Apparatus for diagnosing deterioration of conduit and method for the same

Abstract

To provide a method or an apparatus capable of more quickly and accurately finding abnormality on an image in a conduit using an artificial intelligence technique (AI).SOLUTION: A method for diagnosing deterioration of a conduit includes the steps of: storing an original development image; coupling numerical information to the original development image; generating an undamaged image; taking out the original development image information and the numerical information and creating a damaged image, and extracting an image region where an abnormal place of the original development image is inserted in the image; constructing a program for AI learning using the image and the numerical information as teacher data; constructing a program output as a learned AI determination model, on the image information whose damage is not registered, using the model for AI learning as input data; and outputting an AI determined development image identified and displayed by AI determination, using the learned AI determination model, with the original development image as input data.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present invention constructs a learning AI model from a large number of existing pipe channel inner wall information and numerical information attached to this pipe channel inner wall information, and uses this trained model to perform color-coded display and other identification displays according to the presence or absence of abnormal locations and the types of abnormal locations in a pipe channel image to be newly diagnosed, so as to accurately and quickly diagnose the deterioration of the pipe channel inner wall. The present invention relates to an apparatus and method for diagnosing the deterioration of a pipe channel.

Background Art

[0002] The present applicant has already proposed a storage medium storing a pipe channel internal inspection diagnosis support apparatus, a pipe channel internal inspection diagnosis support method, and a pipe channel internal inspection diagnosis support program (Patent Document 1). This is for performing on a personal computer the processing of an inner surface image of a pipe channel, such as displaying the state of a pipe channel as a video or a still image, creating a longitudinal development view, or inspecting and diagnosing an abnormal state of the inner wall surface of a pipe channel while displaying it as a video or a still image, based on video data obtained by photographing the inner wall surfaces of upper and lower water pipes, traffic tunnels such as railway, road, and sidewalk tunnels, cable buried pipes, and other large and small diameter pipe channels. This conventional invention enables registration of the position information of a point by checking on an image when an abnormal location is found in a video while continuously moving the image taken by a camera before development view processing, and enables recognition of the movement of marked points when abnormal locations are continuous.

[0003] A specific procedure for performing diagnosis using a conventional pipe channel internal inspection diagnosis support apparatus, a pipe channel internal inspection diagnosis support method, and a pipe channel internal inspection diagnosis support program will be described with reference to FIGS. 8 to 14. Figure 10 is a block diagram of a conventional device. In the memory 80, a system program 1, image data 87, and a table 88 are stored, and necessary data is exchanged with the CPU 81. The image data 87 is captured from pipeline video data 86 such as actual captured image data as shown in Fig. 14(a) and developed drawing data processed by the development drawing as shown in Fig. 14(b) into the memory 80 of the personal computer, and the diagnosis of the pipe channel is performed using these image data 87. The table 88 represents various table structures in the system program 1, and the captured image data is managed according to the data structure of this table 88.

[0004] An input device 83, a display device 84, and an output device 85 are connected to the CPU 81 via an input / output control unit 82. The input / output control unit 82 serves to send signals from the input device 83 to the CPU 81 and send the calculation results by the CPU 81 to the display device 84 and the output device 85. The input device 83 consists of a mouse, a keyboard, etc., the display device 84 consists of a personal computer display, a TV screen, etc., and the output device 85 consists of a printer, etc. The system program 1 is loaded into the memory 80, and program information is sent from this memory 80 to the CPU 81 to perform arithmetic processing and start up the system program 1. Thereafter, various processes such as image reproduction, image diagnosis, and input of diagnosis information are performed by the CPU 81 according to signals from the input device 83. Each time, the results are output to the display device 84 and the output device 85, and based on the output results, the diagnostician makes a comprehensive diagnosis of the pipe channel.

[0005] The system program 1 includes a pipeline inspection and diagnosis work support program 1 shown in FIG. 12. This pipeline inspection and diagnosis work support program 1 uses the results of on-site pipeline inspections (hereinafter referred to as pipeline video data) and the developed drawings (hereinafter referred to as diagnosis information) created by importing this pipeline video data into a personal computer to perform video inspection and diagnosis in the office and create inspection results (hereinafter referred to as forms). It consists of an input / output processing unit 2 for diagnosis data, a diagnosis support processing unit 3, an input processing unit 4 for diagnosis results, a display / print processing unit 5 for forms, an input processing unit 6 for form management information, and a display / print processing unit 7 for the diagnosis information list.

[0006] The display / print processing unit 5 for the form performs the display and printing of the form based on the diagnosis information created based on the diagnosis results. The input processing unit 6 for the form management information is for editing the inspection subject of the pipeline, the inspection location, the inspector, the pipeline information, the manhole information, etc. The display / print processing unit 7 for the diagnosis information list performs the display and printing of the list of diagnosed information.

[0007] A process of registering diagnosis information based on FIG. 8 will be described. For this purpose, the pipeline video data 86 in FIG. 10 is read as image data 87 into a memory 80 such as a hard disk of a personal computer. Also, the system program 1 is started. After starting the system program 1, the system program 1 opens the image data 87 of the pipeline video data 86 and displays the "Deck Controller" window (a in FIG. 8). By operating this "Deck Controller", the pipeline video data 86 is played and displayed (b in FIG. 8). In the "Deck Controller", the image frame information 1, 2,... m of the pipeline video data 86 shown in FIG. 9(a) is played and displayed one frame at a time, so that it can be continuously played as a moving image as if the camera 35 is moving forward or backward in the pipeline 14 shown in FIG. 11. More specifically, (1) Based on the image names in the image frame information shown in FIG. 9(a), the display image is displayed on the screen. (2) Based on the characteristics of the camera's mirror lens, the radius of the inspection pipeline, and the imaging distance within the image frame information, determine the distance range of the pipeline shown in the displayed image. (3) Compare the obtained distance range of the pipeline with the pipeline distance (e) stored in the image diagnosis information shown in FIG. 9(b), and display the content of the relevant image diagnosis information on the screen. In addition to videos, it is also possible to view each frame as a still image.

[0008] First, search for the image to be diagnosed and registered by playing it as a video. When a part that may be an abnormal location is found (step c in FIG. 8), switch to a still image and perform detailed diagnosis. When the abnormal location is clicked with the mouse button (step d in FIG. 8), create the image diagnosis information shown in FIG. 9(b), temporarily register the mouse position coordinates in the column (a) of FIG. 9(b), and the image frame number at that time is automatically stored in the column of FIG. 9(b) from the image frame information in FIG. 9(a). Also, a mark point 92 (× mark) is displayed on the screen (FIGS. 11(C), (D) and FIG. 13). Here, while there is an abnormal location, "Continue registration?" becomes YES (step e in FIG. 8), and return to step b in FIG. 8 and repeat.

