TWI420098B - A defect inspection apparatus and method using the image data for defect inspection, a method for manufacturing the same, and a recording medium - Google Patents

Description

Processing device and method for image data for defect inspection, defect inspection device and method using same, manufacturing method of plate body using the same, and recording medium

The present invention relates to a processing apparatus and a processing method for image data for defect inspection for inspecting defects existing in a transparent plate-like body such as a glass plate, and a defect inspection device and a defect inspection method using the same, and using the same A method of manufacturing a plate-like body, and a recording medium readable by a computer for recording a program for processing an image data for performing defect inspection.

Currently, since glass sheets are used in electronic devices such as flat panel displays and thin film solar cells, there is a strong demand for glass sheets having thinner thicknesses, less bubbles or scratches, and few or no defects.

As a defect existing in the glass plate, the flaw formed on the surface of a glass plate is mentioned. For example, in the floating method, a long plate-like body having a fixed thickness is taken out from a melting furnace and conveyed on a driving roller. At this time, the surface of the glass plate is damaged by foreign matter adhering to the driving roller or minute projections on the driving roller, and the like, and scratches are generated. Since a plurality of driving rolls are used for the conveyance of the glass sheets, there are many opportunities for occurrence of minute scratches on the surface of the glass sheets. Since the above-mentioned flaws are periodically generated, since the prior art is not limited to the glass sheet, various methods for inspecting the periodic defects have been proposed in the inspection of the intermediate form of the long strip.

Patent Document 1 discloses a measurement method in which a defect data and a normal data are binarized in order to measure a period of a defect existing in the object to be inspected, and the distance between the defect data is obtained. Each distance is calculated by frequency to calculate a periodic component of the distance, and the basic period of the defect is extracted from the periodic component.

Patent Document 2 discloses a method of classifying defects detected in image data obtained by photographing a subject into periodic defects and aperiodic defects, and inspecting periodic defects.

[Previous Technical Literature]

[Patent Literature]

[Patent Document 1] Japanese Patent Special Fair No. 7-86474

[Patent Document 2] Japanese Patent Laid-Open Publication No. 2006-308473

However, in Patent Document 1, when the periodic component of the distance between the defect data is calculated, since the defect data and the noise data cannot be distinguished, it is difficult to efficiently calculate the periodic component. When the defect is small, the threshold of binarization must be lowered. Therefore, the amount of processing of the noise data as the defect data becomes extremely large, and the accuracy of calculation of the periodic component is further reduced.

On the other hand, in Patent Document 2, when the defects detected in the image data obtained by photographing the object to be inspected are classified into periodic defects and aperiodic defects, the image data is binarized and will be borrowed. The area obtained by binarization which is regarded as the portion of the defect is classified as a non-periodic defect. Therefore, a portion of the defect that is extremely small in size and is considered to be caused by the noise data is misidentified as a periodic defect.

That is, since the defects caused by the small scratches caused by the driving roller or the like on the glass plate are small, when the image data is binarized, it is difficult to reduce the threshold value in order to reduce the portion which is mistakenly regarded as the defect. The generated noise information is distinguished. Therefore, it is extremely difficult to discriminate the periodicity of the defects of the small flaw caused by the driving roller or the like.

Therefore, in order to solve the above problems, an object of the present invention is to provide a defect inspection capable of detecting the presence of a periodic defect even when a captured image contains a noise component when detecting a defect existing in a plate-like body such as a glass plate. The image processing device and the processing method, the defect inspection device and the defect inspection method thereof, the method of manufacturing the plate body using the inspection method or the defect inspection device, and the image data for performing the defect inspection A recording medium that can be read by a computer as a program of processing methods.

In order to achieve the above object, a mode 1 of the present invention provides a processing apparatus characterized in that the plate shape is inspected by using an image obtained by relatively moving a plate-like body in a specific direction and facing the plate-like body. A processing device for image data for defect inspection of a defect existing in a body; the processing device comprising: a processing unit that extracts a plurality of defect candidates from the image using the first signal threshold, and in the moving direction Among the plurality of extracted defect candidates, a defect candidate having the same position in the width direction orthogonal to the specific direction, that is, the moving direction is searched for, and the position of the defect candidate detected by the search in the moving direction of the plate-shaped body is obtained. And an interval between the position of the defect candidate adjacent to the defect candidate detected in the moving direction and the position in the moving direction is obtained by repeating the above processing to obtain a plurality of intervals, and the frequency of generating the plurality of intervals is obtained. When the frequency of occurrence of the interval of interest exceeds the set frequency threshold, it is determined that the plate body has a period in the moving direction a defect; the frequency threshold used in the processing unit is defined according to the interval of the above-mentioned attention, and when the two frequency thresholds are set to different values, the interval between the predetermined frequency thresholds is smaller than The frequency threshold is set in such a manner as to specify the above-mentioned interval of the frequency threshold of the smaller one.

