JP7603528B2 - Inspection support method and inspection support system - Google Patents

Description

This disclosure relates to an inspection support method and an inspection support system.

Radiation testing is known in which radiation such as X-rays is applied to a structure including a welded portion to capture an image, and then potential defects in the welded portion are detected from the image. Patent Document 1 discloses a technology for identifying potential defects by analyzing an image captured by a radiation transmission testing device using a trained model that has previously learned the characteristics of the defect to be identified. Patent Document 2 discloses a technology for sharpening radiographic images captured during piping inspections of nuclear power plants to improve image quality and accurately detect potential defects.

Patent No. 2968442 Patent No. 5743933

As disclosed in Patent Documents 1 and 2, technology is provided that automatically detects defect candidates from images captured by a radiographic testing device. Inspectors check the images of each of the automatically detected defect candidates one by one and make a final judgment visually. However, the automatically detected defect candidates often include over-detected defect candidates, making the inspection work burdensome for inspectors.

This disclosure provides an inspection support method and inspection support system that can solve the above problems.

The inspection support method disclosed herein is an inspection support method that supports image inspection of radiographic testing, and includes the steps of acquiring a plurality of images including the same inspection target area, using the plurality of images to detect defect candidates in the inspection target area contained in the images, creating a converted image that converts the image quality of the plurality of images to a predetermined image quality that takes into account the visibility of an inspector, using the plurality of converted images to detect defect candidates in the inspection target area contained in the converted image, and performing a comprehensive evaluation of the defect candidates detected based on the images and the converted images across the plurality of images and the plurality of converted images , or a comprehensive evaluation of the plurality of images and a comprehensive evaluation of the plurality of converted images .

The inspection support system disclosed herein is an inspection support system that supports image inspection of radiographic testing, and includes: a means for acquiring multiple images including the same inspection target area; a means for using the multiple images to detect defect candidates in the inspection target area contained in the images; a means for creating a converted image that converts the image quality of the multiple images to a predetermined image quality that takes into account the visibility of an inspector; a means for using the multiple converted images to detect defect candidates in the inspection target area contained in the converted image; and a means for performing a comprehensive evaluation of the multiple images and the multiple converted images, or a comprehensive evaluation of the multiple images and a comprehensive evaluation of the multiple converted images, for the defect candidates detected based on the images and the converted images.

The above-mentioned inspection support method and inspection support system can improve the detection accuracy of defect candidates in radiographic testing.

1 is a block diagram showing an example of an examination support system according to an embodiment. 11 is a flowchart illustrating an example of an inspection support process according to the embodiment. 1 is a first flowchart illustrating an example of an integrated automatic analysis process according to the embodiment. 11 is a second flowchart illustrating an example of an integrated automatic analysis process according to the embodiment. 10 is a flowchart illustrating an example of a learning process according to the embodiment. 10 is a flowchart illustrating an example of an automatic analysis process according to the embodiment. 13 is a flowchart showing a modified example of the automatic analysis process according to the embodiment. FIG. 2 is a diagram illustrating an example of a hardware configuration of the inspection support device according to the embodiment.

The inspection support method of the present disclosure will be described below with reference to FIGS.
<Embodiment>
(System Configuration)
FIG. 1 is a block diagram illustrating an example of an examination support system according to an embodiment.
The inspection support system 1 includes a radiation transmission test device 2, a digitizer 3, a display device 4, and an inspection support device 10. The radiation transmission test device 2 irradiates radiation on an object to be inspected M, detects the transmitted radiation with a film, and outputs an analog image (film) of the object to be inspected M. The radiation transmission test device 2 may be capable of attaching a plurality of films to the object to be inspected M and taking a plurality of images with one irradiation. The radiation transmission test device 2 may also be capable of outputting a digital image of the object to be inspected M. The object to be inspected M is, for example, a structure such as a metal plate or a pipe including a welded portion. The radiation transmission test device 2 takes an image of the object to be inspected M so as to include the welded portion. The digitizer 3 converts the analog image (film) taken by the radiation transmission test device 2 into a digital image. Hereinafter, an example will be described in which the radiographic examination device 2 outputs an analog image (film) and converts it into a digital image by the digitizer 3. However, if the radiographic examination device 2 outputs a digital image, the digitizer 3 is not necessary, and the inspection support system 1 can be configured without the digitizer 3. The inspection support device 10 obtains a plurality of digital images of one inspection object M from the digitizer 3 or the radiographic examination device 2, and detects defect candidates of the inspection object M. The defect candidates are those that are considered to have a high probability of being welding defects such as voids and cracks occurring in the welded portion. In order to detect defect candidates of one inspection object M, the influence of noise and the like occurring in the image and overdetection are suppressed, and the detection accuracy of defect candidates is improved by using a plurality of digital images of the inspection object M. It is preferable that the multiple digital images of the object M to be inspected are taken within a predetermined time under the same shooting conditions (exposure dose, positional relationship with the object M to be inspected, etc.), and for example, the multiple digital images are images taken simultaneously using multiple films (the films may be the same or different types) that are stacked, or images obtained by continuously shooting the object M to be inspected set in the radiation transmission testing device 2. The inspection support device 10 outputs the detected defect candidates to the display device 4, paper, electronic file, etc. The display device 4 is configured using a liquid crystal display, etc. The display device 4 displays the image of the object M to be inspected and the defect candidates detected by the inspection support device 10. The image of the object M to be inspected that is output to paper or electronic file, etc., is stored as an inspection record.