[0009] Next, when "Continue registration?" becomes NO, open the diagnosis information window and move to the input of diagnosis information (step f in FIG. 8), and set the diagnosis information. In setting the diagnosis information, as shown in FIG. 8, perform "1. Selection of abnormal classification", "2. Selection of rank information", "3. Selection of abnormal situation", and "4. Input of detailed information". For "1. Abnormal classification", as shown in the registration of the abnormal classification, there are breakage, crack, joint displacement, corrosion, sagging and meandering, water intrusion, protruding attachment pipe, mortar adhesion, lard adhesion, root intrusion, step, attachment pipe right, attachment pipe left, and others, and select from among them. For "2. Rank information", it is handled as "Special A", "A", "B", "C", "Others". "3. Abnormal conditions" are registered as abnormal conditions, and can be selected from the following: attachment pipe, right attachment pipe, left attachment pipe, protruding attachment pipe, damage, crack, joint displacement, corrosion, sagging and meandering, water ingress, mortar adhesion, lard adhesion, root intrusion, step present. "4. Detailed information" is registered as comments on the points noticed regarding the abnormal location. Specific examples of the above 1, 2, and 3 are shown in the visual inspection criteria of Figures 6 and 7.

[0010] While the setting content of the diagnostic information is "no change point" and "NO", this is repeated. When "YES" is confirmed (step g in Figure 8), "select registration button" is selected (step h in Figure 8) to end the registration. When "select registration button" is selected, the abnormal location failure status and registration status are registered in (u) and (o) of Figure 9(b). The "pipe distance" of the registration point in (e) is obtained from the distance of the image frame number and the mouse position coordinates and registered. Here, in the registration status in (o), 0 is set in the case of single registration in the registration process. Also, in the case of continuous registration, 1 is set if it is the continuous start point, 2 if it is during continuous registration, and 3 if it is the continuous point end point.

[0011] The image frame including the mark point 92 and connection line 93 shown in Figure 13 registered as above is displayed as an expansion view in the form, and the image frame can also be displayed as an abnormal diagram. Also, the mark point 92 and connection line 93 are obtained by calculation at the corresponding positions in different image frames, and the mark point is displayed at the obtained positions. For example, in Figure 11(A), it is assumed that a camera 35 discovers a scratch in the image at position a, and an image with a mark point 92 and a connection line 93 as shown in Figure 11(C) is registered according to the above registration procedure. The positions of the mark point 92 and connection line 93 in the image of Figure 11(C) are converted into spatial coordinates in the pipe channel and recognized. Whether the positions of the mark point 92 and connection line 93 in these spatial coordinates also exist in adjacent image frames is obtained by calculation. If they exist, they are registered as image diagnostic information, and the mark point 92 and connection line 93 are displayed at the corresponding positions. Therefore, as shown in Fig. 11(B), assuming that the camera 35 has moved to position b, although the position of the scratch on the image also moves, the position in the spatial coordinates is the same as the scratch registered when the camera was at position a in Fig. 11(A). Therefore, in the image frame when the camera in Fig. 11(B) is at position b, the marking point 92 and the connection line 93 will be automatically displayed. By adopting such a method, the registered marking point 92 and connection line 93 will also be displayed in other image frames, and when viewed as a video, the marking point 92 and the connection line 93 will also appear to move together. As shown in Fig. 11(D), since the marking point 92 and the connection line 93 are all displayed in the image frames during the movement of the camera position from a to b, the scratch also appears to move as shown in the figure. Diagnose the pipeline video data to the end. After the diagnosis information registration is completed, create a form. The display distance range and the development angle of the developed view to be displayed on the form can be specified according to the form image display specification and set to be easy to evaluate. Also, create the form management information by editing the form management information and perform an overall evaluation as the diagnosis result. The developed view, the pipeline diagram, and the abnormal diagram are each displayed.

[0012] In the following Figs. 4, 5, 6, and 7, simply showing "pipe" refers to the "pipe" itself that constitutes the "pipe channel (underground water channel)". It is assumed that pipeline video data 86 composed of a large number of TV camera developed images in an abnormal state as shown in Figs. 4 and 5 is obtained by the conventional device configured as described above. These pipeline video data 86 are stored as the image memory 87 in the memory 80 of Fig. 10 and correspond to the image frame information 1 to m shown in Fig. 9(a). The image diagnosis information corresponding to each image frame information 1 to m is stored as the form management information (a), (i), (u), (e), (o) shown in Fig. 9(b). These document management information (a) to (e) are visually judged based on the various developed images 86 shown in FIGS. 4 and 5 according to the judgment criteria in FIG. 6 (in the case of reinforced concrete pipes and earthenware pipes) and FIG. 7 (in the case of rigid vinyl chloride pipes). Specifically, it consists of coordinate information (mouse position) (a), the corresponding image frame number (b), abnormality location information (c) including the abnormal items and ranks A, B, C described in FIGS. 6 and 7, pipeline distance (d), and registration status (e). A specific example of visual judgment will be described based on FIGS. 4 and 5. In the figures, 15 is the joint of the pipe, 16 is the water stain, 17 is the attached pipe, 18 is the reinforcing bar, and 19 is the stain.

[0013] [1] The damage of the pipe is represented by the axial crack shown in FIG. 4(a) and the inner wall defect shown in FIG. 4(b). Rank A is an example where the width is 5 mm or more (in the case of reinforced concrete pipes) and the length is 1 / 2 or more of the pipe length represented between the joints 15 of two pipes (in the case of earthenware pipes). Ranks B and C are as shown in FIGS. 6 and 7. [2] The crack of the pipe is represented by the circumferential crack shown in FIG. 4(c). Rank A is an example where the width is 5 mm or more (in the case of reinforced concrete pipes) and the length is 2 / 3 or more of the circumference (in the case of earthenware pipes). Ranks B and C are as shown in FIGS. 6 and 7. [3] The displacement of the pipe joint is shown in FIG. 4(d). Rank A is shown in FIG. 4(e) for detachment. Ranks B and C are in a state where a predetermined displacement has occurred at the joint 15 at the connecting part of two pipes as shown in FIGS. 6 and 7.

[0014] [4] Corrosion of the pipe: Rank A is shown in FIG. 5(c) and is the exposure of the reinforcing bar 18. Ranks B and C are the cases where the aggregate is exposed and the surface is rough as shown in FIGS. 6 and 7. [5] Sagging and meandering of the pipe can be judged from the bulge of the water stain 16 as shown in FIG. 5(e). Ranks A, B, and C are as shown in FIGS. 6 and 7, being equal to or greater than the inner diameter, 1 / 2 or more of the inner diameter, and less than 1 / 2 of the inner diameter, respectively. [6] Mortar adhesion is shown in FIG. 5(d). Ranks A, B, and C are as shown in FIGS. 6 and 7, being 30% or more of the inner diameter, 10% or more of the inner diameter, and less than 10% of the inner diameter, respectively. [7] The infiltrating water may blow out from the joint of the pipe as shown in Fig. 5(a), or may blow out from the crack of the pipe as shown in Fig. 5(b): Ranks A, B, and C are, as shown in Figs. 6 and 7, the cases of blowing out, flowing, and seeping. [8] The protruding attachment pipe 7 is the case where it protrudes from the inner wall of the pipe channel 14 as shown in Fig. 5(d): Ranks A, B, and C are, as shown in Figs. 6 and 7, the cases where the inner diameter of the attachment pipe 17 is 1 / 2 or more, 1 / 10 or more, and less than 1 / 10. [9] The adhesion of lard is shown in Fig. 5(f), and the infiltration of tree roots is shown in Fig. 5(f): Ranks A and B are, as shown in Figs. 6 and 7, the cases where the inner diameter is blocked by 1 / 2 or more and less than 1 / 2.

