WO2011019014A1

WO2011019014A1 - Defect inspection method and defect inspection apparatus

Info

Publication number: WO2011019014A1
Application number: PCT/JP2010/063479
Authority: WO
Inventors: 正躬小西; 淳起藤井
Original assignee: 国立大学法人岡山大学
Priority date: 2009-08-11
Filing date: 2010-08-09
Publication date: 2011-02-17
Also published as: JP4543153B1; JP2011038905A

Abstract

Disclosed is a defect inspection method for determining the presence or absence of a defect from an image obtained by image capture of an object to be inspected. The aforementioned defect inspection method is provided with an image analysis step for determining the presence or absence of a defect in the object by processing the obtained image. The aforementioned image analysis step comprises subjecting the image obtained by image capture to processing for extracting the profile of the object , subjecting the contour extraction image obtained by the extraction processing to Hough transform processing so as to have the contour of the object be represented by a plurality of imaginary lines, then determining the presence or absence of a defect by whether or not the contour of the object is included in a circular region which is centered at the intersection of the imaginary lines and has a specified radius.

Description

Defect inspection method and defect inspection apparatus

The present invention relates to a defect inspection method and a defect inspection apparatus for a subject. More specifically, the present invention relates to a defect inspection method and a defect inspection apparatus for determining the presence or absence of a peeling defect or the like generated in a subject.

2. Description of the Related Art Conventionally, in a defect inspection of an object, a technique for obtaining a numerical value of the degree of defect by taking an image of the object and performing binarization processing, boundary detection processing, Hough conversion processing, and the like is known. (See Patent Document 1).

A method for determining the presence or absence of a defect by extracting a region to be inspected from an image of the subject and using a self-organizing neural network model that receives a feature vector representing the shape of the subject, and An apparatus is known (see Patent Document 2).

However, for example, a cutting tool such as an end mill has a complicated shape having a plurality of cutting edges on the outer periphery of a substantially cylindrical shape. Therefore, the operator determines whether or not there is a defect in the cutting edge by visual inspection. Was considered essential. Therefore, a highly reliable automated defect inspection method and defect inspection apparatus have been demanded.
JP 2001-264032 A JP 2006-047098 A

An object of the present invention is to provide a defect inspection method and a defect inspection apparatus that determine the presence or absence of a peeling defect from an image obtained by photographing a subject such as a cutting tool.

According to a first aspect of the present invention, in the defect inspection method for determining the presence or absence of a defect from an image obtained by imaging the subject with an imaging unit, the presence or absence of a defect in the subject by processing the obtained image An image analysis step for determining the contour of the subject with respect to the image obtained by imaging, and a Hough transform process with respect to the contour extracted image obtained by the extraction processing. The contour of the subject is expressed by a plurality of virtual lines, and a defect is determined depending on whether or not the contour of the subject is included in a circular area having a predetermined diameter centered at the intersection of the virtual lines. Judgment is made.

According to a second aspect of the present invention, in the defect inspection method for determining the presence / absence of a defect from an image obtained by imaging the subject with an imaging unit, the presence / absence of a defect in the subject by processing the obtained image An image analysis step for determining the contour of the subject with respect to the image obtained by imaging, and a Hough transform process with respect to the contour extracted image obtained by the extraction processing. The contour of the subject is expressed by a plurality of virtual lines, and an image obtained by imaging is divided into a plurality of partial areas within a circular area having a predetermined diameter centering on the intersection of the virtual lines. Then, discrete Fourier transform processing is performed on each of the images divided into the partial areas to convert them into the frequency domain, and the luminance value in the frequency area is calculated. Based on the luminance value for each partial area, the defect To determine the free.

In a third aspect of the present invention, in the defect inspection method according to the second aspect, the luminance value for each partial area is a value obtained by averaging the luminance in the partial area.

According to a fourth aspect of the present invention, in the defect inspection method according to the second or third aspect, the image analysis step determines the presence / absence of a defect using a neural network model that receives a luminance value for each partial area. To do.