According to a second aspect of the invention, the processing device of the aspect 1, wherein the processing unit includes a reference table indicating a relationship between a density of the defect candidate and the threshold of the first signal, so that a density of defect candidates is set. In the manner in which the target density is generated, the first signal threshold is set using the reference table, and the frequency threshold is a value that varies depending on the value of the target generation density in addition to the change in the interval of interest.

The third aspect of the present invention provides the processing apparatus of the above-described type 1 or type 2, wherein the frequency threshold used in the processing unit is defined in such a manner that noise components are randomly distributed in the area, and The noise component having the position in the width direction at the same position is used as the defect candidate, and the frequency of generation of the noise component with respect to the interval or the simulation image formed by the noise component is obtained analytically. The image of the noise component is used as a defect candidate, and the frequency of generation of the noise component with respect to the interval is obtained, and the frequency threshold is defined based on the obtained generation frequency.

Mode 4 of the present invention provides the processing apparatus of the above-described Type 3, wherein the noise component is generated such that the density of the noise component in the image varies depending on the area of the image.

The processing apparatus according to any one of the above aspects 1 to 4, wherein the processing unit divides the image of the search target for searching for the defect candidate into plural numbers in the width direction and the moving direction And forming a plurality of unit regions having the same size, and when the plurality of unit regions including the plurality of defect candidates are at the same position in the width direction, the defect candidates are set to be the same position in the width direction, and The above interval and the above-mentioned generation frequency are obtained.

The processing apparatus according to any one of the above aspects 1 to 5, wherein the width direction generating frequency distribution indicating the distribution of the generation frequency in the width direction is obtained, and the width direction is used The unevenness of the above-described generation frequency in the above-described width direction in the frequency distribution is generated, and the defect generation pattern is classified.

The present invention provides the processing apparatus according to any one of the above aspects 1 to 6, wherein the plate-like system has a continuous elongated shape in the moving direction; and the processing unit divides the plate-shaped body into The area of the plate-shaped body of the length is set as the inspection target of one unit, and the above-described determination is performed for a plurality of units.

According to a seventh aspect of the invention, the processing device of the seventh aspect, wherein the processing unit records a generation frequency distribution of the interval defined by the interval and the position in the width direction for the plurality of time series units, according to the recorded The generation frequency distribution defines the frequency of occurrence and the position in the width direction to determine the generation frequency, and expresses the generation frequency as time-series data, thereby displaying the defect generation information on the screen.

A mode 9 of the present invention provides the processing device of the above aspect 8, wherein the plurality of generation frequencies obtained by changing at least one of the positions of the interval and the width direction of the above-mentioned attention frequency are determined for the time series data of the frequency of occurrence. Time series data, overwritten in the same chart and displayed on the screen.

The processing apparatus according to any one of the above aspects 1 to 9, wherein the processing unit searches for and detects a defect candidate having the same position in the width direction in the moving direction, except that the processing unit is adjacent The defect candidate is used as the previous defect candidate to obtain the interval in the moving direction, and the interval between the plurality of preceding defect candidates in the moving direction is also obtained, and the above processing is repeated to obtain a plurality of intervals. The frequency at which the plurality of intervals are generated is determined. When the frequency of occurrence of the interval of interest exceeds the set frequency threshold, it is determined that the plate-like body has a periodic defect in the moving direction.

The processing apparatus according to any one of the above aspects 1 to 10, wherein the processing unit is referred to as a spacing interval when the interval between the intervals determined to have a periodic defect in the moving direction is referred to as a spacing interval Further, a region of interest including a defect having the pitch interval in the width direction is defined, and a detailed defect candidate is extracted from the image of the region of interest using the second signal threshold. a search area centered at a position at which the detailed defect candidate obtained from the extraction is separated from the pitch interval in the moving direction, and a detailed defect candidate is searched for using the second signal threshold in the search region, and the search is evaluated separately The detected detailed defect candidate and the attribute of the detailed defect candidate obtained by the extraction determine whether the region of interest includes the periodic detailed defect candidate in the moving direction based on the evaluation result.

The processing apparatus according to any one of the preceding aspects, wherein the processing unit searches for and detects a defect candidate having the same position in the width direction in the moving direction, and obtains the interval. And evaluating the attribute of the defect candidate detected or the similarity between the defect candidate and the specific defect candidate, and determining the interval when at least one of the attributes and the similarity satisfies the set condition.