Prior Art Documents

Patent Documents

[0015]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0016] Based on the criteria shown in Figs. 6 and 7, the operator visually determines the pipeline video data shown in Figs. 4 and 5 individually. Then, the presence or absence of abnormal locations is determined according to the procedure shown in Fig. 8 described above and recorded in the form shown in Fig. 9(b). The determination of the deteriorated locations of the conventional pipe channel was to display the inner surface image of the pipe channel as a developed image or a pipeline diagram on a personal computer, directly view the image, mark points on the abnormal locations that seemed to be damaged, and perform the registration of the abnormal locations and the management of the form management information. Such marking of abnormal locations requires the operator to continuously move the display screen output inside the pipeline with a mouse or the like and visually find locations that seem abnormal and register them in a form. Therefore, it takes a lot of time to search for abnormal locations by moving the display screen one by one for all routes, and it is also necessary to ensure operators who can determine whether it is abnormal. Furthermore, there is a problem that it is difficult to unify the judgment levels.

[0017] The present invention utilizes artificial intelligence technology (AI) to create a developed drawing that has been AI-determined, in which the AI has previously displayed positions and types that seem abnormal on the image inside the pipeline. The purpose is to provide a device and method for diagnosing the deterioration of a pipeline that allows an operator to find abnormalities faster and more accurately based on this drawing.

Means for Solving the Problems

[0018] The method for diagnosing the deterioration of a pipeline according to the present invention is as follows: A step of saving the original developed image created based on the image information of the inner surface of the pipeline; A step of combining numerical information of abnormal positions with the original developed image; A step of generating an undamaged image using the original developed image in order to train an AI learning model to distinguish abnormal locations from other locations; A step of extracting the image information of the original developed image and the numerical information attached to the image to create a damaged image, reading the coordinates of abnormal locations from the numerical information of the original developed image, and extracting the image area where the abnormal locations are within the image; A step of constructing an AI learning program using the undamaged image, the damaged image, and the numerical information attached to them as teaching data; A step of constructing a program that takes the AI learning model constructed in the above step as input data and outputs the image information of the developed drawing that has been AI-determined as a learned AI judgment model for image data with unknown damage positions and damage types using an unregistered developed image consisting only of the image information of the inner surface image of the pipeline where damage has not been registered; A step of using the trained AI determination model constructed in the above step, taking the original developed image as input data, and outputting an AI-determined developed image that is identified and displayed by AI determination consists of

[0019] In addition, the method for diagnosing the deterioration of the conduit of the present invention particularly creates a processed developed image in which abnormal portions of the original developed image are highlighted a step of saving the original developed image created based on the inner surface image information of the conduit a step of combining numerical information of abnormal positions with the original developed image a step of creating and saving a processed developed image in which abnormal portions of the original developed image are highlighted a step of generating an image without damage using the processed developed image in order to train the AI learning model with abnormal locations and other locations a step of extracting the image information of the processed developed image and the numerical information assigned to the image to create a damaged image, reading the coordinates of the abnormal location from the numerical information of the processed developed image, and extracting the image area where the abnormal location enters the center point of the image a step of constructing an AI learning program using the image without damage, the damaged image, and the numerical information assigned to them as teacher data For image data with unknown damage positions and damage types using an unregistered developed image consisting only of the image information of the inner surface image of the conduit where damage has not been registered, using the core information of the AI learning model constructed in the above step as input data, and constructing a program that outputs the image information of the developed drawing determined by AI as a trained AI determination model a step of using the trained AI determination model constructed in the above step, taking the processed developed image as input data, and outputting an AI-determined developed image that is identified and displayed by AI determination It is characterized by consisting of

[0020] In the step of creating and saving the processed developed image with the abnormal part highlighted, the image processing is characterized in that the Canny method is applied to the original developed image as an image processing program to perform edge detection, and further, the abnormal locations of the original developed image are highlighted by line segment extraction through dilation and erosion processing.

[0021] The numerical information of the abnormal locations in the step of combining the numerical information of the abnormal locations is characterized by consisting of the coordinates of the abnormal locations and the abnormal types.

[0022] The abnormal types are characterized by including at least pipe breakage, pipe crack, pipe joint displacement, and pipe corrosion.

[0023] The pipe breakage is an axial crack with a width of 5 mm or more, the pipe crack is a circumferential crack with a width of 5 mm or more, the pipe joint displacement is the detachment of the pipe connection part, and the pipe corrosion is characterized by the exposed state of the reinforcing bars.

[0024] The pipe breakage and the pipe crack are ranked by the width of the crack, the pipe joint displacement is ranked by whether the pipe connection part is detached or displaced by a predetermined amount or more, and the pipe corrosion is ranked in multiple stages based on whether the reinforcing bars or the aggregates are exposed.

[0025] The image area in the step of extracting the image area where the abnormal location enters the image is characterized by being in the shape obtained by dividing the original developed image in the vertical and horizontal directions by a predetermined number of pixels respectively. In this example, it is not limited to entering the center point of the image area where the abnormal location enters the image.

[0026] The image area in the step of extracting the image area where the abnormal location enters the center point of the image is characterized by being in the shape obtained by dividing the processed developed image in the vertical and horizontal directions by a predetermined number of pixels respectively.

[0027] In the step of outputting the AI-determined and developed images that are identified and displayed by AI determination for each of the extracted damage images, the identification and display is characterized in that it is performed by color-coding for each of the extracted image areas. In this example, it is displayed by color-coding.

[0028] In the step of outputting the AI-determined and developed images that are identified and displayed by AI determination for each of the extracted damage images, the identification and display is characterized in that it is performed by different shading for each of the extracted image areas. In this example, it is displayed by different shading.

[0029] In the step of outputting the AI-determined and developed images that are identified and displayed by AI determination for each of the extracted damage images, the identification and display is characterized in that different identification and display are performed for each of the extracted objects. In this example, it is displayed by color-coding or shading for each of the extracted objects.

[0030] In the step of outputting the AI-determined and developed images that are identified and displayed by AI determination for each of the extracted damage images, the identification and display is characterized in that it is displayed with differences according to the rank of the damage determination criteria for each of the extracted images. In this example, it is displayed by color-coding or shading according to the rank for each of the extracted objects.