The test support device 10 includes an image acquisition unit 11 , a control unit 12 , a learning unit 13 , an input reception unit 14 , an output unit 15 , and a memory unit 16 .
The image acquisition unit 11 acquires a plurality of digital images of the object M to be inspected.
The control unit 12 controls the process of detecting defect candidates from the inspection object M. The control unit 12 includes a position adjustment unit 121, an image processing unit 122, and an integrated analysis unit 123.
The position adjustment unit 121 aligns a plurality of digital images using features such as marker information indicating an inspection range within an image and a brightness difference between a specific position within an image and its surroundings.

The image processing unit 122 performs various image processing such as noise removal, brightness value change, filter processing, and integration of multiple images on the digital images acquired by the image acquisition unit 11. For example, the image processing unit 122 integrates multiple digital images to create one integrated image. For example, the image processing unit 122 converts the digital images into images with image quality (brightness, contrast, resolution, number of pixels, gradation, dynamic range, etc.) that makes it easy for an inspector to visually confirm. The image processing unit 122 also performs processing such as removing and homogenizing brightness unevenness in the image and aligning the image quality of multiple digital images to a standard image quality.

The integrated analysis unit 123 performs a process of detecting candidates for welding defects from images using multiple digital images. The detection of defect candidates is performed using a trained model (detection model 161) constructed by learning images showing welding defects. For example, the integrated analysis unit 123 calculates a predetermined feature amount from the digital image acquired by the image acquisition unit 11 and inputs the feature amount to the detection model 161. The detection model 161 outputs the type of defect indicated by the feature amount and its confidence value. The integrated analysis unit 123 treats the confidence output by the detection model 161 as a score indicating the likelihood of the defect candidate, and calculates an overall score for multiple digital images obtained by photographing one inspection object M. For example, the integrated analysis unit 123 calculates an overall score by adding up the scores calculated for each of the multiple digital images. Also, for example, the integrated analysis unit 123 sets the score calculated for the integrated image in which multiple digital images are integrated into one digital image as the overall score. The overall score is a score calculated through multiple digital images. For example, the total score of defect candidates detected from only one of ten digital images will be a smaller value than the total score of defect candidates detected from all ten digital images. Hereinafter, the process of detecting defect candidates from one image using detection model 161 will be referred to as automatic analysis, and the process of detecting defect candidates based on detection model 161 using N images and calculating a total score will be referred to as integrated automatic analysis.

The learning unit 13 learns images that show welding defects, and creates a detection model 161 that outputs information such as the type and certainty of the welding defect contained in the digital image, and the position where the welding defect is shown. For example, the learning unit 13 reads in a large number of digital images for learning, such as image A showing defects a1, a2, ..., and image B showing defects b1, b2, ..., learns the relationship between the image features of the part showing the defect and the type of defect, such as defects a1, a2, ..., and creates the detection model 161.

The input reception unit 14 receives user instructions and information input. For example, the user inputs an operation to instruct the execution of an integrated automatic analysis to the testing support device 10. The input reception unit 14 receives the input operation and outputs it to the control unit 12.

The output unit 15 outputs the defect candidates and the overall score detected by the integrated analysis unit 123 to the display device 4, an electronic file, paper, or the like.
The storage unit 16 stores various information such as the digital images acquired by the image acquisition unit 11 and the detection model 161 .

The inspection support device 10 may be configured with multiple computers capable of communicating with each other. The memory unit 16 may be configured in a storage device separate from the inspection support device 10. The inspection support system 1 may be configured to include one storage device including the memory unit 16, multiple inspection support devices 10 communicably connected to the storage device via a network, and a radiation transmission test device 2 and a digitizer 3 provided for each inspection support device 10.