A type 13 of the present invention provides a defect inspection apparatus characterized in that it is an inspector for a defect existing in a plate-like body, comprising: a light source that illuminates a surface of the plate-like body; and a camera An image of a plate-like body that is irradiated with light by the light source while being relatively moved with respect to the plate-like body; and a processing device according to any one of the above aspects 1 to 12; and the processing device The processing unit extracts the plurality of defect candidates from the image captured by the camera using the first signal threshold, and searches for the above-mentioned plurality of defect candidates in the moving direction. The direction in which the camera relatively moves with respect to the plate-like body, that is, the position in the width direction in which the moving direction is orthogonal is the candidate candidate.

According to a fourth aspect of the present invention, there is provided a method of processing a defect inspection for inspecting a defect existing in the plate-like body by using an image obtained by moving a plate-like body relative to each other in a specific direction while detecting a relative image. a method for processing image data; extracting a plurality of defect candidates from the captured image using the first signal threshold; searching for the specific direction, that is, the moving direction, from the extracted plurality of defect candidates in the moving direction The defect candidates having the same position in the width direction of the orthogonal direction are obtained, and the position of the defect candidate detected by the search in the moving direction is obtained, and the defect candidate adjacent to the defect candidate in the moving direction is obtained in the moving direction. The interval between the positions is obtained by repeating the above processing to obtain a plurality of intervals; determining the frequency of generation of the plurality of intervals; and determining the plate shape when the frequency of occurrence of the interval of interest exceeds the set frequency threshold The body has a periodic defect in the moving direction; the above frequency threshold is specified according to the interval noted above, when the two frequencies Values are different, so that the above-described predetermined frequency interval greater attention by one of the threshold values is less than a predetermined frequency smaller one embodiment the threshold value above attention interval, setting the frequency threshold value.

The mode 15 of the present invention provides the processing method according to the above aspect 14, wherein the inspection condition is set before the discrimination is performed; and in the step of setting the inspection condition, the density of the defect candidate is set to the target generation density. In the method, the first signal threshold is set using a reference table; the frequency threshold is a value that varies according to the value of the target generation density in addition to the change in the interval of interest.

The mode 16 of the present invention provides a processing method of the above-described Type 14 or Type 15, wherein the frequency threshold is specified in such a manner that an image of the above-described noise component of the analog image formed by the noise component is As the defect candidate, the frequency of generation of the above-described noise component with respect to the interval is obtained, and the frequency threshold is defined based on the generated frequency.

The mode 17 of the present invention provides the processing method of the above-described mode 16, wherein the analog image is produced in such a manner that the density of the noise components in the image differs depending on the image area.

The processing method according to any one of the above aspects of the present invention, wherein the interval between the above-mentioned attention intervals and the periodic defect candidate in the moving direction is referred to as a pitch interval, After performing the above-described determination step, further defining a region of interest including a defect candidate having the pitch interval in the width direction; using the second signal threshold, from the image of the region of interest, from the image Extracting a detailed defect candidate from the beginning; defining a search area centered at a position of the detailed defect candidate obtained from the extraction in the moving direction from the position of the pitch interval; and searching for a detailed defect using the second signal threshold in the search area The candidate; the detailed defect candidate detected by the search and the attribute of the detailed defect candidate obtained by the extraction are respectively evaluated; and based on the evaluation result, whether the attention area includes the periodic defect candidate in the moving direction is determined.

According to a tenth aspect of the present invention, in the processing method of the above aspect 18, the image used for the determination of the periodic defect candidate in the region of interest is an image of the plate after the plate-like body is cut at a fixed size.

The method of claim 14, wherein the processing method of any one of the above-mentioned types 14 to 19, wherein the detecting and detecting the defect candidate having the same position in the width direction in the moving direction and determining the interval, evaluating the detected The attribute of the defect candidate, or the similarity between the defect candidate and the specific defect candidate, is obtained when at least one of the attribute and the similarity satisfies the set condition.

According to a mode 21 of the present invention, there is provided a defect inspection method which is characterized in that a defect existing in a plate-like body is inspected, and light is irradiated onto the surface of the plate-like body while the plate-like body is relatively moved. An image of the plate-like body to be irradiated is photographed while the image obtained by the photographing is used to perform the processing method according to any one of the above-described types 14 to 20.