[0031] The device for diagnosing the deterioration of a conduit according to the present invention has a communication unit 10 that transmits and receives conduit inner surface image information via a network with an external device, an input unit 11 operated by a user, an image processing unit 12, and a recording unit 13. The recording unit 13 includes a recording area 30 for the conduit inner surface image information and diagnosis record information input via the communication unit 10, a recording area 31 for input data, and a recording area 32 for information on unregistered damage to the conduit inner surface. The image processing unit 12 includes a data acquisition unit 20 that transmits the information of the communication unit 10 to the recording unit 13, an input data generation unit 21 that creates an original developed image of the inner surface of the pipe conduit based on the inner surface image information of the pipe conduit recorded in the recording area 30, combines the form management information with this original developed image, and creates a developed image obtained by implementing an image processing program on this original developed image, a construction unit 23 of an AI learning model that constructs an AI learning program from the non-damaged image and the damaged image created by this input data generation unit 21, a construction unit 24 of a damage model that uses the learned AI determination model constructed by this construction unit 23 of the AI learning model, takes the developed image as input data, and creates an AI determination completed developed image that is identified and displayed by AI determination, and an output unit 25 that performs identification display by AI determination for each damaged image on the AI determination completed developed image and outputs it. It is characterized by comprising the above.

[0032] The input data generation unit 21 is characterized by creating a developed image with image processing reflected by implementing an image processing program on the original developed image.

[0033] The input data generation unit 21 is characterized by having a built-in image processing program using the Canny method for the original developed image.

[0034] The construction unit 23 of the AI learning model classifies images by the neural network method (AI learning), applies EfficientNet to the algorithm of AI learning, reads teacher data into the constructed AI learning program for learning, and outputs the weight data that becomes the core information of the obtained neural network model.

Effect of the Invention

[0035] According to the invention described in claim 1, a step of saving the original developed image created based on the inner surface image information of the pipe conduit, a step of combining the numerical information of the abnormal position with the original developed image, To make the AI learning model learn abnormal parts and other parts, a step of generating an undamaged image using the original developed image; A step of extracting the image information of the original developed image and the numerical information assigned to the image to create a damaged image, reading the coordinates of the abnormal part from the numerical information of the original developed image, and extracting the image area where the abnormal part enters the image; A step of constructing an AI learning program using the undamaged image, the damaged image, and the numerical information assigned thereto as teacher data; A step of constructing a program that outputs, as a learned AI determination model, the image information of the developed drawing that has been AI-determined, using the unregistered developed image consisting only of the image information of the inner surface image of the pipe conduit with no registered damage as input data for the image data with unknown damage position and damage type, and using the AI learning model constructed in the above step; A step of using the learned AI determination model constructed in the above step, taking the original developed image as input data, and outputting the AI-determined developed image identified and displayed by AI determination; Therefore, the operator can quickly find abnormalities by observing the identification display of the parts that seem to be abnormal in the AI-determined developed image. Moreover, since different identification displays are made for each damaged image, the types of abnormal parts can be easily searched.

[0036] According to the invention described in claim 2, A step of saving the original developed image created based on the inner surface image information of the pipe conduit; A step of combining the numerical information of the abnormal position with the original developed image; A step of creating and saving a processed developed image in which the abnormal part of the original developed image is highlighted; To make the AI learning model learn abnormal parts and other parts, a step of generating an undamaged image using the processed developed image; A step of extracting the image information of the processed developed image and the numerical information assigned to the image to create a damaged image, reading the coordinates of the abnormal part from the numerical information of the processed developed image, and extracting the image area where the abnormal part enters the center point of the image; A step of constructing an AI learning program using the undamaged image, the damaged image, and the numerical information given to them as teacher data; For image data with unknown damage positions and damage types, using an unregistered developed image consisting only of the image information of the inner surface image of the pipe conduit without damage registration, and using the core information of the AI learning model constructed in the above step as input data, a program is constructed to output the image information of the developed drawing that has been AI-determined as a learned AI determination model. A step of using the learned AI determination model constructed in the above step, taking the developed image after image processing as input data, and outputting the developed image that has been AI-determined and identified through AI determination. Therefore, the operator can quickly find the locations that seem abnormal in the developed image that has been AI-determined. Moreover, since different identification displays are made for each damaged image, the types of abnormal locations can be easily searched. In addition, since a developed image after image processing with the abnormal parts of the original developed image highlighted is created, the discrimination accuracy can be improved.

[0037] According to the invention described in claim 3, In the step of creating and saving the developed image after image processing with the abnormal parts highlighted, the image processing is as follows: applying the Canny method as an image processing program to the original developed image to perform edge detection, and further performing line segment extraction by dilation and erosion processing to highlight the abnormal locations of the original developed image. Therefore, the abnormal parts can be recognized more reliably.

[0038] According to the invention described in claim 4, The numerical information of the abnormal location in the step of combining the numerical information of the abnormal location consists of the coordinates of the abnormal location and the abnormal type. Therefore, the abnormal location in the developed image can be reliably displayed.

[0039] According to the invention described in claim 5, The abnormal type includes at least pipe breakage, pipe crack, pipe joint displacement, and pipe corrosion. Therefore, most of the abnormal locations of the pipe can be identified.

[0040] According to the invention described in claim 6, The breakage of the pipe is an axial crack with a width of 5 mm or more, the crack of the pipe is a circumferential crack with a width of 5 mm or more, the misalignment of the pipe joint is regarded as the detachment of the connecting part of the pipe, and the corrosion of the pipe is regarded as the exposed state of the reinforcing bars. Therefore, the abnormal state of rank A can be accurately and reliably recognized.

[0041] According to the invention described in claim 7, The breakage and crack of the pipe are ranked according to the width of the crack, the misalignment of the pipe joint is ranked according to whether the connecting part of the pipe is detached or misaligned by a predetermined amount or more, and the corrosion of the pipe is ranked according to whether the reinforcing bars or aggregates are exposed. Therefore, the abnormalities of the pipe corresponding to the ranks can be distinguished and displayed.

[0042] According to the invention described in claim 8, The image area in the step of extracting the image area where the abnormal part enters the image is in the shape of dividing the original developed image into a plurality of parts in the vertical and horizontal directions by a predetermined number of pixels respectively. Therefore, the developed image can be divided into a plurality of parts to display a plurality of abnormal parts. In the invention described in claim 8, the abnormal part does not necessarily enter the center point of the image area.

[0043] According to the invention described in claim 9, The image area in the step of extracting the image area where the abnormal part enters the center point of the image is in the shape of dividing the processed developed image into a plurality of parts in the vertical and horizontal directions by a predetermined number of pixels respectively. Therefore, the abnormal part of the pipe can be displayed at the center point of the image.

[0044] According to the invention described in claim 10, The identification display in the step of outputting the AI-determined developed image that is identified and displayed by AI determination for each of the extracted damage images is performed by color-coding for each of the extracted image areas. Therefore, the abnormal parts can be color-coded and reliably distinguished.

[0045] According to the invention described in claim 11, In the step of outputting an AI-determined developed image in which the AI determination is discriminatively displayed for each of the extracted damage images, the discriminative display is such that the type of damage, which is displayed by different shading for each of the extracted image regions, can be surely recognized by the difference in the shading.

[0046] According to the invention described in claim 12, In the step of outputting an AI-determined developed image in which the AI determination is discriminatively displayed for each of the extracted damage images, different discriminative displays are made for each of the extracted objects, so that even if an abnormal part spans a plurality of images, it can be surely discovered.

