KR20190011199A - System for defect inspection and method for defect inspection - Google Patents

Abstract

According to a defect inspection system (1), 2D images (F) (t1) to (F) (tm) in which luminance is changed in a return direction (X) is imaged by an imaging unit (3). 2D images (F) (t1) to (F) (tm) and the like are divided into lines (L1) (t1) to (Lk) (tm) and the like by a line division processing unit (9). Thus, lines (L1) (t1) to (Lk) in the same position of 2D images (F) (t1) to (F) (tm) are processed into image data of line division images (DL1) (t1) to (DLk) (t(1-(k−1))) arranged in parallel in time series. Even in an imaged image of the same test object (T), line division images (DL1) (t1) and the like have different brilliances. On the basis of data obtained by accumulating results of machine learning regarding the identification of the species of defects included in line division images (DL1) (t1) to (DLk) (t(1-(k-1))) having different brilliances and viewing types, identification of the species of defects of a test object (T) is performed by a defect species identification unit (10). Thus, it is possible to improve identification accuracy.

Description

FIELD OF THE INVENTION [0001] The present invention relates to a defect inspection system and a defect inspection method,

The present invention relates to a defect inspection system and a defect inspection method.

A defect inspection system for inspecting defects to be inspected based on a sensed image to be inspected, includes a defect inspection system for detecting defects such as, for example, optical films such as polarizing films and retardation films, Is known. This type of defect inspection system carries a film in a carrying direction, captures a two-dimensional image of the film at each discrete time, and conducts defect inspection based on the two-dimensional image picked up. For example, Japanese Patent No. 4726983 discloses a system in which a two-dimensional image is divided into a plurality of lines arranged in parallel in the transport direction, and the lines at the same position in each two-dimensional image picked up at each discrete time are juxtaposed in time series And generates a line segment image. The line segment image is processed into a defect emphasized image emphasizing a luminance change. The presence or position of the defect or position of the film is easily specified by the defect emphasized image.

However, even if a two-dimensional image to be inspected is processed as a defect-emphasized image as in the above-described technique, finally, defect identification is performed by human judgment, and there is room for improvement in defect identification accuracy.

SUMMARY OF THE INVENTION It is an object of the present invention to provide a defect inspection system and a defect inspection method capable of improving defect identification accuracy.

According to the present invention, there is provided an imaging apparatus comprising: a light source for irradiating light to an object to be inspected; an imaging unit for imaging a two-dimensional image irradiated from the light source to the object to be inspected, A transport section for transporting the object to be inspected relatively in the transport direction; and an image processing section for processing the image data of the two-dimensional image picked up by the image pickup section, wherein the image pickup section is moved in the direction coinciding with the transport direction in the two- Dimensional image in which the luminance changes, and the image processing section divides the two-dimensional image into a plurality of lines arranged in parallel in the transport direction so that the line at the same position in each of the two- A line segmentation processing section for processing the image data of the line segment image in parallel in time series; And a defect type identifying section that identifies the type of defect to be inspected based on the data accumulated as a result of the machine learning relating to the identification of the type of defect included in the image.

According to this configuration, there are provided a light source for irradiating light to the object to be inspected, an imaging section for imaging the two-dimensional image irradiated from the light source to the object to be inspected and transmitted or reflected from the object for inspection at each discrete time, A defect inspection system comprising: a transport section for transporting an object to be inspected relatively in a transport direction; and an image processing section for processing image data of a two-dimensional image picked up by the image pickup section, Dimensional image in which the luminance changes in a direction coinciding with the carrying direction of the two-dimensional image, and the two-dimensional image is divided into a plurality of lines arranged in parallel in the carrying direction by the line division processing unit of the image processing unit, The image data of the line-segmented image in which the lines at the same position in each of the two-dimensional images picked up in parallel are processed in a time series order, Each work inspection target may be a captured image line-divided image is the image having different luminance. Further, the defect type identifying unit of the image processing unit may be configured to classify, based on the data obtained by accumulating the result of the machine learning relating to the identification of the type of defect included in the line segment image having two or more different luminance levels respectively processed by the line segmentation processing unit The type of defect to be inspected is identified. Therefore, even if an image of the same inspection object is detected, the type of defect is identified based on the result of the machine learning on two or more line segment images different in luminance Therefore, the accuracy of defect identification can be improved.