Further, a mode 22 of the present invention provides a method for producing a plate-like body, which is characterized in that a plate-like body which is a belt-shaped continuous body which is conveyed by a conveyance roller is manufactured; and the defect inspection using the above-described type 13 The apparatus or the defect inspection method of the above-mentioned type 21, inspecting the above-mentioned plate-like body during the moving process; and determining the conveying roller which causes the plate-shaped body to be defective on the moving path of the above-mentioned plate-shaped body according to the result of the inspection; removal or maintenance The conveyor roller is determined.

Further, a mode 23 of the present invention provides a method for producing a plate-like body, which is characterized in that a plate-like body which is a belt-shaped continuous body which is conveyed by a conveyance roller is manufactured; and the defect inspection using the above-mentioned type 13 In the apparatus or the defect inspection method of the above-described type 21, the plate-like body is inspected during the movement, and the plate-shaped body is cut and taken out while avoiding the position in the width direction of the defect which is determined to have the periodic defect.

Further, the mode 24 of the present invention provides a computer executable program and a computer-readable recording medium on which the program is recorded, and the program executes the image data for defect inspection as in any of the above types 14 to 20. The treatment method.

Further, the present invention provides a processing apparatus according to any one of the above aspects 1 to 10, wherein the above-described processing is performed when an interval in which the periodicity is determined in the moving direction is referred to as a pitch interval. Further, the portion further includes a region of interest including a defect candidate having the pitch interval in the width direction, and extracts a detailed defect candidate from the image of the region of interest using the second signal threshold. a search area in which the position of the detailed defect candidate obtained from the extraction is separated from the position corresponding to the circumferential distance of the transport roller that transports the plate-shaped body in the moving direction, and the second signal threshold search is used in the search area. The detailed defect candidates are respectively evaluated for the detailed defect candidates detected by the search and the attributes of the detailed defect candidates obtained by the extraction, and based on the evaluation result, it is determined whether or not the region of interest includes the periodic detailed defect candidates in the moving direction.

Further, the present invention provides a processing method according to any one of the above aspects 14 to 17, wherein the interval between the above-mentioned attention intervals which is determined to have a periodic defect candidate in the moving direction is referred to as a pitch. At the time of the interval, after the step of determining the above, the region of interest including the defect candidate having the pitch interval in the width direction is further defined; and the second signal threshold is used, from the image of the region of interest The detailed defect candidate is extracted from the beginning of the image, and the search area in which the position of the detailed defect candidate obtained from the extraction is separated from the position corresponding to the circumferential length of the transport roller that transports the plate-shaped body is defined in the moving direction. Searching for the detailed defect candidate using the second signal threshold; respectively, evaluating the detailed defect candidate detected by the search and the attribute of the detailed defect candidate obtained by the extraction; and determining the above-mentioned region of interest based on the evaluation result Whether there is a periodic defect candidate in the direction.

The processing device for image data for defect inspection according to the first aspect of the present invention, the defect inspection device of the type 13 , the processing method of the type 14 and the defect inspection method of the type 21, and the program of the type 24 and the recording of the coal body In the middle, the occurrence frequency and the frequency threshold are used to determine the presence or absence of a periodic defect. Further, in the above, the frequency threshold is defined according to the interval between the two points of interest. When the two frequency thresholds are set to different values, the interval between the above-mentioned attentions of the frequency threshold of the larger one is smaller than the smaller one. The frequency threshold is set in such a manner that the frequency threshold is in the above-mentioned interval.

Therefore, even if the captured image contains noise components, the existence of periodic defects can be discriminated.

The other types of processing methods for the image data for defect inspection and the method for processing image data for defect inspection according to the present invention are as follows. First, in the type 2 and the pattern 15, the density of occurrence of the defect candidate is set. The first signal threshold is set in such a manner that the target density is generated, and the frequency threshold is changed in accordance with the value of the target generation density, and the existence of the periodic defect can be more effectively determined. Even if the density of occurrence of the defect candidate of the plate-like body varies depending on various conditions in the production of the plate-shaped body, the fluctuation can be randomly handled to optimally determine the existence of the periodic defect.

In Type 3 and Type 16, the randomly generated noise component can be overcome, that is, even if there is a noise component, the periodic defect can be easily detected without being affected by the noise component. Further, the frequency threshold can be easily specified based on the resolution frequency and the simulation image obtained analytically.

In the pattern 4 and the pattern 17, the periodic defects locally generated in the plate-like body can be easily detected by dividing and processing the image. Further, it is also possible to perform inspection by excluding locally generated noise components.

In the form 5, since the cell regions having the same size are formed in consideration of the positional deviation of the defect candidates generated in the image to be inspected, it is possible to more effectively determine the existence of the periodic defects in a short time.