[0047] According to the invention described in claim 13, In the step of outputting an AI-determined developed image in which the AI determination is discriminatively displayed for each of the extracted damage images, the discriminative display is made to differ according to the rank of the damage determination criteria for each of the extracted images, so that for each of the extracted images, the type of damage and the determination can be displayed according to the rank of the damage, and thus serious damage and damage parts with a low rank that are easily overlooked can be discovered.

[0048] According to the invention described in claim 14, It has a communication unit that transmits and receives information on the inner surface image of the conduit through a network with an external device, an input unit operated by a user, an image processing unit, and a recording unit. The recording unit includes a recording area for the inner surface image information of the conduit and the diagnosis recording information input through the communication unit, a recording area for the input data, and a recording area for the unregistered damage information on the inner surface of the conduit. The image processing unit includes a data acquisition unit that transmits the information of the communication unit to the recording unit, creates an original developed image of the inner surface of the pipe conduit based on the inner surface image information of the pipe conduit recorded in the recording area, combines the form management information with this original developed image, and creates a developed image obtained by implementing an image processing program on this original developed image; a construction unit for constructing an AI learning model for constructing an AI learning program from the non-damaged image and the damaged image created by this input data generation unit; a construction unit for a damage model that uses the trained AI determination model constructed by this construction unit for the AI learning model, takes the developed image as input data, and creates an AI-determined developed image that is identified and displayed by AI determination; and an output unit that performs identification display for each damaged image on the AI-determined developed image and outputs it. Since it comprises , similar to the operation and effect described in Claim 1, an operator can quickly find the location that seems abnormal by observing the identification display. Moreover, since different identification displays are made for each damaged image, the types of abnormal locations can be easily searched.

[0049] According to the invention described in Claim 15, Since the input data generation unit creates a developed image after reflecting image processing by implementing an image processing program on the original developed image, the types of abnormal locations can be easily searched accurately.

[0050] According to the invention described in Claim 16, Since the input data generation unit incorporates an image processing program using the Canny method for the original developed image, the types of abnormal locations can be easily searched accurately.

[0051] According to the invention described in Claim 17, The construction part of the AI learning model is such that AI learning is image classification by the neural network method, the algorithm of AI learning applies EfficientNet, the constructed AI learning program is made to read teacher data and learn, and the weight data which is the core information of the obtained neural network model is output. Therefore, by AI learning, it becomes possible to improve the determination accuracy in accordance with the accumulation of data.

Brief Description of Drawings

[0052]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12

Figure 13

Figure 14

Embodiments for Carrying Out the Invention

[0053] The present invention A step of saving an original expanded image created based on the in-pipe inner surface image information, A step of combining numerical information of abnormal positions with the original expanded image, A step of generating an undamaged image using the original expanded image in order to train an AI learning model with abnormal parts and other parts, A step of extracting the image information of the original expanded image and the numerical information given to the image to create a damaged image, reading the coordinates of the abnormal part from the numerical information of the original expanded image, and extracting the image area where the abnormal part enters the image, A step of constructing an AI learning program using the undamaged image, the damaged image, and the numerical information given to them as teacher data, Using an unregistered developed image consisting only of the image information of the inner surface image of the conduit with no damage registration, for image data with unknown damage positions and damage types, constructing a program that outputs the image information of the developed drawing that has been AI-determined as a learned AI determination model using the AI learning model constructed in the above step as input data. Using the learned AI determination model constructed in the above step, with the original developed image as input data, and outputting the AI-determined developed image that has been identified and displayed by AI determination. It consists of.

[0054] Also, the method for diagnosing the deterioration of the conduit of the present invention particularly creates an image-processed developed image in which abnormal parts are highlighted in the original developed image. A step of saving the original developed image created based on the inner surface image information of the conduit. A step of combining numerical information of abnormal positions with the original developed image. A step of creating and saving an image-processed developed image in which abnormal parts are highlighted in the original developed image. In order to let the AI learning model learn abnormal locations and other locations, a step of generating a non-damage image using the image-processed developed image. A step of extracting the image information of the image-processed developed image and the numerical information attached to the image to create a damage image, reading the coordinates of the abnormal location from the numerical information of the image-processed developed image, and extracting the image area where the abnormal location enters the center point of the image. A step of constructing an AI learning program using the non-damage image, the damage image, and the numerical information attached to them as teacher data. Using an unregistered developed image consisting only of the image information of the inner surface image of the conduit with no damage registration, for image data with unknown damage positions and damage types, constructing a program that outputs the image information of the developed drawing that has been AI-determined as a learned AI determination model using the core information of the AI learning model constructed in the above step as input data. Using the learned AI determination model constructed in the above step, with the image-processed developed image as input data, and outputting the AI-determined developed image that has been identified and displayed by AI determination. It consists of.

[0055] In the step of creating and storing the image - processed developed image with the abnormal part highlighted, the image processing is as follows: Apply the Canny method as an image - processing program to the original developed image for edge detection. Further, it is desirable to highlight the abnormal locations in the original developed image by line - segment extraction through dilation and erosion processing.

[0056] The numerical information of the abnormal positions in the step of combining the numerical information of the abnormal positions shall consist of, for example, the coordinates of the abnormal positions and the abnormal types.

[0057] The abnormal types shall include at least pipe breakage, pipe crack, pipe joint displacement, and pipe corrosion.

[0058] Pipe breakage is an axial crack with a width of 5 mm or more. Pipe crack is a circumferential crack with a width of 5 mm or more. Pipe joint displacement is the detachment of the pipe connection part. Pipe corrosion is the state of rebar exposure.

[0059] The pipe breakage and pipe crack are ranked by the width of the crack. The pipe joint displacement is ranked by whether the pipe connection part is detached or displaced by a predetermined amount or more. The pipe corrosion is ranked in multiple levels based on whether the rebar or aggregate is exposed.

[0060] The image area in the step of extracting the image area where the abnormal location enters the image shall be in a shape obtained by dividing the original developed image in the vertical and horizontal directions by a predetermined number of pixels respectively, so that the developed image can be divided into multiple parts to display multiple abnormal locations.

[0061] The image area in the step of extracting the image area where the abnormal location enters the center point of the image shall be in a shape obtained by dividing the image - processed developed image in the vertical and horizontal directions by a predetermined number of pixels respectively.

[0062] In the step of outputting the AI - judged developed image that is identified and displayed by AI judgment for each of the extracted damage images, the identification and display are performed by color - coding for each of the extracted image areas.

[0063] In the step of outputting an AI-determined and expanded image in which the AI determination is discriminatively displayed for each of the extracted damage images, the discriminative display is performed by different line shadows for each of the extracted image regions.

[0064] In the step of outputting an AI-determined and expanded image in which the AI determination is discriminatively displayed for each of the extracted damage images, different discriminative displays may be applied to each of the extracted objects.

[0065] In the step of outputting an AI-determined and expanded image in which the AI determination is discriminatively displayed for each of the extracted damage images, the discriminative display may be made different according to the rank of the damage determination criteria for each of the extracted images.