(Operation)
Next, the operation of the test support device 10 will be described with reference to FIGS.
2 shows the overall flow of support processing for a radiographic examination using the inspection support device 10.
First, the radiographic examination device 2 is used to take N images of the object M (step S1). As a method of taking an image of the same location of the object M, there is a method of repeatedly taking images by changing the film, as well as a method of taking images of multiple films at once by overlapping them. The radiographic examination device 2 outputs the N analog images (films) taken by the film. Next, the digitizer 3 is used to convert the N analog images (films) into digital images (step S2). Next, the inspector inputs the N digital images into the inspection support device 10. In the inspection support device 10, the image acquisition unit 11 acquires the N digital images and records them in the storage unit 16 (step S3). Next, the inspector instructs to perform an integrated automatic analysis. The input reception unit 14 receives the instruction to perform an integrated automatic analysis and outputs it to the control unit 12. Then, the position adjustment unit 121 first aligns the N digital images (step S4). For example, the position adjustment unit 121 adjusts the positions of the N digital images based on predetermined marker information captured in each of the N digital images so that the pixel groups in which the marker information is captured overlap. Alternatively, the position adjustment unit 121 extracts a pixel group having a characteristic and local brightness value or feature value different from that of the surrounding pixels from each of the N digital images, and matches the positions of the N digital images so that the extracted pixel groups overlap. When N images are captured under the same shooting conditions, in theory, defects should be captured at approximately the same positions in the digital images. In reality, the positions and sizes of defect candidates captured in the images do not exactly match, and there may be a slight positional shift. In addition, due to the characteristics of radiation, the overall brightness level is somewhat different, and brightness unevenness and noise contained in the digital image may cause overdetection. In step S4, positioning is performed using a reliable mark on the image as a reference, and the detection accuracy is improved by the process described later. Note that the positioning in step S4 is not essential. If the positions of the N digital images are originally aligned, the process of step S4 can be omitted.

Next, the integrated analysis unit 123 performs an integrated automatic analysis (step S5). In the integrated automatic analysis, multiple digital images of the same area are used to perform an automatic analysis, thereby reducing the effects of noise caused by the film and noise caused by digitization, and improving the accuracy of the automatic analysis. There are two types of integrated automatic analysis:

(1) Digital images are processed one by one: An automatic analysis is performed on each of the N digital images, and the results are then integrated. Examples of integration methods include (a) a method in which, if a defect candidate is detected at the same position in all digital images, it is assumed that there is a defect detected by the detection model 161 at that position; (b) a method in which, if the sum of the scores calculated for each of the N digital images for a defect candidate detected at a certain position is equal to or greater than a predetermined threshold, it is assumed that there is a defect detected by the detection model 161 at that position; and (c) a method in which, if a defect candidate is detected with a score equal to or greater than a predetermined value at a position that can be assumed to be the same in at least a predetermined number of the N images, it is assumed that there is a defect detected by the detection model 161 at that position. This process will be described in more detail later with reference to FIG. 3.

(2) Merge and process N digital images: First, N digital images are merged into one to create an integrated image, and then automatic analysis is performed on the integrated image. One method for merging digital images is to use the average brightness value of each pixel in multiple images as the brightness value of the corresponding pixel in the integrated image. This process will be described in more detail later with reference to Figure 4.

Next, the output unit 15 outputs the result of the integrated automatic analysis to the display device 4 (step S6). For example, the output unit 15 displays N digital images or one integrated image, and displays the positions of the defect candidates detected in the integrated automatic analysis and the overall score superimposed on those images, or separately from the images, for example in a list format. The output unit 15 may display, for example, by color according to the type of defect or the magnitude of the overall score. The inspector makes a final evaluation of the radiographic test by referring to the displayed result of the integrated automatic analysis (step S7). The final evaluation is to determine whether the defect candidates detected by the integrated automatic analysis are true defects or overdetection. The inspector visually checks the images used in the integrated automatic analysis and makes a final evaluation by referring to the overall score. At this time, the inspector may apply a filter to the image (digital image, integrated image, etc.) displayed on the display device 4 so that the inspector can easily check it. For example, when the inspector specifies the type of filter to be applied to the image, the image processing unit 122 performs the specified filter processing on the displayed image. The output unit 15 then outputs the filtered image to the display device 4. By trying out various filters, the inspector can determine, for example, whether the area of interest is image noise or a real defect. This allows the inspector to make a final evaluation more accurately and efficiently.

2. Integrated automatic analysis (processing one sheet at a time)
Next, the process of automatically analyzing N digital images one by one in the integrated automatic analysis in step S5 will be described with reference to FIG.
The control unit 12 reads out the first of the N digital images from the storage unit 16 and sets the variable j to 1 (step S11). The control unit 12 instructs the integrated analysis unit 123 to perform an automatic analysis. The integrated analysis unit 123 performs an automatic analysis of the first digital image using the detection model 161 (step S12). For example, the integrated analysis unit 123 calculates a predetermined feature amount for the first digital image and inputs the calculation result to the detection model 161. The integrated analysis unit 123 acquires the position information (for example, coordinate information when a predetermined position of the digital image is set as the origin) of the detection position of the defect candidate output by the detection model 161, the type of defect (small cavity, poor joint, foreign material contamination, etc.), and a score (certainty), and records them in the storage unit 16 in association with the first digital image. Next, the control unit 12 determines whether the variable j is equal to or less than N (step S13). If the variable j is equal to or smaller than N (step S13; Yes), the control unit 12 adds 1 to the variable j (step S14), and repeats and executes the processes from step S12 onward. Note that if the same defect candidate as the defect candidate detected in the first image is detected by the automatic analysis of the second or subsequent digital images, the same position information as that of the first image is detected because the alignment was performed in step S4 of FIG.