In the pattern 6, the frequency distribution in the width direction of the defect candidate can be obtained, and the pattern of the defect generation can be classified using the unevenness of the frequency of generation along the width direction, so that it can be effectively used for the cause of the defect. In addition to speculation, it also contributes to the stable and continuous production of the condition of the plate-like body, which is very helpful for the management of production steps.

In the type 7, the plate-shaped body is divided into the plate-like body region of the set length, and the image of the region is used as a time-series unit to examine a plurality of time series units, and the defect can be determined. The information on the frequency of occurrence of the candidates is expressed as time series data. Therefore, it is possible to more reliably determine the existence of the periodic defect, and it is also possible to grasp the state of the defect which changes in time series, which contributes to the determination of the quality of the plate-like body (glass substrate) in units of sheets, and subsequent steps. Processing in the middle.

In Type 8, the frequency distribution is recorded in a plurality of time series units, so that information on how long the periodic defects continue to be generated can be obtained, and in addition, it facilitates real-time management in the production steps.

In Type 9, the time series data of a plurality of generating frequencies are overwritten in the same graph, so that the cause of the defect that may be complicatedly changed can be easily detected in a short time.

In the pattern 10, the interval between the candidate and the candidate adjacent to the previous one is obtained, and the interval between the plurality of defect candidates and the previous plurality of candidate candidates is obtained, and the frequency of occurrence is determined, thereby determining the periodic defect. The presence or absence of the periodic defect can be easily detected even in the case where there are defects caused by a plurality of different generation sources, and the positions where the generations are partially overlapped in the plane.

In the pattern 11, the pattern 18, the pattern 25, and the pattern 26, the region of interest and the search region are defined and the detailed defect candidates are searched to determine whether or not the periodic detailed defect candidate is included. Therefore, it is possible to detect the presence or absence of a predetermined assumption in a short time. The occurrence of defects. Conversely, if the time required for the detection is relatively extended, the accuracy of the defect detection can be improved.

In the type 12 and the pattern 20, when the similarity between the defect candidate attribute and the defect candidate satisfies the set condition, the interval between the defect candidates is obtained, so that the periodic defect can be reliably and completely detected, thereby improving Determine whether it is the reliability of periodic defects.

Further, in the method for producing a plate-like body of the form 22 of the present invention, the time required for the repairing operation in the step of providing the conveying roller can be shortened as compared with the prior art, and the management of the step can be continuously performed. It's easy, so you can stabilize your yield. Further, the conveying roller that is being used in the production step can be easily managed. Further, the period in which the transfer roller is to be replaced in the future can be estimated in advance.

In the method for producing a plate-like body of the type 23 of the present invention, it is possible to effectively cut and take out a strip-shaped continuous body, that is, a plate-like body, even if a conveyance roller or the like has a defect, so that good The rate is stable.

Hereinafter, a processing apparatus and a processing method for image data for defect inspection according to the present invention, a defect inspection apparatus and a defect inspection method using the same, and the like, and the like, will be described in detail based on preferred embodiments shown in the drawings. A method of manufacturing a plate-shaped body, and a recording medium readable by a computer that records a program for executing the processing method.

The defect inspection device 1 of Fig. 1A is a defect inspection device of the present invention which performs the defect inspection method of the present invention, and determines whether or not the periodicity of the defect is present. The defect inspection device 1 mainly has a defect inspection unit 10, a processing portion 16, and a defect inspection unit 26.

As the plate-like body of the present invention, a glass plate G having transparency is used in the following description. However, the plate-shaped body of the present invention is not limited thereto, and for example, a long-length acrylic plate or a long film is also included. Or paper, etc.

Further, the glass sheet G described below is cut into a continuous strip-shaped continuous body before a specific size, and the conveyance state will be mainly described. As the type of the mother glass substrate to be produced, for example, G6, G8, G10, and G12, etc., tend to have a larger size.

In the present invention, a glass plate cut to a specific size may be used as a target. However, in order to determine the periodicity of the defect candidate, it is preferable to use a strip-shaped continuous body, that is, a long glass plate, and a glass plate in a conveyed state.

Further, in the following embodiments, the glass plate G to be conveyed is photographed using a stationary light source and a camera. However, the present invention may be in the form of a glass plate G that is stationary while moving the light source and the camera. In the present invention, the light source and the camera and the glass sheet G may be relatively moved. The transport direction described in the following embodiments corresponds to the first moving direction of the present invention, and the width direction corresponds to the second moving direction of the present invention.

The defect inspection unit 10 shown in Fig. 1A is disposed on a transport path formed by a plurality of transport rollers 11 having different radii. The glass plate G is a strip-shaped continuous body which is taken out from the melt path at a fixed thickness, and is continuously conveyed and moved on the conveyance path.