In this case, the defect type identifying section identifies the type of the defect to be inspected based on the data that accumulates the result of the machine learning relating to the identification of the type of the defect included in the two or more line segment images whose luminance is different by 10% or more Suitable.

According to this configuration, the defect type identifying section identifies the type of defect to be inspected based on the data that accumulates the result of the machine learning relating to the identification of the type of defect included in the two line-split images whose luminance is 10% or more different The type of the defect is identified based on the result of the machine learning on the two line-split images whose luminance greatly differs by 10% or more even in the case of the image of the same inspection object, The defect identification accuracy can be further improved.

The light shielding body further includes a light shielding member which is located between the light source and the inspection target and shields a part of the light irradiated from the light source to the inspection target to thereby form the name portion and the arm portion in the two- Wherein the defect type identification unit is configured to carry the inspection target relative to the light source, the light shielding body, and the imaging unit relatively in the conveying direction crossing the boundary line of the list portion and the dark portion, Based on the data obtained by accumulating the result of the machine learning relating to the identification of the type of defect included in the line segment image obtained by parallelizing the lines of the positions of the arm portions in the two- It is appropriate to identify the type of the object.

According to this configuration, the light shielding body located between the light source and the inspection object shadows a part of the light irradiated from the light source to the inspection object, so that the list portion and the dark portion are formed in the two- The light source, the light shielding body, and the imaging unit are relatively transported in the transport direction in which the inspection target crosses the boundary line between the bright portion and the dark portion. Therefore, in each of the two-dimensional images picked up at each discrete time, Enters both the list and the dark areas. Further, the defect type identifying section may be a defect type identifying section that includes a line segment image in which lines of the position of the list in the two-dimensional image are arranged in a time series order and a line segment image in which the positions of the arm sections in the two- Since the type of defect to be inspected is identified on the basis of data obtained by accumulating the result of the machine learning relating to the identification of the type of defect, it is possible to identify the types of defects to be inspected on two line segment images belonging to each of the list part and the dark part, By identifying the type of defect based on the result of the machine learning, it is possible to further improve the defect identification accuracy.

According to another aspect of the present invention, there is provided a defect inspection system comprising: an irradiation step of irradiating light to an object to be inspected from a light source of a defect inspection system; and an irradiation step of irradiating the object to be inspected from the light source, An image pickup step of picking up a two-dimensional image by light at each discrete time; a transporting step of relatively transporting the inspection object to the light source and the image pickup unit in the transport direction by the transporting unit of the defect inspection system; And an image processing step of processing the image data of the two-dimensional image picked up in the image pickup step by the processing unit, wherein in the image pickup step, the two-dimensional image in which the luminance changes in the direction coinciding with the transport direction in the two- In the image processing step, the two-dimensional image is divided into a plurality of lines arranged in parallel in the carrying direction, A line dividing processing step of processing lines of the same position in each of the two-dimensional images picked up each time as image data of line-divided images obtained by paralleling in a time series order; And a defect type identifying step of identifying the type of defect to be inspected based on the data accumulated as a result of the machine learning relating to the identification of the type of defect.

In this case, in the defect type identifying step, the type of the defect to be inspected is identified based on the data obtained by accumulating the result of the machine learning relating to the identification of the type of defect included in the two line segment images whose luminance is 10% or more different Is suitable.

In the irradiation step, a light shielding member which is located between the light source and the inspection target and shields a part of the light irradiated from the light source to the inspection target forms a list portion and a dark portion in the two-dimensional image picked up at every discrete time in the image pickup process , In the conveying step, the inspection target is relatively conveyed to the light source, the light shielding body, and the imaging section in the conveying direction crossing the boundary between the list portion and the dark portion, and in the defect type discriminating step, the line of the position of the list portion in the two- Based on data obtained by accumulating the results of the machine learning relating to the identification of the types of defects included in the line segment images obtained by paralleling the lines of the positions of the arm portions in the two- It is appropriate to identify the type of defect to be inspected.

1 is a perspective view showing a defect inspection system according to an embodiment.
Fig. 2 is a view showing the arrangement of a light source, an image pickup section, a light shielding body and an object to be inspected in the defect inspection system of Fig. 1;
3 is a block diagram showing details of an image processing unit of the defect inspection system of FIG.
4 is a flowchart showing a process of a defect inspection method according to the embodiment.
5 is a flowchart showing the details of the image processing process of FIG.
6 (A), 6 (B), 6 C, 6 D, 6 E, 6 F, and 6 G show the images processed by the line division processing unit of the image processing unit of the defect inspection system of FIG. Fig.
FIG. 7A is a view showing a two-dimensional image in a time series, FIG. 7B is a view showing each line-divided image obtained by line-by-line sequencing the lines at respective positions, Fig. 7 is a view showing a position-aligned image in which the images are displaced so that each of the images indicates the same position of the object to be inspected.
8 is a diagram showing a convolutional neural network.