According to a fifth aspect of the present invention, in the defect inspection apparatus that determines the presence / absence of a defect from an image obtained by imaging the subject with an imaging unit, the presence / absence of a defect in the subject is processed by processing the obtained image. An image analysis unit for determining the contour of the subject from the image obtained by the imaging, and the Hough transform process for the contour extracted image obtained by the extraction process. The contour of the subject is expressed by a plurality of virtual lines, and a defect is determined depending on whether or not the contour of the subject is included in a circular area having a predetermined diameter centered at the intersection of the virtual lines. Judgment is made.

According to a sixth aspect of the present invention, in the defect inspection apparatus that determines the presence / absence of a defect from an image obtained by imaging the subject with an imaging unit, the presence / absence of a defect in the subject is processed by processing the obtained image. An image analysis unit for determining the contour of the subject from the image obtained by the imaging, and the Hough transform process for the contour extracted image obtained by the extraction process. The contour of the subject is expressed by a plurality of virtual lines, and an image obtained by imaging is divided into a plurality of partial areas within a circular area having a predetermined diameter centering on the intersection of the virtual lines. The image divided into the partial areas is subjected to discrete Fourier transform processing to be converted into the frequency domain, the luminance value in the frequency domain is calculated, and the presence / absence of a defect is determined based on the luminance value for each partial area Judge .

According to a seventh aspect of the present invention, in the defect inspection apparatus according to the sixth aspect, the luminance value for each partial area is a value obtained by averaging the luminance in the partial area.

According to an eighth aspect of the present invention, in the defect inspection apparatus according to the sixth or seventh aspect, the image analysis unit determines the presence / absence of a defect using a neural network model that receives a luminance value for each partial region. To do.

As the effects of the present invention, the following effects are obtained.

According to the first aspect of the present invention, the contour of the subject can be represented by a virtual line by performing the Hough transform processing after extracting the contour of the subject. Then, in order to determine whether or not the contour of the subject is included in the region specified based on the shape of the subject expressed by the virtual line, it is possible to determine the presence or absence of a highly reliable defect. It becomes possible.

According to the second and third aspects of the present invention, the contour of the subject can be expressed by a virtual line by performing the Hough transform processing after extracting the contour of the subject. In addition, the image obtained by imaging within the area specified based on the shape of the subject represented by the virtual line is divided into a plurality of partial areas, and the luminance value for each partial area is taken into consideration, so reliability It is possible to determine whether there is a high defect.

According to the fourth aspect of the present invention, since a neural network model that is learned to be suitable for defect inspection is input using the luminance value for each partial region, the presence or absence of a highly reliable defect is determined. Is possible.

According to the fifth aspect of the present invention, the contour of the subject can be expressed by a virtual line by performing the Hough transform processing after extracting the contour of the subject. Then, in order to determine whether or not the contour of the subject is included in the region specified based on the shape of the subject expressed by the virtual line, it is possible to determine the presence or absence of a highly reliable defect. It becomes possible.

According to the sixth and seventh aspects of the present invention, the contour of the subject can be expressed by a virtual line by performing the Hough transform processing after extracting the contour of the subject. In addition, the image obtained by imaging within the area specified based on the shape of the subject represented by the virtual line is divided into a plurality of partial areas, and the luminance value for each partial area is taken into consideration, so reliability It is possible to determine whether there is a high defect.

According to the eighth aspect of the present invention, since a neural network model that is learned to be suitable for defect inspection is input using the luminance value of each partial region, the presence / absence of a highly reliable defect is determined. Is possible.