[0066] In the step of outputting an AI-determined and expanded image in which the AI determination is discriminatively displayed for each of the extracted damage images, if the discriminative display is made different according to the rank of the damage determination criteria for each of the extracted images, it is possible to discover damage locations with a low rank that are easily overlooked.

[0067] It has a communication unit that transmits and receives pipe inner surface image information via a network with an external device, an input unit operated by a user, an image processing unit, and a recording unit. The recording unit includes a recording area for pipe inner surface image information and diagnostic recording information input via the communication unit, a recording area for input data, and a recording area for unregistered damage information on the pipe inner surface. The image processing unit includes a data acquisition unit that transmits the information of the communication unit to the recording unit, creates an original unfolded image of the inner surface of the conduit based on the inner surface image information of the conduit recorded in the recording area, combines the form management information with this original unfolded image, and creates an unfolded image obtained by performing an image processing program on this original unfolded image; an input data generation unit; a construction unit for an AI learning model that constructs an AI learning program from the undamaged image and the damaged image created by this input data generation unit; a construction unit for a damage model that uses the learned AI determination model constructed by this construction unit for AI learning, takes the unfolded image as input data, and creates an AI-determined unfolded image identified and displayed by AI determination; and an output unit that performs identification and display by AI determination for each damaged image on the AI-determined unfolded image and outputs it. It is characterized by comprising the same.

[0068] If the input data generation unit creates an unfolded image after reflecting the image processing program on the original unfolded image, the type of abnormal location can be easily searched accurately.

[0069] The input data generation unit incorporates an image processing program using the Canny method for the original unfolded image.

[0070] The construction unit for the AI learning model uses neural network method-based image classification for AI learning, applies EfficientNet to the algorithm of AI learning, reads teacher data into the constructed AI learning program for learning, and outputs the weight data that becomes the core information of the obtained neural network model.

Example

[0071] Example 1 of the apparatus and method for diagnosing the deterioration of a conduit according to the present invention will be described based on the drawings. In FIG. 1, the apparatus for diagnosing the deterioration of the conduit of the present invention includes a communication unit 10, an input unit 11, an image processing unit 12, and a recording unit 13. The communication unit 10 is connected to a network such as a LAN and transmits and receives information to and from an external device via the network. The input unit 11 consists of a mouse, a keyboard, etc., and is operated by the user of the image processing unit 12. The image processing unit 12 includes a data acquisition unit 20, a generation unit 21 of input data incorporating an image processing program 22, a construction unit 23 of an AI learning model incorporating an AI learning program 23a, a construction unit 24 of a damage determination model, and an output unit 25. The recording unit 13 consists of a recording area 30 for the inner surface image information and diagnostic record information of the pipe channel, a recording area 31 for the input data, and a recording area 32 for the inner surface image information (unregistered damage information) of the pipe channel.

[0072] Based on FIG. 2, the flow of the abnormality diagnosis performed using AI according to the present invention will be described. Steps 1 to 4 are the generation process of the input data. Step 1: Creation of the developed image (original developed image) of the inner surface of the pipe channel In the recording area 30 for the inner surface image information and diagnostic record information of the pipe channel in the recording unit 13, as shown in FIG. 9(a), the pipeline video data 86 divided into frames as shown in FIGS. 4 and 5 obtained by the conventional method is recorded. This pipeline video data 86 includes not only abnormal states but also a large number of developed images including those not recognized as abnormal, which are recorded through the acquisition unit 20 of the image processing unit 12. Further, in the recording area 30 for the inner surface image information and diagnostic record information of the pipe channel, as the form management information (image diagnosis information) for each frame of the pipeline video data 86 as shown in FIGS. 4 and 5 obtained by the conventional method shown in FIG. 9(b), the coordinate information (a) of the mouse position information described above, the image frame number (i), the abnormal location information (u), the pipeline distance (e), the registration state (o), etc. are recorded. In response to an instruction from the input unit 11, a database of still image information, moving image information, abnormal type, damage position, etc. in the recording area 30 of the recording unit 13 such as a personal computer is called by the input data generation unit 21 of the image processing unit 12. Based on this input data, a pipe inner developed image (original developed image) for AI learning as shown in Fig. 3(a) is created and saved in the recording area 31 of the input data. This original developed image is an image with the pipe opened, with the center of the image being the bottom of the pipe in the pipe channel and having a water mark 16, and the top and bottom of the image being the pipe top. In the image, in addition to cracks, attachment pipes 17, water marks 16, etc., dirt 19, etc. are also shown. The original developed image created in this step 1 is information only in the form of an image.

[0073] Step 2: Combining form management information In order to train the AI to determine abnormal locations, numerical information for the computer to read in addition to the image information of the original developed image and what kind of abnormalities exist in which parts of the original developed image is also required. Therefore, in step 2, the form registration management information (information determined and added by past visual inspections) recorded in the recording area 30 of the pipe inner surface image information and diagnostic record information is combined, the file name (ID information) of the original developed image is read, and numerical information (coordinates of abnormal positions, abnormal types, ranks of abnormalities, etc.) is linked from the form management information in which the inspection investigation results matching the ID information are accumulated. Specifically, in addition to the image frame information as shown in Figs. 9(a) and 9(b) described above, numerical information such as coordinate information (a), image frame number (i), abnormal location information (u), pipeline distance (e), registration status (o), and other information are added as image diagnostic information. As a result, numerical information and other information explaining the image information are added to the image information of the original developed image (image + text data).

[0074] Step 3: Execution of image processing program Since the original developed images created in steps 1 and 2 are mixed with dirt, stains, dust, etc., it is difficult to identify abnormal locations on the visual image. Even when performing AI learning, there is a concern that the accuracy of determination is low with only the original developed image. Therefore, by applying image processing technology to the original developed image and highlighting the parts that may be abnormal, it becomes easier to identify whether there is an abnormality and the judgment accuracy of the AI can be further improved. For specific image processing, the Canny method (Canny algorithm), which is known from Japanese Patent Application Laid-Open No. 2005-227055 as an image processing program, is applied to the original developed image to perform edge detection. Furthermore, by extracting line segments through dilation and erosion processing, it becomes possible to highlight abnormal locations in the original developed image. In the verification experiment, the judgment accuracy of the AI was improved as a result of summarizing the image processing technology of this technology compared to the case where no image processing technology was used. In this way, an image processing program based on the Canny method (Canny algorithm) is implemented for the original developed image.

[0075] Step 4: Creation of a developed drawing image with image processing reflected (hereinafter referred to as an image-processed developed image) By performing the image processing in Step 3 on the original developed image, an image-processed developed image is created as a developed drawing image with image processing reflected. An example of the created image-processed developed image is shown in Fig. 3(b). The linear pattern in Fig. 3(b) is obtained by applying image processing technology to highlight abnormal locations. In Fig. 3(b), it can be confirmed that many linear patterns are generated around the actual abnormal locations (damage cracks and protruding attachment pipes). The image-processed developed image is saved in the recording area 31 of the input data. By having the AI learn (perform deep learning) on this image-processed developed image with image processing reflected, the judgment accuracy of abnormalities can be improved. This Step 4 can also be omitted for other purposes such as simplification of data processing.