Hereinafter, a preferred embodiment of a defect inspection system and a defect inspection method of the present invention will be described in detail with reference to the drawings. 1 and 2, a defect inspection system 1 according to an embodiment of the present invention includes a light source 2, an image pickup unit 3, a carry unit 4, an image processing unit 5, A lens 6, a collimated optical lens 7, and a display device 8 are provided. The defect inspection system of the present embodiment detects defects in the inspection target T by using an optical film such as a polarizing film and a retardation film, a film such as a laminated film used in a separator of a battery as an inspection target T, The inspection target T extends in the carrying direction X of the carry section 4 and has a preset width in the width direction Y orthogonal to the carrying direction X. [ Defects generated in the inspection object T indicate different states from the desired state, and examples thereof include foreign objects, impact scratches, bubbles (such as those generated during molding), foreign matter bubbles Etc.), scratches, nicks (such as those caused by folding line marks), and stripes (such as those caused by differences in thickness). The defect inspection system 1 identifies the type of these defects. The defect inspection system 1 can specify on which side of the inspection target T the defect is occurring, in addition to the identification of the defect type.

As shown in Figs. 1 and 2, the light source 2 irradiates the inspection target T with light. The light source 2 is arranged to emit linear light parallel to the width direction (Y). The light source 2 is not particularly limited as long as it emits light that does not affect the composition and properties of the film to be inspected (T), such as a metal halide lamp, a halogen transmission light, or a fluorescent lamp.

The imaging unit 3 captures a two-dimensional image by the light irradiated to the inspection target T from the light source 2 and transmitted or reflected by the inspection target T at each discrete time. The imaging section 3 has a plurality of optical members and a photoelectric conversion element. The optical member is composed of an optical lens, a shutter, or the like, and images the light transmitted through the film to be inspected T on the surface of the photoelectric conversion element. The photoelectric conversion element is an area sensor composed of an image pickup element such as a CCD (Charge Coupled Device) or a CMOS (Complementary Metal-Oxide Semiconductor) for picking up a two-dimensional image. The image pickup section 3 may pick up any of a two-dimensional image having no color and a two-dimensional image having a color.

The carrying section 4 relatively conveys the inspection target T in the carrying direction X to the light source 2 and the imaging section 3. [ The carry section 4 includes, for example, a delivery roller and a delivery roller for transporting a film to be inspected T in the transport direction X, and measures the transport distance by a rotary encoder or the like. In the present embodiment, the conveying speed of the inspection target T by the conveying unit 4 is set to about 2 to 100 m / min in the conveying direction X. The conveying speed in the carry section 4 is set and controlled by the image processing section 5 or the like.

The image processing section 5 processes the image data of the two-dimensional image picked up by the image pickup section 3. [ The image processing unit 5 is not particularly limited as long as it performs image processing of two-dimensional image data. For example, a PC (personal computer) in which image processing software is installed, an FPGA (Field Programmable Gate An image capturing board on which an image sensor is mounted) can be applied.

The light shield 6 is disposed between the light source 2 and the inspection target T and shields a part of the light irradiated from the light source 2 to the inspection target T so that the light shield 6 is shielded by the imaging unit 3 at every discrete time A list portion and a dark portion are formed in the two-dimensional image to be captured. The light shielding body 6 picks up a two-dimensional image whose luminance changes in a direction coinciding with the carrying direction X in the two-dimensional image. More specifically, the carry section 4 is configured to cross the inspection target T with the boundary line between the list section and the dark section with respect to the light source 2, the parallel optical lens 7, the light shielding body 6 and the imaging section 3 And is transported relatively in the transport direction (X). In this embodiment, the boundary line is parallel to the width direction Y perpendicular to the carrying direction X. If the image pickup section 3 can capture a two-dimensional image whose luminance changes in the direction in which the two-dimensional image coincides with the carrying direction X, the light shielding body 6 may not be provided. The parallel optical lens 7 makes the traveling direction of light irradiated from the light source 2 to the inspection target T and the light shielding body 6 parallel. The parallel optical lens 7 can be constituted by, for example, a telecentric optical system.