1 is an external view showing the overall configuration of a defect inspection apparatus according to the present invention. Schematic which shows the whole structure of the defect inspection apparatus which concerns on this invention. External view of end mill. The figure which shows the inspection flow of the defect inspection apparatus which concerns on this invention. The figure which shows the operation | movement flow in an automatic focus adjustment step. The figure which shows creation of frequency conversion image data. The figure which shows the process of focusing using the total of a luminance value. The figure which shows the operation | movement flow of the 1st image analysis process in an image analysis step. The figure which shows an outline extraction image and Hough conversion image data. The figure which shows the process of determining the presence or absence of a defect. The figure which shows the operation | movement flow of the 2nd image analysis process in an image analysis step. The figure which shows the process of dividing | segmenting an image into a some partial area | region. The figure which shows the process of judging the presence or absence of the defect by a neural network model. The figure which shows the monitor image which displayed the result of the defect inspection. The figure which shows the aspect of the defect which can be inspected by the defect inspection method and defect inspection apparatus which concern on this invention. The figure which shows the correct answer rate of the defect inspection apparatus which concerns on this invention.

DESCRIPTION OF SYMBOLS 1 Defect inspection apparatus 2 Imaging device (imaging means)
3 Moving Device 4 Lifting Device 5 Rotating Device 6 Control Device 7 Reflector 8 End Mill (Subject)
8A Cutting edge part 8a Cutting edge 9i Virtual line 9s Virtual circle 9t Virtual circle 10 Image processing part 11 Image analysis part 11a First image analysis part 11b Second image analysis part 12 Output part 13 Control part 14 Storage part 15 Automatic focus adjustment part

Next, an embodiment of the invention will be described.

First, the defect inspection apparatus 1 according to the present invention will be described with reference to FIG. 1 and FIG. FIG. 1 is an external view showing the overall configuration of a defect inspection apparatus 1 according to the present invention. FIG. 2 is a schematic diagram showing the overall configuration of the defect inspection apparatus 1 according to the present invention. In addition, it shows in FIG. 1 by making the action direction of gravity into the up-down direction.

The defect inspection apparatus 1 is an apparatus that captures an object to be provided and determines the presence or absence of a peeling defect or the like generated on the surface of the object from the obtained image. The defect inspection apparatus 1 mainly includes an imaging device 2 as an imaging means, a moving device 3 of the imaging device 2, a subject lifting device 4, a subject rotating device 5, and image processing and presence / absence of defects. And a control device 6 that performs the above determination and the like.

The imaging apparatus 2 is an apparatus for imaging the provided subject. In the defect inspection apparatus 1, a C-MOS camera is adopted as the photographing apparatus 2. The imaging device 2 images the subject as a color image and outputs the obtained image to the control device 6. The defect inspection apparatus 1 includes a plurality of illumination devices that illuminate the subject, the reflection plate 7, and other diffuser plates that make illumination light uniform, so that a good subject image can be obtained. Has been.

The moving device 3 is a device that enables the imaging device 2 to move in order to adjust the distance between the subject and the imaging device 2. The moving device 3 can adjust the focus of the imaging device 2 regardless of the shape of the subject by adjusting the distance between the subject and the imaging device 2. However, for example, a configuration in which the focal length of the photographing apparatus 2 can be adjusted while the photographing apparatus 2 is fixed may be employed.

The elevating device 4 is a device that enables the attached subject to move in the vertical direction. The elevating device 4 allows the imaging device 2 to image an arbitrary position in the vertical direction of the subject by moving the subject in the vertical direction. However, for example, the imaging apparatus 2 may move in the vertical direction so that an arbitrary position in the vertical direction of the subject can be imaged.

Rotating device 5 is a device that can rotate about the vertical direction of the provided subject as a central axis. The rotating device 5 allows the imaging device 2 to image an arbitrary position in the circumferential direction of the subject by rotating the subject in the circumferential direction. Note that the stepping motor that constitutes the rotating device 5 and drives the rotating device 5 can be freely operated by the control device 6 to start and stop rotating.

