputs the divided regions into the classification model, and detects, for each divided region, whether the divided region is a corroded region or a non-corroded region. The extraction unit 12 extracts corrosion images from regions detected as corroded regions in the photographed image.

[0071] By detecting corroded areas and extracting corrosion images from the detected corroded areas, it is possible to limit the area where the amount of metal loss is estimated (by excluding non-corroded areas), thereby speeding up the estimation of the amount of metal loss.

[0072] The above-described estimation devices 10, 10A, and 10B can be realized by a computer 20 shown in FIG. 20. A program for causing the computer 20 to function as the estimation devices 10, 10A, and 10B may be provided. The program may be stored in a storage medium or provided via a network. FIG. 20 is a block diagram showing a schematic configuration of the computer 20 functioning as the estimation devices 10, 10A, and 10B. The computer 20 may be a general-purpose computer, a dedicated computer, a workstation, a PC (Personal Computer), an electronic notepad, or the like. The program instructions may be program code, code segments, or the like for executing necessary tasks.

[0073] 20, a computer 20 includes a processor 21, a ROM (Read Only Memory) 22, a RAM (Random Access Memory) 23, a storage 24, an input unit 25, a display unit 26, and a communication interface (I / F) 27. Each component is communicably connected to one another via a bus 29. The processor 21 is specifically a CPU (Central Processing Unit), an MPU (Micro Processing Unit), a GPU (Graphics Processing Unit), a DSP (Digital Signal Processor), a SoC (System on a Chip), or the like, and may be configured with multiple processors of the same or different types.

[0074] The processor 21 is a control unit that controls each component and performs various arithmetic processing. That is, the processor 21 reads a program from the ROM 22 or the storage 24 and executes the program using the RAM 23 as a work area. The processor 21 controls each component and performs various arithmetic processing in accordance with the program stored in the ROM 22 or the storage 24. In this embodiment, the ROM 22 or the storage 24 stores a program for operating the computer 20 as the estimation devices 10, 10A, and 10B according to the present disclosure. The processor 21 reads and executes the program to realize each component of the estimation devices 10, 10A, and 10B.

[0075] The program may be provided in a form stored on a non-transitory storage medium such as a CD-ROM (Compact Disk Read Only Memory), a DVD-ROM (Digital Versatile Disk Read Only Memory), a USB (Universal Serial Bus) memory, etc. The program may also be provided in a form downloaded from an external device via a network.

[0076] The ROM 22 stores various programs and various data. The RAM 23 temporarily stores programs or data as a working area. The storage 24 is configured with an HDD (Hard Disk Drive) or an SSD (Solid State Drive) and stores various programs including the operating system and various data.

[0077] The input unit 25 includes a pointing device such as a mouse and a keyboard, and is used to input various types of information.

[0078] The display unit 26 is, for example, a liquid crystal display, and displays various information. The display unit 26 may employ a touch panel system and function as the input unit 25. The display unit 26 displays, for example, the estimated result of the amount of wall thinning output by the output unit 15.

[0079] The communication interface 27 is an interface for communicating with other devices, for example, an interface for a LAN.

[0080] The following additional notes are provided regarding the above-described embodiments.

[0081] [Supplementary Item 1] An estimation device for estimating the amount of wall thinning of a metallic pipeline, comprising a control unit, wherein the control unit is configured to: extract corrosion images that are images of corroded portions from photographed images of the inside of the metallic pipeline to be inspected; generate an estimation model that estimates the amount of wall thinning in the corroded portions inside the metallic pipeline to be inspected from the photographed images of the inside of the metallic pipeline to be inspected by learning from a dataset that pairs training images of the corroded portions inside the metallic pipeline with actual measured values ​​of the amount of wall thinning of the metallic pipeline at the corroded portions; and input the corrosion images into the estimation model to estimate the amount of wall thinning in the corroded portions corresponding to the corrosion images.

[0082] [Supplementary Item 2] The estimation device according to Supplementary Item 1, wherein the control unit outputs the estimated amount of wall-thinning by superimposing it on the captured image.

[0083] [Supplementary Item 3] In the estimation device described in Supplementary Item 2, the control unit divides the captured image into a plurality of transformation regions, converts each of the plurality of transformation regions into a rectangular region with the same number of pixels by projective transformation, and generates a development of the captured image, extracts the corrosion image from the development, restores the original development from the development from which the corrosion image has been extracted and the extracted corrosion image, inversely transforms each of the plurality of transformation regions constituting the restored development to their original shapes, and restores the captured image from the development, and outputs the estimated amount of metal loss by superimposing it on the restored captured image.