The display device 8 connected to the image processing section 5 includes a PC (personal computer) or the like and displays the types of defects identified by the image processing section 5 on an LC (Liquid Crystal) display panel, a plasma A display panel, an EL (Electro Luminescence) display panel, and the like. Further, the image processing section 5 may have a display device for displaying the processed image.

Hereinafter, the details of the image processing section 5 will be described. As shown in Fig. 3, the image processing section 5 has a line division processing section 9 and a defect type identification section 10. As shown in Fig. The line dividing processing section 9 divides the two-dimensional image into a plurality of lines arranged in parallel in the carrying direction X and outputs a line at the same position in each of the two-dimensional images picked up by the image pickup section 3 at each discrete time Processed image data in a line-segmented image in parallel in time series. The defect type identification unit 10 identifies the type of defects included in two or more line segment images processed by the line segmentation processing unit 9 based on the data obtained by accumulating the results of the machine learning, Lt; RTI ID = 0.0 > defects < / RTI > The data that accumulates the results of the machine learning is stored in a storage device such as a hard disk of a PC including the defect type identifying section 10, and is updated with the result of the machine learning.

In the present embodiment, data obtained by accumulating the result of the machine learning relating to the identification of the type of defect included in the line segment image refers to a series of images captured at the discrete time in the internal image pickup section 3 of the defect inspection system 1 In addition to the data that accumulates the result of the machine learning relating to the identification of the type of defect included in the line segment image on which the two-dimensional image of the defect has been processed, And also includes data that accumulates the results of machine learning related to identification of the type. That is, in the present embodiment, in addition to the mode in which the type of defect is identified in the state where the machine learning is performed in the defect inspection system 1, in the defect inspection system 1, A mode in which the type of defect is identified based on data accumulated as a result of the machine learning separately manufactured outside the inspection system 1 is also included.

The defect type identification unit 10 identifies the types of defects in the inspection target T based on the data obtained by accumulating the results of the machine learning relating to the identification of the types of defects included in the two line segment images whose luminance is 10% Lt; / RTI > The defect type identifying section 10 is a device for discriminating the defect type identifying section 10 from the line dividing image in which the lines of the positions of the names in the two-dimensional image are juxtaposed in time series by the light shield 6, The type of defects in the inspection target T is identified based on the data obtained by accumulating the results of the machine learning relating to the identification of the types of defects included in the line segment images in parallel in time series.

Hereinafter, a defect inspection method according to the present embodiment will be described. An irradiation step of irradiating light to the inspection target T from the light source 2 of the defect inspection system 1 is performed (S1) as shown in Fig. 6A, in the irradiating step, a part of light irradiated from the light source 2 to the inspection target T is placed between the light source 2 and the inspection target T, Dimensional image F picked up at every discrete time in the imaging process by the light shield 6 of the inspection system 1 and the arm portion d with the boundary line b as a boundary, . 6 (A), the two-dimensional image F (t1) at time t = t1 is a region in which the light from the light source 2 is shielded by the light shielding body 6, X), the brightness in the two-dimensional image F (t1) becomes high. Further, the defect D on the film of the inspection target T is displayed on the two-dimensional image F (t1). Time t = t2, t3, ... , two-dimensional images (F) (t2), (F) (t3), and , And (F) (tm) (m is an arbitrary natural number).

As shown in Fig. 4, the imaging unit 3 of the defect inspection system 1 irradiates the inspection target T from the light source 2 by the irradiation step and transmits or reflects the inspection target T An image pickup step of picking up a two-dimensional image (F) t1 by light is performed every discrete time (S2). 6 (A), in the imaging step, a part of the light irradiated from the light source 2 to the inspection target T is shielded by the light shield 6, so that the two-dimensional image F a two-dimensional image F (t1) whose luminance changes in a direction coinciding with the carrying direction X at the time t1 is picked up. Time t = t2, t3 ... The two-dimensional images (F) (t2), (F) (t3), and , And (F) (tm).