If the variable j exceeds N (step S13; No), the integrated analysis unit 123 integrates the results of the automatic analysis for the N digital images (step S15). (a) For example, when a defect candidate is detected at a position that can be considered to be the same in N digital images (including not only when the coordinate information of the N positions where the defect candidate is detected is completely identical, but also when the coordinate information is included within a predetermined range), the integrated analysis unit 123 determines that there is a welding defect at that position and sets the sum or average of the scores of the N images as the overall score of the defect candidate. (b) For example, the integrated analysis unit 123 sums up the scores of the defect candidates detected at positions that can be considered to be the same in each of the N digital images, and when the sum is equal to or greater than a predetermined threshold, determines that there is a welding defect at that position and sets the sum or average of the scores as the overall score of the defect candidate. (c) For example, when a defect candidate is detected with a score equal to or greater than a predetermined value in a position that can be considered the same on at least a predetermined number of sheets out of the N sheets, the integrated analysis unit 123 determines that there is a welding defect at that position, and sets the sum or average of the scores as the overall score of the defect candidate. The integrated analysis unit 123 outputs the position information of the defect candidate, the type of defect, and the overall score to the output unit 15. According to the method described in FIG. 3, only defects detected from multiple digital images are output as defect candidates, so that overdetection due to image noise and overdetection by the detection model 161 can be suppressed, and the detection accuracy of defect candidates can be improved.

3. Integrated automatic analysis (merging N sheets for processing)
Next, the process of integrating N digital images into one integrated image and then automatically analyzing the integrated image in the automatic integration analysis in step S5 will be described with reference to FIG.
The control unit 12 reads out N digital images from the storage unit 16 (step S21). The control unit 12 instructs the image processing unit 122 to create an integrated image. The image processing unit 122 creates an integrated image by integrating N digital images (step S22). For example, the image processing unit 122 calculates the average of the luminance values of the pixels of the N digital images that correspond as a result of the alignment process in step S4, and creates an integrated image having the average value as the luminance value of the corresponding pixels. In addition, a method of creating an integrated image by setting the sum of the luminance values of each pixel of multiple images to the luminance value of the corresponding pixel of the integrated image, or a method of creating an integrated image by setting the maximum value, minimum value, median value, etc. of the luminance values of each pixel of multiple images to the luminance value of the corresponding pixel of the integrated image, etc. are conceivable. The image processing unit 122 records the integrated image in the storage unit 16. Next, the integrated analysis unit 123 performs automatic analysis on the integrated image using the detection model 161 (step S23). For example, the integrated analysis unit 123 inputs a predetermined feature amount calculated for the integrated image to the detection model 161, obtains the position information, the type of defect, and the score of the defect candidate, and records them in the storage unit 16 in association with the integrated image. When N images are merged and processed, the score calculated based on the integrated image becomes the overall score. The integrated analysis unit 123 may select only those images whose overall score is equal to or greater than a predetermined value as defect candidates. The integrated analysis unit 123 outputs the position information, the type of defect, and the overall score of the defect candidate to the output unit 15. According to the method described in FIG. 4, one digital image (integrated image) is created by averaging multiple digital images, and defect candidates are detected based on the created integrated image. This smooths out the variation in image quality between images, suppresses overdetection due to noise or the like and overdetection by the detection model 161, and improves the detection accuracy of defect candidates.

Next, the automatic analysis in step S12 in FIG. 3 and step S23 in FIG. 4 will be described with reference to FIGS.
4. Creation of a detection model First, the creation of a detection model 161 used in the automatic analysis of steps S12 and S23 will be described with reference to FIG. 5. First, the image acquisition unit 11 acquires a plurality of digital images for learning (step S31). A welding defect is captured in the digital image for learning, and the type of the defect and the position where the welding defect is captured are specified by a prior analysis. Next, the learning unit 13 performs machine learning on the digital image for learning (step S32). For example, the learning unit 13 extracts a predetermined feature amount from the digital image by a predetermined calculation method. The learning unit 13 learns the relationship between the feature amount extracted from the area where the welding defect is captured and the welding defect specified by the prior analysis, and the relationship between the feature amount extracted from the sound area where the welding defect is not captured and "no welding defect" by machine learning, and creates a detection model 161 that outputs the presence or absence of a welding defect, the type and certainty of the defect if the welding defect can be detected, and position information within the image when the feature amount extracted from the digital image is input. The learning unit 13 records the created detection model 161 in the storage unit 16 .

The detection model 161 may be created according to the characteristics of the image to be automatically analyzed. For example, for the process described in FIG. 3, the detection model 161 may be created by using each of a large number of learning images as learning data, or multiple integrated learning images may be created by selecting N images of the same inspection target area from the large number of learning images, and the detection model 161 may be created by using the multiple integrated images as learning data. In addition, a converted image may be created for each of the large number of learning images in the same manner as the converted image that takes into account the visibility of the inspector described next using FIG. 6, and the detection model 161 may be created by using these converted images as learning data. In addition, as described later using FIG. 7, image quality adjustment may be performed on each of the large number of learning images to smooth out the luminance variation between images and the luminance unevenness within one image, and the detection model 161 may be created by using the learning images after image quality adjustment as learning data.