The control device 6 mainly includes an image processing unit 10, an image analysis unit 11, an output unit 12, a control unit 13, and a storage unit 14. The image processing unit 10 processes an image photographed by the photographing device 2. The image analysis unit 11 includes a first image analysis unit 11a and a second image analysis unit 11b, and determines the presence / absence of a defect based on information from the image processing unit 10. The output unit 12 outputs the result obtained by analysis by the image analysis unit 11 to a monitor or the like. The control unit 13 controls the moving device 3 and the like based on information from the image processing unit 10. The storage unit 14 stores an image obtained by photographing. In addition, the automatic focus adjustment unit 15 in the defect inspection apparatus 1 is a focus adjustment mechanism by autonomous control configured by the image processing unit 10, the control unit 13, the imaging device 2, the moving device 3, the lifting device 4, and the rotating device 5. Say.

With the configuration as described above, the defect inspection apparatus 1 can adjust the distance between the subject and the imaging apparatus 2 by the moving apparatus 3, so that the imaging apparatus 2 can be focused. Further, in the defect inspection apparatus 1, since the lifting device 4 can move the subject in the vertical direction and the rotating device 5 can rotate the subject, the imaging device 2 can photograph an arbitrary position of the subject. Is possible. The defect inspection apparatus 1 can determine the presence / absence of a defect by analyzing an image of the subject imaged by the imaging apparatus 2.

In addition, below, the case where the end mill 8 which is a typical cutting tool is used as a test object is demonstrated. As shown in FIG. 3, the end mill 8 is a cutting tool having a plurality of cutting edge portions 8A formed in a spiral shape on the outer periphery thereof. The end mill 8 cuts the workpiece by being rotated about its axis.

The inspection flow of the defect inspection apparatus 1 according to the present invention will be described with reference to FIG.

First, the defect inspection apparatus 1 according to the present invention controls the position of the end mill 8 that is a subject as a placement step (step S110). This is a step of adjusting the vertical position and rotation phase of the end mill 8 by transmitting control signals from the control unit 13 of the control device 6 to the lifting device 4 and the rotating device 5. Thereby, it becomes possible to image | photograph the arbitrary positions of the end mill 8 with the imaging device 2. FIG.

Next, as the automatic focus adjustment step, the focus of the photographing apparatus 2 is adjusted (step S120). This is a step in which the control unit 13 adjusts the distance between the end mill 8 and the imaging device 2 by transmitting a control signal to the moving device 3 based on the frequency converted image data created by the image processing unit 10. . As a result, the photographing apparatus 2 can be focused.

Then, as a first image analysis step in the image analysis step, it is determined whether or not there is a defect in the cutting edge 8a of the end mill 8 (step S130). This is a step in which the image analysis unit 11 determines whether or not the cutting edge 8a has a peeling defect or the like based on the Hough transform image data or the like created by the image processing unit 10.

Also, as a second image analysis step performed simultaneously with the first image analysis step (step S130), the presence / absence of a defect using a neural network model is determined (step S140). This is a step in which the image analysis unit 11 determines whether or not the cutting edge 8a has a peeling defect or the like based on the Hough transform image data or the like created by the image processing unit 10.

In addition, in this defect inspection apparatus 1, it is comprised so that a 1st image analysis process (step S130) and a 2nd image analysis process (step S140) may be performed simultaneously. However, a configuration may be adopted in which one of them is performed first and the other is performed thereafter. Further, the order is not limited. Furthermore, it is good also as a structure which performs only any one.

After that, the processes from step S110 to step S140 are performed in all the inspection areas of the end mill 8, and when it is determined that all the inspection areas are completed, the inspection flow is finished.

Here, the automatic focus adjustment step (step S120) will be described in detail with reference to FIG.

The automatic focus adjustment step (step S120) includes a photographing operation, an extraction processing operation, a frequency domain conversion operation, a sum calculation operation, a distance specifying operation, and a focusing operation.

The imaging operation is an operation for imaging while moving the imaging device 2 so that the distance between the subject end mill 8 and the imaging device 2 is gradually separated (step S121). The defect inspection apparatus 1 is configured such that photographing is automatically performed every time the photographing apparatus 2 moves 0.1 mm.