4, the conveyance unit 4 of the defect inspection system 1 allows the light source 2 and the imaging unit 3 to be inspected relative to each other in the conveying direction X Is carried out (S3). 6A, in the carrying step, the inspection target T is irradiated to the light source 2, the parallel optical lens 7, the light shielding body 6, In the carrying direction X crossing the boundary line b of the arm portion d and the arm portion d. In this embodiment, the boundary line b is parallel to the width direction Y orthogonal to the carrying direction X, but the angle formed by the boundary line b and the carrying direction X may be other than 90 degrees. The boundary line b is not always strict and the boundary line b is a boundary between the brightest portion of the two-dimensional image F (t1) included in the list portion 1 and the two- Means a line in the middle of the portion having the smallest brightness of the image F. [

The image processing unit 5 of the defect inspection system 1 processes the image data of the two-dimensional images F (t1) to (F) (tm) picked up in the imaging process An image processing step is performed (S4). Hereinafter, the details of the image processing step will be described. 5, in the image processing step, the line division processing unit 9 of the image processing unit 5 of the defect inspection system 1 performs a line division processing step (S41). 6B, in the line dividing processing step, the line dividing processing section 9 divides the two-dimensional image F (t1) into a plurality of first lines ( (J and k are arbitrary natural numbers, j < RTI ID = 0.0 > k) < / RTI > The widths of the conveying direction X of the lines L1 (t1) to Lk (t1) are the widths of the time t1, the time t2, ... , Visual tj, ... , The distance to be inspected T is the same as the distance at which the inspection target T is transported in the transport direction X at one frame interval of each time tm. Time t = t2, t3 ... The two-dimensional images (F) (t2), (F) (t3), and , And (F) (tm) are also performed.

The line dividing processing section 9 divides the two-dimensional images F (t1) to (F) (tm) picked up at the discrete time in the imaging process into two-dimensional images F (L1) (t1), (L1), (t2) and the like at the same position in the time series image data are processed in the form of image data of a line segment image obtained by paralleling in a time series order. The first line segment image will be described as an example. As shown in Fig. 6 (C), the line dividing processing section 9 divides the two-dimensional images F (t1), (f) (t2), (F) The first line L1 (t1), (L1), (t2), (L1), (t3), ... on the most downstream side in the conveying direction X (In the transport direction X). As shown in FIG. 6D, the line dividing processing section 9 divides the first line L1 (t1) to (t2) in each of the two-dimensional images F (t1) to (F) L1) (tm) are arranged in a time series order to generate a first line segment image DL1 (t1).

As shown in FIG. 6 (E), FIG. 6 (F) and FIG. 6 (G), the line segmentation processing section 9 divides the two-dimensional images F (t1) The first line L1 (t1) to (L1) (tm), ..., , the j-th line (Lj) (t1) to (Lj) (tm), ... , and the k-th lines Lk (t1) to Lk (tm), the first line segment image DL1 (t1), ..., , the j-th line segment image DLJ (t1), ... , and a k-th line-divided image DLk (t1). As shown in FIG. 6E, the line segment image DL1 (t1) is the line (L1 (t1)) at the position of the list portion 1 in the two-dimensional images F ) (t1) to (L1) (tm) in parallel in time series. 6 (F), the line-divided image DLj (t1) is the image of the position in the vicinity of the boundary line b in the two-dimensional images F (t1) to F (tm) And the lines Lj (t1) to (L1) (tm) are arranged in parallel in time series. 6 (G), the line segment image DLk (t1) further includes a position of the arm portion d in the two-dimensional images F (t1) to F (tm) And the lines Lk (t1) to (Lk) (tm) are arranged in parallel in time series.

As shown in (E) to (G) of FIG. 6, the line segment images DL1 (t1) to DLk (t1) are divided into two-dimensional images F (T1) to (Lk) (t1) in the same position in each of the time-series images DL1 (t1) to ) To DLk (t1) indicate different positions of the inspection target T and the positions of the defects D in the line segment images DL1 (t1) to DLk (t1) are also deviated from each other. Thus, in the present embodiment, line-segmented images in which the lines at the same position in each of two-dimensional images picked up in different time ranges are arranged in a time-series manner are produced so that each of the line-split images is the same Position alignment is performed to indicate the position.

As shown in Fig. 7 (A), two-dimensional images F (t1) to (F) (tm) are picked up at discrete time in the imaging process. The positions of the defects D in the two-dimensional images F (t1) to (F) (tm) are shifted from each other because the inspection target T is transported in the transport direction X. The line segment images DL1 (t1) to DLj (t1) to DLk (t1) are generated as shown in Fig. 7 (B). Since the line segment images DL1 (t1) to DLk (t1) in the same time range represent different positions of the inspection target T, the line segment images DL1 (t1) to DLk The positions of the defects D in the defects are also deviated from each other.