Here, the driving roller is used as the conveying roller 11 for conveying the glass G, but one or more driven rollers may be used between the driving rollers. Further, as the conveying roller 11, various types of conveying rollers can be used. For example, it may be a normal conveying roller which is in contact with the glass G as a whole, such as a cleaning roller, a stationary roller, a sheathed roller, a coating roller, a coating roller, or the like. Further, it may be a conveying roller having a shoulder roller or a stepping roller and the like, which is spaced apart from the glass G in the width direction.

The defect inspection unit 10 includes a light source 22 that irradiates light to the surface of the glass sheet G, a camera 14 that images an image of the plate-shaped body that is irradiated with the light source 22, and a processing unit 16 that determines a candidate candidate having a periodicity. Further, the defect inspection unit 10 may be connected to a display 18a that displays the inspection result or the like as a soft copy image on the screen, and an output system 18 such as a printer 18b that outputs the inspection result as a hard copy image. Or enter the operating system 20, such as a mouse or a keyboard. The defect inspection unit 10 is a device that inspects a defect candidate in a transmission image by transmitting light transmitted through the glass plate G. Further, the processing unit 16 is connected to the defect inspection unit 26 that reflects the image, and the light source 22 and the camera 24 are disposed on one side of the surface of the glass sheet G, and the image of the defect of the glass sheet G is made by the camera 22. Reflect the image and extract the location of the defect candidate.

When the glass sheet G is taken out from the melting furnace at a fixed thickness and continuously conveyed by the conveying roller 11, on the surface of the glass sheet G, as shown in FIG. 1B, the conveying roller 11 is periodically generated. Small defect D. Further, a bit-shaped defect X is generated on the surface of the glass sheet G. Further, a contaminated area Y which is diffused in a certain area is also generated. Further, in the image read by the defect inspection unit 10, in addition to the above-described respective defects, the dot noise randomly generated by the processing at the time of image reading is also regarded as the defect X.

The defect inspection unit 10 periodically generates the image of the glass sheet G at substantially the same position in the width direction of the glass sheet G which is generated on the surface of the glass sheet G to be conveyed and which is orthogonal to the conveyance direction. The gap P between the defects D is distinguished from the defect X by the discriminant.

The light source 12 is a light source that emits substantially parallel light, and is a linear light source that emits substantially parallel light having a substantially uniform light intensity in a width direction of the glass sheet G (a direction perpendicular to the plane of the paper of FIG. 1A). As the light source, a halogen light source or an LED (Light Emitting Diode) light source or the like can be used. The type of light is not particularly limited, and white light is preferably used.

The camera 14 is a line sensor type camera that is placed at a position facing the glass plate G with respect to the light source 12, and directly reads the transmitted light transmitted through the glass sheet G from the light receiving surface. The type of the line sensor of the camera 14 is not particularly limited, and may be of a CCD (Charge Coupled Device) type or a CMOS (Complementary Metal-Oxide-Semiconductor) type. The camera 14 is provided with a plurality of stages in a direction perpendicular to the paper surface in FIG. 1A, and photographs the same position in the transport direction, and the plurality of cameras are set to partially overlap each other in the field of view in the width direction of the glass sheet G.

The image signal captured by the camera 14 is transmitted to the processing unit 16.

The processing unit 16 is a part constituting the processing apparatus of the present invention which performs the processing method of the image data for defect inspection of the present invention. The processing unit 16 generates an image data of the inspection target of the glass sheet G based on the transmitted image signal, and performs a defect inspection using the image data. The image signals transmitted from the camera 14 are averaged as described above to generate image data constituting one image. The above image data is used to determine whether or not the detected defect candidate has periodicity. This determination will be described below.

The light source 22 of the defect inspection unit 26 emits a light source that emits illumination light to the surface of the glass sheet G, that is, a light source that emits substantially parallel light, and causes light to enter from a direction inclined with respect to the surface of the glass sheet G. Like the light source 12 of the defect inspection unit 10, the light source 22 is a linear light source extending in a direction perpendicular to the plane of the paper of Fig. 1A, preferably in the width direction of the glass sheet G (direction perpendicular to the plane of the paper of Fig. 1A). A substantially parallel light having a substantially uniform light intensity is emitted. The present invention is provided with two light sources. For example, an LED light source or a halogen light source can be used as the light source 22. The type of light is not particularly limited and may be red, blue, white, or the like, and is preferably white.