The extraction processing operation first performs so-called gray scale representing each color image captured by the photographing operation (step S121) with black and white light and dark, and then applies a Prewitt filter or the like to the gray scale black and white image. This is an operation for extracting the contour of the blade edge 8a by using a differential filter such as a Sobel filter, so-called edge extraction filter (step S122). The defect inspection apparatus 1 employs a Prewitt filter.

The frequency domain transform operation is an operation for creating frequency transformed image data by performing discrete Fourier transform processing on each contour extracted image created by the extraction processing operation (step S122) (step S123). The discrete Fourier transform process replaces a two-dimensional image in the horizontal direction and the vertical direction with a waveform of a luminance value that changes in each direction, and develops the waveform into a sine wave and a cosine wave having different frequencies and amplitudes. That is, by performing discrete Fourier transform processing, an m × n image can be expressed by a frequency region of k × l. Formula 1 for performing the discrete Fourier transform processing is shown in Equation 1.

As described above, the defect inspection apparatus 1 processes the image of the cutting edge 8a of the end mill 8 photographed by the photographing apparatus 2 to create frequency conversion image data expressed in the frequency domain. 6A is a captured image of the cutting edge 8a, and FIG. 6B is a contour extraction image of the cutting edge 8a. FIG. 6C shows frequency converted image data of the blade edge 8a.

Further, the defect inspection apparatus 1 is characterized in that it performs frequency transform image data by performing discrete Fourier transform processing without performing discrete cosine transform processing. This is because the discrete cosine transform process has an advantage that the processing speed is high, but the discrete Fourier transform process can create more precise frequency transformed image data. Since this defect inspection apparatus 1 performs a defect inspection of an end mill cylinder having a complicated shape and a metallic luster, high accuracy is ensured by performing a discrete Fourier transform process.

The sum calculation operation is an operation for calculating the sum LS of luminance values from the frequency converted image data created by the frequency domain conversion operation (step S123) (step S124). The total luminance value LS is calculated by the calculation formula shown in FIG. 7A.

The distance specifying operation is an operation of finding the distance L between the end mill 8 and the photographing apparatus 2 that maximizes the sum LS of the luminance values calculated by the sum calculation operation (step S124) (step S125). The horizontal axis in FIG. 7B indicates the distance L between the end mill 8 and the imaging device 2, and the vertical axis in FIG. 7B indicates the total luminance value LS. “R1” in FIG. 7B indicates the result when the frequency domain conversion operation (step S123) is performed after the extraction processing operation (step S122). Note that accuracy is improved by performing function approximation on the sum LS of luminance values that change depending on the distance between the end mill 8 and the imaging device 2. From FIG. 7B, the distance Lx between the end mill 8 and the photographing apparatus 2 that maximizes the total luminance value LS is the position where the photographing apparatus 2 is in focus.

This is because the distance L between the end mill 8 and the photographing apparatus 2 is closer than the position where the photographing apparatus 2 is in focus (L1) or the distance L between the end mill 8 and the photographing apparatus 2 is from the position where the photographing apparatus 2 is in focus. If the distance is too far (L2), the image photographed by the photographing device 2 is blurred, so that the contour of the blade edge 8a is not clearly extracted even by the extraction processing operation (step S122), and as a result, the sum of the luminance values becomes small. by.

That is, at the distance Lx between the end mill 8 and the photographing device 2 where the sum of luminance values is maximum, the image photographed by the photographing device 2 is clear, that is, the photographing device 2 is in focus. ing. FIG. 7B shows “R2” that is the result of performing only the extraction processing operation (step S122) and “R3” that is the result of performing only the frequency domain conversion operation (step S123). “R1”, which is the result of performing the frequency domain conversion operation (step S123) after the processing operation (step S122), most clearly indicates the focal length of the photographing apparatus 2. 7C is a captured image of the cutting edge 8a, and FIG. 7D is an outline extraction image of the cutting edge 8a. FIG. 7E shows frequency converted image data of the cutting edge 8a.