The first line L1 (t1) to the first line (tm) from the downstream side in the carrying direction X is shifted from the downstream side in the carrying direction X of the same time range, for example, (Lj) (t1) to (Lj) (tm) on the upstream side in the carrying direction X of the inspection target T by the distance that the inspection target T is transported at the frame interval of (j- Respectively. Therefore, as shown in Fig. 7C, the line segment image DL1 (tm) of the first line L1 (tm) to L1 (t (m + (m-1) (J- 1)) or the time (m-1), which has been traced by the time of the frame interval of (j-1) with respect to the range of the time t1 to the time tm, the line segment image DLj (t (m- (j-1)) in the range of t (m + (mj)) represents the same position of the inspection target T.

Similarly, for the line segment image DL1 (tm) of the first line L1 (tm) to L1 (t (m + (m-1) (M- (k-1)) to time t (m + (mk)) of the range of the time t (m-1) from the time t1 to the time tm The line segment image DLk (t (m- (k-1))) indicates the same position of the inspection target T. [

Or the line segment image DL1 (t1) of the first line L1 (t1) to L1 (t (1 + (m-1) (J) of the line-divided image DLj (t (1- (j-1)) in the range of time t (1- (j-1) T). The line segment image DL1 (t1) of the first line L1 (t1) to L1 (t (1 + (m-1) The line segment image DLk (t (1- (k-1)) in the range of time t (1- (k-1)) to time t T). By shifting the range of the time in this manner, the alignment can be performed so that each of the line segment images represents the same position of the inspection target T.

When the amount of the positional deviation is known or when the size of the line-divided image is sufficiently large for the defect, the defect is always converged in the line-divided image. Therefore, It is possible to use it for learning. Therefore, in such a case, the alignment does not need to be performed.

As shown in Fig. 5, the defect type identification step is performed by the defect type identification unit 10 of the image processing unit 5 of the defect inspection system 1 (S42). In the defect type identifying step, the defect type identifying section 10 judges whether or not two or more line segment images DL1 (t1), ..., DL2 , (DLj) (t (1- (j-1))), ... (D) of the inspection target (T) on the basis of data obtained by accumulating the results of the machine learning relating to the identification of the types of defects included in the defect types (DLk) and (DL (k .

In the defect type identification step, the defect type identification unit 10 identifies the types of defects included in the two line segment images DL1 (t1) and DLk (t1) whose luminance is 10% or more different And identifies the type of the defect (D) of the inspection target (T) based on the data accumulated as a result of the learning. More specifically, in the defect type identification step, the defect type identification unit 10 identifies the line L1 ((L1)) of the position of the list portion 1 in the two-dimensional images F (t1) to (t1) to (L1) (t1) in which a line segment image DL1 (t1) and a line segment image L1 (tk) (1 (k-1))) to Lk (t (1+ (mk)) at the position of the arm portion (d) Defects of the inspection target T based on the data obtained by accumulating the results of the machine learning relating to the identification of the types of defects included in the line-divided image DLk (t (1- (k-1) D). In the machine learning, for example, a composite product neural network is performed. It is also possible to use a neural network other than the composite neural network or other methods as long as the type of the defect can be identified by the machine learning.

As shown in Fig. 8, the composite-articulated network 100 includes an input layer 110, a hidden layer 120, and an output layer 130. As shown in Fig. The input layer 110 is provided with the line segment images DL1 (t1) to DLk (t (1- (k-1)) processed in the line segmentation process by the image processing unit 5 of the defect inspection system 1, 1))) are input. The hidden layer 120 is composed of a composite layer 121 and 123 for performing image processing by a weighted filter and a pulling function for reducing the two-dimensional array output from the composite multiplication layers 121 and 123, Layer 122 and an overall bonding layer 124 in which the weighting factor n of each layer is updated. In the output layer 130, the identification result of the type of the defect (D) by the machine learning is output. In the composite neural network 100, the weight of each layer is learned by back propagating the error between the output identification result and the correct answer value in the reverse direction (R).