[0076] In Steps 5, 6, and 7, the image information of the image-processed developed image captured from the recording area 31 of the input data and the numerical information attached to the image are used as teacher data to construct a learning AI model for finding abnormal locations in the inner surface image of the pipe (original developed image). The learned parameters (weight information), which are the output values, are obtained. The AI learning model uses the method of image classification of a neural network (CNN), which is a method of deep learning. For a more detailed explanation.

[0077] Steps 5 to 7 construct the AI learning model. Step 5: Creation of an image without damage Extract the image information of the image after image processing and the numerical information attached to the image from the recording area 31 of the input data. To let the AI learning model learn abnormal parts and others, an image without damage is generated using the original developed image recorded in the recording area 31 of the input data. To generate an image without damage, read the coordinates of the abnormal part from the numerical information of the image after image processing given in Step 2, and randomly extract an image area without abnormality (image without damage [0] in Fig. 3(b)) within a range that does not overlap with the abnormal range. More specifically, cut out an image without damage that does not fall into the damage items of the judgment criteria in Figs. 6 and 7 and does not fall into ranks A, B, or C from the original developed image to generate an image without damage. Save this image without damage [0] again in the recording area 31 of the input data.

[0078] Step 6: Creation of a damaged image Extract the image information of the image after image processing shown in Fig. 3(b) and the numerical information attached to the image from the recording area 31 of the input data. To let the AI learning model learn abnormal parts and others, create a damaged image using the image information [0] without abnormality created in Step 5 and the image after image processing with abnormal parts in the recording area 31 of the input data (Fig. 3(b)). Read the coordinates of the abnormal part from the numerical information of the image after image processing, and extract the image area where the abnormal part enters the center point of the image. The types of abnormalities are considered as pipe breakage [1], pipe crack [2], pipe joint displacement [3], pipe corrosion [4], pipe sagging and serpentine [5], mortar adhesion [6], water intrusion [7], protruding attachment pipe [8], lard adhesion and root intrusion [9], etc. shown in Figs. 4 and 5. The size of the extracted image is, for example, in FIG. 3(b), a square of 300×300 pixels with a side length of d obtained by horizontally dividing the developed drawing into three equal parts. Save the damaged location again in the recording area 31 of the input data. The size of the extracted image is not limited to a square and can be a rectangle or other shapes. Note that the creation of the non-damaged image in step 5 and the creation of the damaged image in step 6 are generated based on the image information of the processed developed image and the numerical information attached to the image. Although there are problems with the determination accuracy, they may also be generated based on the image information of the original developed image and the numerical information attached to the image.

[0079] Step 7: Construction of the AI learning program Construct an AI learning program using the non-damaged image (step 5), the damaged image (step 6), and the numerical information attached to them (coordinate information (a), image frame number (i), abnormal location information (u), pipeline distance (e), registration status (o), etc. as shown in FIG. 9 above) as teacher data. The AI learning is based on the neural network method (image classification), and the algorithm of the AI learning applies EfficientNet. Let the constructed AI learning program read the teacher data for learning and output the weight data that is the core information of the obtained neural network model. Execute the AI learning program using a large number of developed image data accumulated so far, specifically about 20,000, and output the weight data.

[0080] Step 8: Construction of the trained damage location determination model Using the original developed image (only image information, without using numerical information) to be determined in the determination area 32 of the inner surface image of the pipe (unregistered damage information), for the image data with unknown damage location and type, construct a program that outputs the image information of the developed drawing after AI determination, using the core information of the AI learning model constructed in step 7 as input data. Here, the inner surface image of the pipe channel (damage unregistered information) refers to the unregistered original unfolded image recorded in the recording area 32 of the inner surface image of the pipe channel (damage unregistered information) in the original unfolded image created in Step 1, for which the presence, type, rank, etc. of damage have not yet been determined.

[0081] Step 9: Creation of the AI-determined unfolded image Using the trained AI determination model constructed in Step 7, with the processed unfolded image (Figure 3(b)) as input data, output an AI-determined unfolded image as shown in Figure 3(c) where AI determination is identified and displayed (for example, color-coded display or line and shadow separation display). The input data may be the original unfolded image instead of the processed unfolded image. When the original unfolded image is used as input data, an identification display of AI determination is attached to the original unfolded image as shown in Figure 3(a). In Step 6, the size of the extracted image was a square with a side length of d, which is one-third of the unfolded drawing in the horizontal direction. However, the size and shape are not limited to this.

[0082] The damage images to be identified and displayed may be made different for each of the ranks A, B, and C of the determination criteria shown in Figure 6 or Figure 7. The damage images have a specific shape and size, but are not limited to this. For example, in the case where the damaged part spans two or more sections, different identification displays may be applied to the damaged object itself. When the identification display is a color-coded display, for example, no damage [0] is white, breakage [1] is red, crack [2] is orange, joint displacement [3] is yellow, corrosion [4] is light green, pipe sagging / snake-like [5] is green, mortar adhesion [6] is light blue, water intrusion [7] is blue, attachment pipe protrusion [8] is gray, lard adhesion / root intrusion [9] is purple, etc. It is desirable that the identification display be a light color that can be distinguished from other colors and does not affect the presence or absence of abnormalities in the damage image. When the identification display is a shaded image, for example, as shown in Fig. 3(d), no damage [0] is represented by no shading, breakage [1] is represented by a grid pattern, crack [2] is represented by a diagonal line slanting upwards to the right, joint misalignment [3] is represented by a horizontal line, corrosion [4] is represented by a diagonal line slanting downwards to the right, sagging and meandering of the pipe [5] is represented by a vertical wavy line, mortar adhesion [6] is represented by a horizontal wavy line, infiltrated water [7] is represented by a vertical line, protruding attachment pipe [8] is represented by intersecting diagonal lines, adhesion of lard and infiltration of tree roots [9] is represented by a wavy line with vertical and horizontal intersections, etc. In Fig. 11, the camera 35 is of the mirror lens type, but this is not restrictive, and it is not limited to the type of camera for capturing images such as a fish-eye lens type or a wide-angle lens type.

Explanation of symbols

[0083] 1... System program (sewer inspection diagnosis support program), 2... Input / output processing unit for diagnostic data, 3... Diagnostic support processing unit, 4... Input processing unit for diagnostic results, 5... Display / print processing unit for forms, 6... Input processing unit for form management information, 7... Display / print processing unit for diagnostic information list, 10... Communication unit, 11... Input unit, 12... Image processing unit, 13... Recording unit, 14... Sewer, 15... Joint of the pipe, 16... Water stain, 17... Attachment pipe, 18... Reinforcing bar, 19... Dirt, 20... Data acquisition unit, 21... Generation unit for input data, 22... Image processing program, 23... Construction unit for AI learning model, 23a... AI learning program, 24... Construction unit for damage determination model, 25... Output unit, 30... Recording area for sewer inner surface drawing information / diagnostic record information, 31... Recording area for input data, 32... Recording area for sewer inner surface image information (undocumented damage information), 35... Camera, 80... Memory, 81... CPU, 82... Input / output control unit, 83... Input device, 84... Display device, 85... Output device, 86... Pipeline video data, 87... Image data, 88... Table, 92... Mark point, 93... Connection line.