The camera 24 is a line sensor type camera that collects reflected light emitted from the surface of the glass sheet G and captures a reflected image, and a camera of the same type as the camera 14 can be used. The camera 24 is disposed on the same side as the light source 22 as viewed from the glass sheet G. The camera 24 and the light source 22 are disposed in a positional relationship between the upstream side and the downstream side in the transport direction, and the light emission direction of the light source 22 and the field of view direction of the camera 24 are incident on the back surface of the glass plate G. The camera 24 is adjusted in the same manner.

The image captured by the camera 24 is an image that is illuminated by the light source 22 and reflected on the front and back surfaces of the glass sheet G, and is an image in which a defect existing inside the glass sheet G is an image of a dark portion. This image first contains a real image of a defect formed by the defect region passing through the optical path of the reflected light which is incident on the surface of the glass sheet G in the direction inclined with respect to the surface of the glass sheet G and reflected on the back surface of the glass sheet G.

Further, a mirror image of the defect formed by the incident light incident on the surface of the glass sheet G from the direction inclined with respect to the surface of the glass sheet G passing through the defective region in the optical path inside the glass sheet G on the back surface of the glass sheet G is included. .

In this manner, the image data obtained by the camera 24 is read linearly each time and then sent to the processing unit 16 one by one. As in the case of the defect inspection unit 10, the processing unit 16 performs defect inspection using the transmitted image data.

The reason why the defect is inspected by the defect inspection unit 26 is that the defect is located on the surface of the glass sheet G (the surface on the upper side of the glass sheet G in Fig. 1A) and the back surface based on the positional deviation of the real image of the defect and the mirror image in the conveyance direction. (While the surface on the lower side of the glass sheet G in FIG. 1A and the surface on the side of the conveying roller 11) is defined as an attribute of the defect. That is, when there is no positional deviation and one image is observed, it is understood that the defect is located on the back surface, and when there is a positional deviation amount and two images are observed, it is understood that the defect is located in the glass sheet G or the surface of the glass sheet G.

The defect inspection device 1 shown in FIG. 1A includes two defect inspection units 10 and 26. However, the present invention is not limited thereto, and only one of them may be provided.

The processing unit 16 performs processing for image data for defect inspection from the image of the obtained inspection target in the following manner. Fig. 2 is a flow chart showing the flow of a defect inspection method, particularly a method of processing image data for defect inspection.

First, the inspection condition of the defect to be detected is set (step S100). Specifically, it is preferable to input the operating system 20 (see FIG. 1A), and set an allowable amount of deviation in the width direction position when detecting a candidate candidate having a periodicity by an input from an operator, and the detection has a periodicity. The allowable amount of the deviation of the transport direction position from the specific pitch interval at the time of the defect candidate, the extent to which the periodic defect candidate is continuously generated, and the degree to which the one group defect candidate is periodically and continuously generated Information generated and information on the frequency of occurrence of periodic defect candidates.

Next, the cell size is determined based on the set inspection conditions with respect to the image to be inspected for the defect inspection (step S110).

As shown in FIG. 3, the unit size is obtained by dividing an image of a search target of a search defect candidate into a plurality of parts in the width direction and the transport direction (moving direction) to form a plurality of unit regions having the same size. The length of the unit area in the width direction and the transport direction. The cell size is determined based on the allowable amount of the positional deviation in the width direction and the transport direction set as the inspection conditions. For example, the length in the width direction × the length in the transport direction is determined to be 10 mm × 10 mm.

The reason why the cell regions are formed in this manner is that the interval between the following defect candidates located in the cell region and the defect candidates located in the other regions is determined in accordance with the interval (separation distance) of the cell regions. . Therefore, as long as the defect candidate is located in one unit area, the position of the defect candidate can be regarded as being processed regardless of the position. Even if the position of the periodic defect candidate varies within the allowable range in the width direction and the transport direction, the positional deviation of the defect candidate in the cell region can be absorbed or reduced or eliminated by setting the cell size. The interval between stable candidate candidates.

Next, it is decided to check the unit length (step S120). The inspection unit length refers to the length of the image in the transport direction of the defect inspection target. By repeating the defect inspection by setting the length of the inspection unit to be fixed, the result of the defect inspection in the time series can be obtained, and the cause of the defect or the like can be estimated.

The inspection unit length is determined based on the length of the defect set as the inspection condition, and whether or not the information is generated periodically and continuously. For example, the length of the glass sheet G conveyed for one hour or one day, or the length of 100 m or 1000 m or the like is determined.

Then, the target of the defect candidate is determined to have a value of density (step S130). The defect candidate is a dark portion that is divided into a dark portion in the image when the captured image of the photographed glass sheet G is binarized. In the present embodiment, the periodicity of the fine scratches generated by the conveyance roller 11 of the glass sheet G is checked. Therefore, when the density of the defect candidates is high, a plurality of spots due to noise components are formed. Dark area.