Thereafter, as the focusing operation, the photographing apparatus 2 is moved so as to be the distance Lx found by the distance specifying operation (step S125) (step S126), and the automatic focusing step (step S120) is completed.

As described above, even if the subject is an end mill 8 or the like having a complicated shape, the distance between the subject and the imaging device 2 is changed a plurality of times, and the obtained image is processed to obtain accuracy. It is possible to focus well.

Next, the first image analysis step (step S130) in the image analysis step will be described in detail with reference to FIG.

The first image analysis step (step S130) includes an extraction processing operation, a virtual line creation operation, and a defect determination operation.

The extraction processing operation first performs so-called gray scale representing each color image captured by the photographing operation (step S121) with black and white light and dark, and then applies a Prewitt filter or the like to the gray scale black and white image. This is an operation for extracting the contour of the cutting edge 8a by using a differential filter such as a Sobel filter, so-called edge extraction filter (step S131). The defect inspection apparatus 1 employs a Prewitt filter. The contour extraction image created in this way is shown in FIG. 9A.

The virtual line creation operation is an operation of expressing the contour of the blade edge 8a with two virtual lines 9i by performing a Hough transform process on the contour extracted image created by the extraction processing operation (step S131) (step S131). S132). Hough transform processing is one of feature extraction methods used as digital image processing. FIG. 9B shows the Hough transform image data created in this way.

Further, as shown in FIG. 9B, a virtual circle 9s having a predetermined diameter centered on the intersection of the virtual line 9i is created on the Hough transform image data in which the contour of the blade edge 8a is expressed by two virtual lines 9i. Is done. The diameter of the virtual circle 9s is determined by finding in advance by a test or the like how much the size of the defect will be when a defect occurs at the tip of the blade edge 8a.

The defect determination operation is an operation for determining whether or not there is a defect depending on whether or not the contour of the blade edge 8a in the contour extraction image is included in the circular area of the virtual circle 9s created by the virtual line creation operation (step S132) ( Step S133). This utilizes a defect mode in which there is a high possibility that the blade edge 8a is not included inside the virtual circle 9s when a defect occurs in the tip portion of the blade edge 8a. In FIG. 10, it is determined that there is a defect because the cutting edge 8a is not included in the region of the virtual circle 9s.

As described above, the contour of the cutting edge 8a can be expressed by the virtual line 9i by photographing the cutting edge 8a of the end mill 8 which is the subject and processing the obtained image. In order to determine whether or not the contour of the blade edge 8a in the contour extraction image is included in the circular area specified based on the shape of the blade edge 8a represented by the virtual line 9i, the presence or absence of a highly reliable defect This makes it possible to make a judgment.

Next, the second image analysis step (step S140) in the image analysis step will be described in detail with reference to FIG.

The second image analysis step (step S140) includes an extraction processing operation, a virtual line creation operation, a region division operation, a frequency domain conversion operation, an average calculation operation, and a defect determination operation.

In the extraction processing operation, first, so-called gray scale conversion is performed, in which each color image captured by the image capture operation (step S121) is represented only by a luminance value, and then a Prewitt filter or the like is applied to the gray scale monochrome image. This is an operation for extracting the contour of the blade edge 8a by using a differential filter such as a Sobel filter, so-called edge extraction filter (step S141). The defect inspection apparatus 1 employs a Prewitt filter. The contour extraction image created in this way is shown in FIG. 9A.

The virtual line creation operation is an operation of expressing the contour of the blade edge 8a with two virtual lines 9i by performing a Hough transform process on the contour extraction image created by the extraction processing operation (step S141) (step S1). S142). Hough transform processing is one of feature extraction methods used as digital image processing. FIG. 9B shows the Hough transform image data created in this way.