[0084] [Supplementary Item 4] In the estimation device described in any one of Supplementary Items 1 to 3, the control unit generates a classification model that classifies each predetermined area of ​​a photographed image of the interior of the metal pipeline to be inspected as either a corroded area or a non-corroded area by learning from a data set that classifies images of corroded areas in a metal pipeline where corrosion is visible and images of non-corroded areas in a metal pipeline where corrosion is not visible or no corrosion is present, divides the photographed image into a plurality of divided areas and inputs them to the classification model, detects for each divided area whether the divided area is a corroded area or a non-corroded area, and extracts the corrosion image from the area detected as a corroded area in the photographed image.

[0085] [Supplementary Item 5] An estimation method executed by an estimation device that estimates the amount of wall thinning of a metallic pipeline, the estimation method comprising: extracting corrosion images that are images of corroded portions from photographed images of the inside of the metallic pipeline to be inspected; generating an estimation model that estimates the amount of wall thinning in the corroded portions inside the metallic pipeline to be inspected from the photographed images of the inside of the metallic pipeline to be inspected by learning from a dataset that pairs training images of the corroded portions inside the metallic pipeline with actual measured values ​​of the amount of wall thinning of the metallic pipeline in the corroded portions; and inputting the corrosion images into the estimation model to estimate the amount of wall thinning in the corroded portions corresponding to the corrosion images.

[0086] [Supplementary Item 6] A non-transitory storage medium storing a program executable by a computer, the non-transitory storage medium storing the program causing the computer to operate as the estimation device according to any one of Supplementary Items 1 to 4.

[0087] Although the above-described embodiments have been described as typical examples, it will be apparent to those skilled in the art that many modifications and substitutions can be made within the spirit and scope of the present disclosure. Therefore, the present invention should not be interpreted as being limited by the above-described embodiments, and various modifications and alterations are possible without departing from the scope of the claims. For example, multiple building blocks shown in the block diagrams of the embodiments can be combined into one, or one building block can be divided.

[0088] 10, 10A, 10B Estimation device 11 Input unit 12 Extraction unit 13 Estimation model learning unit 14 Estimation unit 15 Output unit 16 Expansion unit 17 Restoration unit 18 Classification model learning unit 19 Detection unit 121 Cutting unit 122 Coordinate storage unit 131 Data set input unit 132 Allocation unit 133 Thinning amount learning unit 134 Evaluation unit 151 Combination unit 152 Superposition unit 161 Compression unit 162 Exclusion unit 163 Separation unit 164 Development drawing creation unit 171 Development drawing restoration unit 172 Original image restoration unit 181 Data set input unit 182 Allocation unit 183 Thinning amount learning unit 184 Evaluation unit 191 Division unit 192 Classification unit 193 Result output unit 20 Computer 21 Processor 22 ROM 23 RAM 24 Storage 25 Input unit 26 Display unit 27 Communication I / F 29 Path

Claims

1. An estimation device for estimating the amount of thinning of a metal pipeline, comprising: an extraction unit that extracts corrosion images that are images of corroded portions from images taken of the inside of the metal pipeline to be inspected; an estimation model training unit that generates an estimation model that estimates the amount of thinning of corroded portions inside the metal pipeline to be inspected from the images taken of the inside of the metal pipeline to be inspected by training a data set that pairs training images of corroded portions inside the metal pipeline with actual measured values ​​of the amount of thinning of the metal pipeline at the corroded portions; and an estimation unit that inputs the corrosion images to the estimation model and estimates the amount of thinning of the corroded portions corresponding to the corrosion images.

2. The estimation device according to claim 1, further comprising an output unit that outputs the estimated amount of wall-thinning by superimposing it on the captured image.

3. An estimation device according to claim 2, further comprising: an expansion unit that divides the captured image into a plurality of transformation regions, converts each of the plurality of transformation regions into a rectangular region with the same number of pixels by projective transformation, and generates a development of the captured image; and a restoration unit that inversely transforms each of the plurality of transformation regions after projective transformation back to their original shapes and restores the captured image from the development, wherein the extraction unit extracts the corrosion image from the development, the restoration unit restores the original development from the development from which the corrosion image has been extracted and the extracted corrosion image, and inversely transforms each of the plurality of transformation regions constituting the restored development back to their original shapes, thereby restoring the captured image from the development, and the output unit superimposes the estimated amount of metal loss on the restored captured image and outputs it.

4. An estimation device according to claim 1, further comprising: a classification model learning unit that generates a classification model that classifies each predetermined area of ​​a photographed image of the interior of the metal pipeline to be inspected as either a corroded area or a non-corroded area by learning from a data set that classifies images of corroded areas in the metal pipeline where corrosion is visible and images of non-corroded areas in the metal pipeline where corrosion is not visible or no corrosion is present; and a detection unit that divides the photographed image into a plurality of divided areas and inputs them into the classification model, and detects for each divided area whether the divided area is a corroded area or a non-corroded area, wherein the extraction unit extracts the corrosion image from an area in the photographed image that is detected as a corroded area.

Images