5. Automatic Analysis Processing Next, the automatic analysis in steps S12 and S23 will be described in more detail with reference to FIG. 6. As already described, the automatic analysis is a process for detecting defect candidates from a digital image of an inspection target using the detection model 161. Here, a process for converting an original image into a converted image that takes into account the visibility of an inspector when performing the automatic analysis, and performing automatic analysis on the converted image in addition to the original image will be described. The original image is a digital image acquired by the image acquisition unit 11, or an integrated image created by the image processing unit 122.

First, the integrated analysis unit 123 performs an automatic analysis on the original image (step S41). This process is the same as the process described in step S12 of FIG. 3 or step S23 of FIG. 4. Next, the control unit 12 determines whether to perform an automatic analysis using a converted image (step S42). For example, if an inspector desires to detect defects using not only the original image but also the converted image, the inspector inputs a setting for automatic analysis using the converted image to the inspection support device 10 in advance. This setting input is acquired by the input reception unit 14, and setting information for performing automatic analysis using the converted image is set in the storage unit 16. The control unit 12 performs the determination of step S42 based on this setting information. If the setting information is not registered, the control unit 12 determines that automatic analysis will not be performed using the converted image (step S42; No), and ends the process of FIG. 6.

If the setting information is registered, the control unit 12 determines that automatic analysis is to be performed using the converted image (step S42; Yes) and executes the processes of steps S43 and S44. In step S43, the control unit 12 instructs the image processing unit 122 to create a converted image. The image processing unit 122 creates a new converted image by performing image conversion taking into account the visibility of the inspector (step S43). In general, the luminance distribution of an image captured by the radiographic test device 2 or a digital image converted from the image often has too wide a dynamic range of luminance and is not suitable for confirmation by the human eye. In such cases, the inspector adjusts the luminance and contrast of the original image before performing the analysis. Since the analysis of the original image in the radiographic test (for example, visually checking for defect candidates) is also an aspect of sensory inspection, automatic analysis is performed by combining not only the original image but also a converted image that is easily visible to the inspector. This can improve the accuracy of the automatic analysis.

For example, the output unit 15 outputs the original image to the display device 4, and the inspector inputs instruction information to adjust the brightness and contrast to the inspection support device 10. The image processing unit 122 acquires the instruction information via the input reception unit 14, changes the brightness and contrast of the original image according to the instruction information, and outputs the changed converted image to the display device 4 via the output unit 15. The inspector repeats the adjustment of the brightness, etc. until the original image becomes a converted image suitable for the inspection. When the desired converted image is obtained, the inspector inputs the completion of the image adjustment to the inspection support device 10. The image processing unit 122 records the image currently output to the display device 4 in the storage unit 16 as a converted image that takes into account the visibility of the inspector. If there are remaining digital images, for example, if the shooting conditions of the N images are the same, the remaining N-1 digital images may be processed in the same way as the first digital image to create converted images, or if the shooting conditions are different, the inspector may create converted images for the remaining N-1 images while checking them one by one.

Alternatively, the converted image converted to an image quality suitable for visual inspection may be registered in advance in the storage unit 16 as a representative image, and the image processing unit 122 may create the converted image by adjusting the brightness and contrast of the image to be converted to that of the representative image. For example, the image processing unit 122 may create brightness histogram distribution information indicating the distribution of brightness for each pixel of the representative image. The image processing unit 122 may create the converted image by automatically adjusting the brightness of all pixels so that the brightness distribution for each pixel in the converted image matches the brightness histogram distribution information. When creating an integrated image, a converted image may be created for each of the N digital images and then integrated to create the integrated image, or an integrated image may be created and then converted into the converted image. The converted image may also be an image with less information than the original image.

Once the converted image is obtained, the integrated analysis unit 123 performs an automatic analysis of the converted image (step S44). The integrated analysis unit 123 may use the same detection model 161 as in step S41, but it is preferable to perform the automatic analysis of the converted image using a detection model 161 created using a learning image converted into the converted image as learning data.

When automatic analysis using the converted image is performed in step S12 of FIG. 3, the integrated analysis unit 123 may integrate the results of the automatic analysis using the results of the automatic analysis of the N original images and the N converted images in step S15 of FIG. 3, or may integrate the results of the automatic analysis of the N original images and separately integrate the results of the automatic analysis of the N converted images. Also, when automatic analysis using the converted image is performed in step S23 of FIG. 4, the results of the automatic analysis performed on the integrated image and the results of the automatic analysis performed on the converted image may be integrated in the same manner as in step S15 of FIG. 3, or the results of the automatic analysis of the integrated image and the converted image may be handled separately without being integrated.