For example, a plurality of line-divided images are previously input to the image processing section 5 together with correct answers for discrimination of the type of the defect (D), so that they are included in the newly input line-divided image DL1 Whether or not the defect is a specific defect D is sequentially identified, and the identification result is output sequentially. The sequentially output identification result and the error of the correct answer are reversely propagated in the reverse direction (R), and the weighting coefficient n of each layer is sequentially updated and accumulated as data. Whether or not the types of defects included in the newly input line segment image DL1 (t1) are sequentially identified and the identification results are sequentially output in the state in which the weight of each phase is updated in sequence, and the identification result And the weighting coefficient n of each layer are sequentially updated and accumulated as data on the basis of the error of the correct answer is repeated so that the difference between the identification result and the correct answer is reduced and the identification accuracy of the type of the defect D is improved.

According to the present embodiment, there are provided a light source 2 for irradiating light to an inspection target T, a light source 2 for emitting light to the inspection target T from the light source 2, An image pickup section 3 for picking up images F at time t1 to t at every discrete time and an image pickup section 3 for subjecting the light source 2 and the image pickup section 3 to an inspection target T in the conveying direction X And an image processing section 5 for processing image data such as two-dimensional images F (t1) to (F) (tm) picked up by the image pickup section 3, Dimensional image (F) (t1) to (F) (tm) by the imaging unit 3 in the direction matching the carrying direction X The two-dimensional images F (t1) to F (tm) are picked up by the line division processing unit 9 of the image processing unit 5, ) tm and the like are arranged in parallel in the conveying direction X by a plurality of lines L1 (t1) to Lk (tm) (L1) through (L1) (L1 (t1) to L1 (L1)) at the same position in each of the two-dimensional images F captured during the discrete time by the image pickup unit 3 (t1) to (DLk) (t (1- (k-1))) in which the same inspection object is picked up in the time-series order Each of the line-divided images DL1 (t1) to DLk (t (1- (k-1)) becomes an image having different luminance. The defect type identifying section 10 of the image processing section 5 also extracts the line segment images DL1 (t1) to DLk (t (t)) having different intensities of two or more processed by the line segmentation processing section 9, Since the type of the defect D of the inspection target T is identified based on the data obtained by accumulating the result of the machine learning related to the identification of the type of defect included in the inspection target (1- (k-1)), (T1) to (DLk) (t (1- (k-1))), etc., whose luminance is different even in the case where the image of the object T is photographed, Since the type of the defect D is identified based on the result of the learning, the accuracy of identifying the defect D can be improved.

According to the present embodiment, the defect type identifying section 10 has two line segment images DL1 (t1) and DLk (t (1- (k-1)) with luminance different from each other by 10% The type of the defect D of the inspection target T is identified on the basis of the data obtained by accumulating the result of the machine learning relating to the identification of the type of the defect included in the inspection target T. Therefore, Based on the results of the machine learning on the two line segment images DL1 (t1) and DLk (t (1- (k-1)) in which the luminance is significantly different from 10% And the type of the defect D is identified, the identification accuracy of the defect D can be further improved.

According to the present embodiment, a part of the light irradiated from the light source 2 to the inspection target T is shielded by the light shield 6 positioned between the light source 2 and the inspection target T, The image portion 1 and the arm portion d are formed on the two-dimensional images F (t1) to (F) (tm) picked up at each discrete time in the light source 3, Since the inspection target T is conveyed relative to the light shielding body 6 and the imaging section 3 in the conveying direction X crossing the boundary line b between the shade portion 1 and the arm portion d, The respective portions of the inspection target T in a series of two-dimensional images F (t1) to (F) (tm) picked up at each discrete time enter both the list portion 1 and the arm portion (d). The defect type identifying section 10 also identifies the line L1 (t1) to (L1) (tm) of the position of the list portion 1 in the two-dimensional images F (t1) to (F) (T1) (t1) and t2 (t) are obtained by dividing a line segment image DL1 (t1) and a two-dimensional image F (t Divided line image DLk in which the lines Lk (t (1- (k-1)) through Lk (t (1+ since the type of the defect D of the inspection target T is identified on the basis of the data obtained by accumulating the result of the machine learning relating to the identification of the type of defect included in the inspection target T (t (1- (k-1))) (K) of the line segment images DL1 (t1) and DLk (t (1- (k-1)) belonging to the list portion 1 and the arm portion The type of the defect D is identified based on the result of the determination of the defect D, thereby further improving the identification accuracy of the defect D. [

Although the embodiments of the present invention have been described above, the present invention is not limited to the above-described embodiments, but is implemented in various forms. For example, in the above-described embodiment, the case where the inspection target T is a film has been mainly described. However, the defect inspection system and the defect inspection method of the present invention are applicable to, for example, It can be applied to charge quantity inspection. The defect inspection system 1 and the defect inspection method of the present embodiment can detect defects such as whether the liquid does not reach a desired position in the container or whether the liquid does not exceed a desired position in the container.