Also, as shown in FIG. 12, a virtual circle 9t having a predetermined diameter centered on the intersection of the virtual line 9i is created on the Hough transform image data in which the contour of the blade edge 8a is expressed by two virtual lines 9i. Is done. The diameter of the virtual circle 9t is determined by finding in advance a place where the defect is likely to occur and the size of the defect when a defect occurs in the cutting edge 8a.

The area dividing operation is performed in the circular area of the virtual circle 9t created by the virtual line creating operation (step S142) and the

virtual circles

9u, 9v,. This is an operation of dividing an image obtained by photographing into regions divided by

line segments

9j, 9k,... Divided at equal intervals so as to pass through the intersection of the lines 9i (step S143). As shown in FIG. 12, in this defect inspection apparatus 1, it is divided into 17 partial areas, but the specific number is not limited.

The frequency domain transforming operation is an operation of creating frequency transformed image data by performing discrete Fourier transform processing on the image segmented by the region segmenting operation (step S143) (step S144). The discrete Fourier transform process replaces a two-dimensional image in the vertical direction and the horizontal direction with a waveform of a luminance value changing in each direction, and develops the waveform into a sine wave and a cosine wave having different frequencies and amplitudes. That is, an image can be expressed in the frequency domain by performing discrete Fourier transform processing. In the defect inspection apparatus 1, since the image obtained by the photographing apparatus 2 is a color image, the image divided by the region dividing operation (step S143) is a monochrome image after being grayscaled.

The average calculation operation is an operation for calculating an average value of luminance from the frequency converted image data created by the frequency domain conversion operation (step S144) (step S145). In the defect inspection apparatus 1, the luminance of all pixels is averaged in each partial region in order to reduce misjudgment about the presence / absence of a defect.

The defect determination operation is an operation for determining the presence / absence of a defect by using a neural network model NN that receives an average value of luminance for each partial region (step S146). As shown in FIG. 13C, the neural network model NN is a hierarchical neural network having a three-layer structure including an input layer IL, an intermediate layer HL, and an output layer OL. As for the weights of the input layer IL and the intermediate layer HL and the weights of the intermediate layer HL and the output layer OL, a desired output is obtained from teacher data (an image of the end mill 8 having a defect and an image of the end mill 8 having no defect) prepared in advance. Has been learned to obtain.

That is, the presence / absence of a defect is determined by using a neural network model NN that is trained so as to be suitable for defect inspection, using an average value of luminance for each partial region as an input. FIG. 13A is a black and white image of the divided cutting edge 8a, and FIG. 13B is frequency-converted image data of one partial region.

Formula 2 for learning the neural network model NN is shown in Equation 2. Weight W _kj of the output o _k and the teacher output t _k and the input layer so that the value of the error E is minimized representing an error of the intermediate layer the weight W _ji and the intermediate layer and the output layer of is determined.

As described above, the contour of the cutting edge 8a can be expressed by the virtual line 9i by photographing the cutting edge 8a of the end mill 8 which is the subject and processing the obtained image. Then, by using a neural network model that divides a black and white image in a circular area specified based on the shape of the cutting edge 8a represented by the virtual line 9i, and inputs an average value of luminance for each partial area, This makes it possible to determine whether there is a highly reliable defect.

FIG. 14 shows a monitor image displaying the result of defect inspection. In the monitor image, in addition to the presence or absence of defects (see a part in the figure), the image processing process (see part b in the figure), buttons for instructing execution of a series of operations (see part c in the figure), and parameters are changed. An operation section (see section d in the figure) that can be displayed is displayed.

As described above, it is possible to focus with high accuracy by changing the distance L between the subject end mill 8 and the imaging device 2 a plurality of times and processing the obtained image. Further, by processing the photographed image, the contour of the blade edge 8a can be expressed by a virtual line 9i, and a defect is determined depending on whether or not the contour of the blade edge 8a is included in an area specified from the virtual line 9i. Can be determined. Further, the black-and-white image after gray scale is divided within the area specified based on the virtual line 9i, and the presence / absence of a defect is detected by using a neural network model NN that receives an average value of luminance for each partial area. Can be judged.