6. Modified Example of Automatic Analysis Processing FIG. 7 shows a modified example of the automatic analysis processing (FIG. 6). In the processing of FIG. 7, image quality is adjusted prior to automatic analysis to match certain conditions. When imaging for radiographic examination, the image quality of the obtained image changes due to individual differences of the object M, differences in the positional relationship between the irradiation port and the object M and the film (such as close or far depending on the film superposition position), and subtle differences in irradiation energy and irradiation time. By matching this to the image quality of a representative image (this representative image is not an image created with the visibility of the inspector in mind, but may be, for example, one of N digital images or another reference digital image) in advance, it is possible to improve the accuracy of automatic analysis. This image quality adjustment is particularly effective in the case of a configuration in which images are processed one by one (FIG. 3), but is also effective in the case of creating an integrated image and performing automatic analysis (FIG. 4). In the processing example shown in FIG. 7, image quality adjustment is performed prior to automatic analysis, but when creating an integrated image, image quality adjustment may be performed prior to creating the integrated image (before step S22 in FIG. 4). Furthermore, when adjusting the image quality, it is preferable to perform automatic analysis by using a detection model 161 created using the training images after the image quality adjustment as training data (as in the case of Figure 3), or by creating multiple integrated images from the training images after the image quality adjustment and using these as training data to create a detection model 161 (as in the case of Figure 4).

When performing image quality adjustment, the image processing unit 122 performs image quality adjustment on the digital image (step S40). For example, the image processing unit 122 adjusts the brightness and contrast by matching the brightness distribution of the original image to the brightness histogram distribution of the representative image, similar to the process of step S43 in FIG. 6. In addition, for example, when the brightness offset differs between multiple images, the brightness distribution between the images may be aligned by adding an offset amount X1 to the pixel value of each pixel of the original image 1 and adding an offset amount X2 to the pixel value of each pixel of the original image 2. In addition, when shooting with the radiation transmission test device 2, the center of the image (near the irradiation port) is bright and the periphery is dark, so that low-frequency component brightness unevenness is likely to occur. Therefore, for example, when the image processing unit 122 detects that the original image contains low-frequency components by frequency decomposing the original image, it applies a filter that cuts low-frequency components to suppress this brightness unevenness. Through these processes, the original images to be automatically analyzed can be unified under the same conditions for automatic analysis.

The processing from step S41 onwards is the same as in FIG. 6. The integrated analysis unit 123 performs automatic analysis on the original image after image quality adjustment (step S41). When performing automatic analysis using the converted image (step S42; Yes), the image processing unit 122 creates a converted image for the original image after image quality adjustment (step S43). Since the image quality of the N original images is already uniform through image quality adjustment, converted images can be created for the remaining original images using a process similar to the process for creating a converted image for the first image. Next, the integrated analysis unit 123 performs automatic analysis on the converted image (step S44).

As described above, according to this embodiment, defect candidates can be detected based on multiple images before an inspector performs visual evaluation. By detecting defect candidates based on multiple images, the effects of characteristics unique to each image (uneven brightness, noise, etc.) can be reduced and defect candidates can be detected. In addition, defect candidates detected from one image include overdetection, which places a burden on the inspector's confirmation work. However, by using multiple images, it is possible to determine, for example, that defect candidates detected from only a small number of images are due to overdetection, and that defect candidates detected in the majority of images are likely to have defects, thereby reducing the inspector's workload. Alternatively, defect candidates detected from only a small number of images are not included in the detection results, and only defect candidates detected in the majority of images are presented to the inspector, thereby reducing the burden of the inspector's confirmation work. In addition, by displaying the original image to be analyzed, the converted image, the image quality adjusted image, and the result of the integrated automatic analysis on the display device 4, the inspector can grasp the characteristics (confidence, position, etc.) of each defect candidate, and can perform a final evaluation of the automatic analysis results smoothly and easily. This will improve the workability of inspectors.

In the embodiment, an example of detecting welding defect candidates has been described, but the application of this embodiment is not limited to the detection of welding defect candidates in radiographic testing, and can be applied to the detection of other features of a structure based on images captured by a radiographic testing device.

FIG. 8 is a diagram illustrating an example of a hardware configuration of the inspection support device according to the embodiment.
The computer 900 includes a CPU 901, a main storage device 902, an auxiliary storage device 903, an input/output interface 904, and a communication interface 905. The above-mentioned test support device 10 is implemented in the computer 900. The above-mentioned functions are stored in the auxiliary storage device 903 in the form of a program. The CPU 901 reads the program from the auxiliary storage device 903, loads it in the main storage device 902, and executes the above-mentioned processing according to the program. The CPU 901 also secures a storage area in the main storage device 902 according to the program. The CPU 901 also secures a storage area in the auxiliary storage device 903 for storing data being processed according to the program.