Claims

1. A step of saving the original developed image created based on the inner surface image information of the pipe channel; A step of combining numerical information of abnormal positions with the original developed image; A step of generating an image without damage using the original developed image in order to train an AI learning model to distinguish abnormal parts from other parts; A step of extracting the image information of the original developed image and the numerical information attached to the image to create a damaged image, reading the coordinates of the abnormal part from the numerical information of the original developed image, and extracting the image area where the abnormal part is located within the image; A step of constructing an AI learning program using the image without damage, the damaged image, and the numerical information attached thereto as teacher data; A step of constructing a program that takes the core information of the AI learning model constructed in the step of constructing the AI learning program as input data and outputs the image information of the developed drawing that has been AI-determined as a learned AI determination model for image data with unknown damage positions and damage types using an unregistered developed image consisting only of the image information of the inner surface image of the pipe channel without registered damage; A step of using the learned AI determination model constructed in the above step, taking the original developed image as input data, and outputting an AI-determined developed image identified and displayed by AI determination; A method for diagnosing the deterioration of a pipe channel, characterized by comprising the above steps.

2. A step of saving the original developed image created based on the inner surface image information of the pipe channel; A step of combining numerical information of abnormal positions with the original developed image; A step of creating and saving a processed developed image in which abnormal parts of the original developed image are highlighted; A step of generating an image without damage using the processed developed image in order to train an AI learning model to distinguish abnormal parts from other parts; A step of extracting the image information of the processed developed image and the numerical information attached to the image to create a damaged image, reading the coordinates of the abnormal part from the numerical information of the processed developed image, and extracting the image area where the abnormal part is located at the center point of the image; A step of constructing an AI learning program using the image without damage, the damaged image, and the numerical information attached thereto as teacher data; Using the unregistered developed image consisting only of the image information of the inner surface image of the culvert with unregistered damage, for the image data with unknown damage position and damage type, taking the core information of the AI learning model constructed in the step of constructing the AI learning program as input data, and constructing a program that outputs the image information of the developed drawing with AI determination as the learned AI determination model. Using the learned AI determination model constructed in the above step, taking the processed developed image as input data, and outputting the developed image with AI determination identified and displayed by AI determination. A method for diagnosing the deterioration of a culvert, characterized by comprising the above steps.

3. In the step of creating and storing the processed developed image with the abnormal part highlighted, the image processing is to apply the Canny method as an image processing program to the original developed image for edge detection, and further, by line segment extraction using dilation and erosion processing, highlight the abnormal locations of the original developed image. The method for diagnosing the deterioration of a culvert according to claim 2, characterized in that.

4. The numerical information of the abnormal position in the step of combining the numerical information of the abnormal position consists of the coordinates of the abnormal position and the abnormal type. The method for diagnosing the deterioration of a culvert according to claim 1 or 2, characterized in that.

5. The abnormal type includes at least pipe breakage, pipe crack, pipe joint displacement, and pipe corrosion. The method for diagnosing the deterioration of a culvert according to claim 4, characterized in that.

6. The pipe breakage is an axial crack with a width of 5 mm or more, the pipe crack is a circumferential crack with a width of 5 mm or more, the pipe joint displacement is the detachment of the pipe connection part, and the pipe corrosion is the state of exposed steel bars. The method for diagnosing the deterioration of a culvert according to claim 5, characterized in that.

7. The pipe breakage and the pipe crack are ranked by the width of the crack, the pipe joint displacement is ranked by whether the pipe connection part is detached or displaced by a predetermined amount or more, and the pipe corrosion is ranked by whether the steel bars are exposed or the aggregates are exposed. The method for diagnosing the deterioration of a culvert according to claim 5, characterized in that.

8. The image area in the step of extracting the image area where the abnormal location enters the image is in the shape of dividing the original developed image in the vertical and horizontal directions by a predetermined number of pixels respectively. The method for diagnosing the deterioration of a culvert according to claim 1, characterized in that.

9. The image area in the step of extracting the image area where the abnormal part enters the center point of the image is in a shape obtained by dividing the processed and expanded image in the vertical and horizontal directions by a predetermined number of pixels respectively. The method for diagnosing deterioration of a conduit according to claim 1 or 2, characterized in that.

10. The identification display in the step of outputting the AI-determined and expanded image identified and displayed by AI determination on the extracted damaged image is performed by color-coding for each extracted image area of different abnormal types. The method for diagnosing deterioration of a conduit according to claim 1 or 2, characterized in that.

11. The identification display in the step of outputting the AI-determined and expanded image identified and displayed by AI determination for each of the extracted damaged images is performed by line shading for each extracted image area of different abnormal types. The method for diagnosing deterioration of a conduit according to claim 1 or 2, characterized in that.

12. The identification display in the step of outputting the AI-determined and expanded image identified and displayed by AI determination for each of the extracted damaged images is performed by applying different identification displays for each extracted object. The method for diagnosing deterioration of a conduit according to claim 1 or 2, characterized in that.

13. The identification display in the step of outputting the AI-determined and expanded image identified and displayed by AI determination for each of the extracted damaged images is performed by making different identification displays according to the ranks of the damage determination criteria for each extracted image. The method for diagnosing deterioration of a conduit according to claim 1 or 2, characterized in that.

14. It has a communication unit for transmitting and receiving conduit inner surface image information via a network with an external device, an input unit operated by a user, an image processing unit, and a recording unit. The recording unit includes a recording area for conduit inner surface image information / diagnosis recording information input via the communication unit, a recording area for input data, and a recording area for unregistered damage information on the conduit inner surface. The image processing unit includes a data acquisition unit that transmits the information of the communication unit to the recording unit, creates an original developed image of the inner surface of the conduit based on the inner surface image information of the conduit recorded in the recording area, combines the form management information with this original developed image, and creates a developed image obtained by implementing an image processing program on this original developed image; a construction unit of an AI learning model that constructs an AI learning program from the non-damaged image and the damaged image created by this input data generation unit; a construction unit of a damage model that uses the learned AI determination model constructed by this construction unit of the AI learning model, uses the original developed image as input data, and creates an AI-determined developed image identified and displayed by AI determination; and an output unit that performs identification display for each damaged image on the AI-determined developed image and outputs it. An apparatus for diagnosing deterioration of a conduit, characterized by comprising the above.

15. The apparatus for diagnosing deterioration of a conduit according to claim 14, wherein the input data generation unit creates a developed image after reflecting image processing by implementing an image processing program on the original developed image.

16. The apparatus for diagnosing deterioration of a conduit according to claim 15, wherein the input data generation unit incorporates an image processing program using the Canny method for the original developed image.

17. The apparatus for diagnosing deterioration of a conduit according to claim 14 or 15, wherein the construction unit of the AI learning model performs AI learning as image classification by the neural network method, applies EfficientNet as the algorithm of the AI learning, reads teacher data into the constructed AI learning program for learning, and outputs weight data that becomes the core information of the obtained neural network model.

Images