Thus, in the defect inspection apparatus 1 that embodies the defect inspection method according to the present invention, it is possible to inspect the defect mode as shown in FIG. As shown in FIG. 16, the defect inspection method and the defect inspection apparatus 1 according to the present invention can determine the presence or absence of a highly reliable defect.

The present invention can be used in a technique for determining the presence or absence of a peeling defect or the like that has occurred in a subject.

Claims

In a defect inspection method for determining the presence or absence of a defect from an image obtained by imaging a subject with an imaging means,
Comprising an image analysis step of determining the presence or absence of a defect in the subject by processing the obtained image;
In the image analysis step, the contour of the subject is extracted from an image obtained by imaging, and the contour of the subject is obtained by performing a Hough transform process on the contour extracted image obtained by the extraction processing. It is expressed by a plurality of virtual lines, and whether or not there is a defect is determined by whether or not the contour of the subject is included in a circular area having a predetermined diameter centered on the intersection of the virtual lines. Defect inspection method.
In a defect inspection method for determining the presence or absence of a defect from an image obtained by imaging a subject with an imaging means,
Comprising an image analysis step of determining the presence or absence of a defect in the subject by processing the obtained image;
In the image analysis step, the contour of the subject is extracted from an image obtained by imaging, and the contour of the subject is obtained by performing a Hough transform process on the contour extracted image obtained by the extraction processing. An image represented by a plurality of virtual lines and obtained by photographing within a circular area having a predetermined diameter centered on the intersection of the virtual lines is divided into a plurality of partial areas, and each image divided into the partial areas The defect is converted to a frequency domain by performing a discrete Fourier transform process on the image, a luminance value in the frequency domain is calculated, and the presence or absence of a defect is determined based on the luminance value of each partial area Inspection method.
3. The defect inspection method according to claim 2, wherein the luminance value for each partial area is a value obtained by averaging the luminance in the partial area.
4. The defect inspection method according to claim 2, wherein the image analysis step determines the presence / absence of a defect using a neural network model that receives a luminance value for each of the partial areas.
In a defect inspection apparatus that determines the presence or absence of defects from an image obtained by imaging a subject with an imaging means,
An image analysis unit that determines the presence or absence of defects in the subject by processing the obtained image,
The image analysis unit extracts the contour of the subject from an image obtained by imaging, and performs a Hough transform process on the contour extracted image obtained by the extraction processing, thereby determining the contour of the subject. It is expressed by a plurality of virtual lines, and whether or not there is a defect is determined by whether or not the contour of the subject is included in a circular area having a predetermined diameter centered on the intersection of the virtual lines. Defect inspection equipment.
In a defect inspection apparatus that determines the presence or absence of defects from an image obtained by imaging a subject with an imaging means,
An image analysis unit that determines the presence or absence of defects in the subject by processing the obtained image,
The image analysis unit extracts the contour of the subject from an image obtained by imaging, and performs a Hough transform process on the contour extracted image obtained by the extraction processing, thereby determining the contour of the subject. An image represented by a plurality of virtual lines and obtained by photographing within a circular area having a predetermined diameter centered on the intersection of the virtual lines is divided into a plurality of partial areas, and each image divided into the partial areas The defect is converted to a frequency domain by performing a discrete Fourier transform process on the image, a luminance value in the frequency domain is calculated, and the presence or absence of a defect is determined based on the luminance value of each partial area Inspection device.
The defect inspection apparatus according to claim 6, wherein the luminance value for each partial area is a value obtained by averaging the luminance in the partial area.
The defect inspection apparatus according to claim 6 or 7, wherein the image analysis unit determines the presence or absence of a defect using a neural network model that receives a luminance value for each of the partial areas.