A program for implementing all or part of the functions of the test support device 10 may be recorded on a computer-readable recording medium, and the program recorded on the recording medium may be read into a computer system and executed to perform processing by each functional unit. The term "computer system" as used herein includes hardware such as an OS and peripheral devices. In addition, if a WWW system is used, the term "computer system" also includes a home page providing environment (or display environment). In addition, the term "computer-readable recording medium" refers to portable media such as CDs, DVDs, and USBs, and storage devices such as hard disks built into a computer system. In addition, if the program is distributed to a computer 900 via a communication line, the computer 900 that receives the program may expand the program into the main storage device 902 and execute the above processing. In addition, the program may be for implementing part of the above-mentioned functions, and may further be capable of implementing the above-mentioned functions in combination with a program already recorded in the computer system.

As described above, several embodiments of the present disclosure have been described, but all of these embodiments are presented as examples and are not intended to limit the scope of the invention. These embodiments can be implemented in various other forms, and various omissions, substitutions, and modifications can be made without departing from the gist of the invention. These embodiments and their modifications are included in the scope of the invention and its equivalents as described in the claims, as well as in the scope and gist of the invention.

<Additional Notes>
The inspection support method and inspection support system described in each embodiment can be understood, for example, as follows.

(1) An inspection support method according to a first aspect is an inspection support method for supporting image inspection of a radiographic test, and includes the steps of: acquiring a plurality of images including the same inspection target area (S3); detecting defect candidates in the inspection target area included in the images using the plurality of images (automatic analysis of digital images, integrated images, converted images, etc.); and performing a comprehensive evaluation of the detected defect candidates across the plurality of images (calculating a comprehensive score) (S5, S15, S23).
By detecting defect candidates using multiple images and evaluating the results comprehensively, the accuracy of automatic analysis of image inspection in radiographic testing can be improved.

(2) A second aspect of the present invention is an inspection assistance method according to (1), further comprising a step (S4) of aligning the multiple images based on a landmark shown in the images.
By performing alignment even when multiple images are taken under different conditions, it is possible to prevent the confusion of defect candidates that exist in different positions in subsequent processing, thereby ensuring detection accuracy.

(3) An inspection support method according to a third aspect is the inspection support method of (1) to (2), wherein in the step of detecting defect candidates, the defect candidates are detected for each of the multiple images and a score indicating the likelihood of each of the detected defect candidates is calculated (S12), and in the step of evaluating, the overall evaluation is performed based on the number of images in which the defect candidates are detected and the score (S15).
It is possible to reduce noise in the image and prevent overdetection.

(4) An inspection support method according to a fourth aspect is the inspection support method of (1) to (2), wherein in the step of detecting the defect candidates, the defect candidates are detected based on an integrated image created from a plurality of the images (an integrated image created in such a way that a value calculated based on the pixel value of each pixel of the images is set as the brightness value of the corresponding pixel) and a score indicating the likelihood of the detected defect candidates is calculated (S23), and in the step of performing the evaluation, the overall evaluation is performed based on the defect candidates detected based on the integrated image and the score (S23).
It is possible to reduce noise in the image and prevent overdetection.

(5) An inspection support method according to a fifth aspect is the inspection support method of (3) to (4), further comprising a step of displaying the image, the detected defect candidates, and the score.
By displaying the evaluation results for each of the multiple images (or for the integrated image in the case of an integrated image), inspectors can make a final visual evaluation while referring to the location and certainty of potential defects, thereby improving work productivity.

(6) An inspection support method according to a sixth aspect is an inspection support method according to any one of (1) to (5), in which in the step of detecting defect candidates, a process of detecting the defect candidates is performed after aligning the image quality of the multiple images to a predetermined standard (S30).
By standardizing the image quality of images under certain conditions, it is possible to evaluate multiple images equally and fairly.

(7) An inspection support method according to a seventh aspect is an inspection support method according to any one of (1) to (6), further comprising a step (S33) of creating a converted image that converts the image quality of the image to a predetermined image quality that takes into account the visibility of an inspector, wherein in the step of detecting defect candidates, the defect candidates are detected from the photographed image and also from the converted image, and in the step of evaluating, the comprehensive evaluation is performed based on the defect candidates detected from the image and the converted image.
Assuming that a digital image (RAW data) has a dynamic range larger than that of a screen, automatic defect detection can be performed on the RAW data, and then the RAW data can be converted into an image (e.g., 8-bit image data) after adjusting the image quality (brightness, contrast) so that the defects are easier for people to see, and automatic detection of defect candidates can be performed on the converted image, thereby obtaining detection results for the RAW data and the converted image for each of a plurality of images. This improves visibility for inspectors, and ultimately improves detection accuracy.

(8) An inspection support method according to an eighth aspect is the inspection support method according to (6) to (7), wherein the image quality is the brightness and contrast of the image.
The brightness and contrast of the image can be adjusted.