In addition, the defect inspection system 1 and the defect inspection method of the present embodiment can be applied to appearance inspection such as cracks and scratches of glass products in a production line. When a glass product has defects such as cracks or scratches, it is possible to extract defects by using those whose luminance is higher than other parts.

Claims (6)

A light source for irradiating an object to be inspected,
An image pickup section for picking up a two-dimensional image by the light irradiated to the inspection target from the light source and transmitted or reflected by the inspection target at every discrete time,
A conveying unit for relatively conveying the inspection object to the light source and the imaging unit in a conveying direction,
An image processing section for processing the image data of the two-dimensional image picked up by the image pickup section
And,
Wherein,
Dimensional image in which the luminance changes in a direction coinciding with the carrying direction in the two-dimensional image,
Wherein the image processing unit comprises:
Dimensional image is divided into a plurality of lines that are arranged in parallel in the transport direction and a line segment image in which the lines at the same position in each of the two- A line division processing unit for processing the image data of the image data,
A defect classification identifying unit for identifying the type of the defect to be inspected based on the data obtained by accumulating the result of the machine learning relating to the identification of the type of defect included in the two or more line segment images processed by the line segmentation processing unit,
And a defect inspection system. The defect inspection apparatus according to claim 1, wherein the defect type identifying section identifies a type of defect included in the two line segment images whose luminance is 10% or more different from the defect classification identification data, The type of the defect. The imaging apparatus according to any one of claims 1 to 3, further comprising: a light source which is located between the light source and the inspection object and shields a part of the light emitted from the light source to the inspection target, Further comprising a light shielding member for forming a light shielding portion and a light shielding portion,
Wherein,
The light source, the light shielding body, and the imaging unit, the inspection target is relatively conveyed in the conveying direction crossing the boundary line between the name portion and the arm portion,
Wherein the defect type identifying section includes a line segment image in which the lines at the position of the list portion in the two-dimensional image are arranged in a time series order and a line segment image in which the lines at the positions of the arm portions in the two- And identifies the type of the defect to be inspected based on data obtained by accumulating the result of the machine learning relating to the identification of the type of defect included in the divided image. An irradiating step of irradiating light to an object to be inspected from a light source of the defect inspection system,
An image capturing step of capturing a two-dimensional image by the light emitted by the light source onto the inspection target and transmitted or reflected by the inspection target by the imaging unit of the defect inspection system at each discrete time,
A transporting step of transporting the inspection object relative to the light source and the imaging unit in a transport direction by the transport unit of the defect inspection system;
An image processing step of processing the image data of the two-dimensional image picked up in the imaging step by the image processing unit of the defect inspection system
And,
In the imaging step,
Dimensional image in which the luminance changes in a direction coinciding with the carrying direction in the two-dimensional image,
In the image processing step,
Dimensional image obtained by dividing the two-dimensional image into a plurality of lines arranged in parallel in the conveying direction and arranging the lines at the same position in each of the two- A line segmentation process step of processing the image data,
A defect kind identification step of identifying a defect type of the object to be inspected based on data obtained by accumulating the result of the machine learning relating to the identification of the type of defect included in the two or more line segment images processed in the line segmentation processing step
And a defect inspection method. The method according to claim 4, wherein, in the defect kind identification step, the defect type identification step identifies the types of defects included in the two line segment images whose luminance is 10% or more different, A defect inspection method for identifying a defect type. The method according to claim 4 or 5, wherein in the irradiation step,
A light shielding member which is located between a light source and an inspection target and shields a part of light irradiated from the light source to the inspection target forms a list portion and an arm portion in the two-
In the carrying step,
The light source, the light shielding body, and the imaging unit, the inspection target is relatively conveyed in the conveying direction crossing the boundary line between the name portion and the arm portion,
In the defect type identifying step,
Dimensional image, a line segment image in which the line at the position of the list portion in the two-dimensional image is arranged in a time series order, and a defect included in a line segment image in which the line at the position of the arm portion in the two- And classifying the type of defect to be inspected on the basis of data accumulated as a result of the machine learning relating to the identification of the type of defect.

Images