(9) An inspection support method according to a ninth aspect is an inspection support method according to any one of (1) to (8), wherein the multiple images include at least one of images obtained by photographing the inspection target area multiple times and images obtained by photographing the inspection target area by overlapping multiple films.
By stacking multiple films or taking consecutive shots, it is possible to obtain multiple images (e.g., images taken under similar conditions) that contribute to improving detection accuracy.

(10) An inspection support method according to a tenth aspect is the inspection support method of any one of (1) to (9), wherein in the step of detecting the defect candidates, the defect candidates are detected using a detection model 161 created by machine learning the relationship between image features and defects.
By creating a detection model, defect candidates can be automatically detected.

(11) The inspection support system according to the eleventh aspect is an inspection support system 1 that supports image inspection of a radiographic test, and includes an image acquisition unit 11 that acquires multiple images including the same inspection target area, and a control unit 12 that uses the multiple images to detect defect candidates in the inspection target area contained in the images, and the control unit 12 performs a comprehensive evaluation of the detected defect candidates through the multiple images.

1... Inspection support system 2... Radiation transmission test device 3... Digitizer 4... Display device 10... Inspection support device 11... Image acquisition unit 12... Control unit 121... Position adjustment unit 122... Image processing unit 123... Integrated analysis unit 13... Learning unit 14... Input reception unit 15... Output unit 16... Memory unit 161... Detection model 900... Computer 901... CPU
902: Main memory device 903: Auxiliary memory device 904: Input/output interface 905: Communication interface

Claims (10)

An inspection support method for supporting image inspection of a radiographic examination, comprising:
acquiring a plurality of images including the same inspection target area;
detecting defect candidates in the inspection target area included in a plurality of the images using the plurality of the images;
creating a converted image that converts the image quality of the plurality of images into a predetermined image quality that takes into consideration the visibility of an inspector;
detecting defect candidates in the inspection target area included in the converted images using a plurality of the converted images;
performing a comprehensive evaluation of the defect candidates detected based on the images and the converted images , or a comprehensive evaluation of the defect candidates based on the images and the converted images ;
The inspection support method includes the steps of: A step of aligning the images based on landmarks shown in the images;
The method of claim 1 , further comprising: In the step of detecting the defect candidates in the inspection target region included in the image , the defect candidates are detected for each of the plurality of images, and a first score indicating a likelihood of each of the detected defect candidates is calculated;
In the step of detecting defect candidates in the inspection target region included in the converted image, the defect candidates are detected for each of the plurality of converted images, and a second score indicating a likelihood of each of the detected defect candidates is calculated;
In the step of performing the evaluation, a first overall score is calculated based on the number of the images in which the defect candidates are detected and the first score , and a second overall score is calculated based on the number of the converted images in which the defect candidates are detected and the second score,
outputting the image, the defect candidates detected from the image, and the first overall score, and the converted image, the defect candidates detected from the converted image, and the second overall score;
The inspection support method according to claim 1 or 2. In the step of detecting the defect candidate, the defect candidate is detected based on a plurality of the images and an integrated image created from the plurality of the converted images , and a score indicating a likelihood of the detected defect candidate is calculated;
In the step of performing the evaluation, a comprehensive score is calculated based on the defect candidates detected based on the integrated image and the score,
outputting the integrated image, the defect candidates detected from the integrated image, and the overall score;
The inspection support method according to claim 1 or 2. In the step of detecting the defect candidate in the inspection target area included in the image and the step of detecting the defect candidate in the inspection target area included in the converted image , a process of detecting the defect candidate is performed after aligning the image quality of the plurality of images to a predetermined standard.
The inspection support method according to any one of claims 1 to 4 . In the step of creating the converted image, the process of outputting the image to a display device, accepting input of instruction information for adjusting brightness and contrast, changing the brightness and contrast of the image in accordance with the instruction information, and outputting the converted image after the change to the display device is repeated until the converted image suitable for inspection is obtained.
The inspection support method according to any one of claims 1 to 5. The image quality is the brightness and contrast of the image.
The inspection support method according to claim 1 or 5 . The plurality of images include at least one of an image obtained by photographing the inspection target area a plurality of times and an image obtained by photographing the inspection target area by overlapping a plurality of films.
The inspection support method according to any one of claims 1 to 7 . In the step of detecting the defect candidate, the defect candidate is detected using a detection model created by machine learning the relationship between the feature amount of the image and the defect.
The inspection support method according to any one of claims 1 to 8 . An inspection support device that supports image inspection of a radiographic examination,
A means for acquiring a plurality of images including the same inspection target area;
a means for detecting defect candidates in the inspection target area included in a plurality of the images using the images;
a means for generating converted images for converting the image quality of the plurality of images into a predetermined image quality taking into consideration the visibility of an inspector;
a means for detecting defect candidates in the inspection target area included in a plurality of the converted images using the converted images;
a means for performing a comprehensive evaluation of the defect candidates detected based on the images and the converted images, using a plurality of the images and a plurality of the converted images, or a comprehensive evaluation of the plurality of the images and a comprehensive evaluation of the plurality of the converted images;
An inspection support system comprising :

Images