Therefore, it is difficult to accurately discriminate whether or not the fine flaws to be inspected are periodic. On the other hand, when the first signal threshold value described below is lowered and the generation density of the defect candidate is reduced, there is a case where the fine flaw to be inspected cannot be divided as the dark portion of the defect candidate.

Therefore, the target generation density of the dark portion region in the image is determined using the information of the frequency of occurrence of the periodic defect candidate set as the inspection condition.

Next, the first signal threshold is determined based on the determined target generation frequency (step S140). The first signal threshold is a threshold value of image data when the captured image of the glass plate G is binarized. When the value of the image data is lower than the first signal threshold, it is divided into a dark region (defect candidate). The processing unit 16 includes a reference table indicating the relationship between the first signal threshold and the density of the dark region, and obtains the density based on the determined target, and obtains and determines the first signal threshold by referring to the reference table.

Generally, the smaller the target generation density is, the lower the first signal threshold is set. In the defect inspection unit 10, the relationship between the first signal threshold value and the density of occurrence of the dark portion region is obtained in advance for the captured image of the specific glass plate G, and is stored in the memory in advance.

Next, the frequency threshold is determined based on the value of the target occurrence density of the determined defect candidate (step S150). The frequency threshold refers to a threshold for discriminating whether or not the glass sheet G has a periodic defect in the defect inspection (step S160) described below. The frequency threshold is such that the horizontal axis represents the interval of the defect candidates and the vertical axis represents the threshold for discriminating the frequency having the periodicity in the histogram obtained with respect to the frequency of the interval.

For example, when the processing unit 16 creates a histogram as shown in FIG. 4, it is determined whether or not there is periodicity based on whether or not the frequency of occurrence of the interval B is larger than the frequency threshold A determined for the interval B. When the frequency of occurrence of the interval B is high with respect to the frequency threshold A, it is determined that there is periodicity, and the interval B is set as the pitch interval.

Whether or not the periodicity is determined by each interval of the horizontal axis of the histogram is determined, and the frequency threshold is changed according to the interval of the attention, and the smaller the interval, the larger the frequency threshold is set. Further, it is preferable that the frequency threshold is changed in accordance with the value of the target generation density of the determined defect candidate, in addition to the change in accordance with the interval. Specifically, it is preferable to set the frequency threshold to be smaller as the target generation density becomes smaller.

The reason why the frequency threshold is set in such a manner is that since the defect candidate includes the defect candidate due to the noise component as described above, the periodicity is set so as to prevent the defect candidate due to the noise component. Misjudgment.

5A and 5B are graphs showing the relationship between the intervals of defect candidates in the simulation image created only based on the noise component and the frequency of occurrence of the interval. It is assumed that the noise components are randomly generated, and a noise component is generated in the region of the glass plate G and having a length of 2,500 mm and a cell size of 10 mm × 10 mm.

The vertical axes of Figs. 5A and 5B each indicate the number of generations per 1 m ² (production frequency). The generation density of the noise component in Fig. 5A is divided into three variations (50/m ² , 100/m ² , 200/m ² ). According to FIG. 5A, for any of the generation densities, the smaller the interval, the higher the frequency of generation, and the higher the generation density of the noise component, the higher the frequency of occurrence. Therefore, in order to prevent erroneous discrimination of the periodicity of the defect candidate due to the noise component, it is preferable to set the frequency threshold to be higher with respect to the generation frequency (with margin) of the vertical axis shown in FIG. 5A. For example, it is sufficient to set the value of 1.1 to 2 times the frequency of occurrence of each interval as the frequency threshold.

Of course, the above values can be set larger depending on the inspection conditions. Therefore, the larger the interval and the smaller the generation density of the noise component, the lower the frequency threshold is set to be substantially lower. Here, the fact that the setting is substantially lower means that the frequency threshold does not change (equal) even if the interval of attention is different.

For example, it refers to the case where the frequency threshold of the interval of 200 mm is larger than the frequency threshold of 500 mm and 1000 mm, but the frequency threshold of 500 mm is equal to the frequency threshold of 1000 mm. In the present invention, when the two frequency thresholds are different, the frequency threshold is set such that the interval at which the frequency threshold of the larger one is specified is smaller than the interval of the frequency threshold of the smaller one.

Fig. 5B is a graph showing the generation frequencies of the setting intervals of 1000 mm, 750 mm, and 500 mm with respect to the density of the noise component. According to FIG. 5B, the smaller the generation density of the noise component becomes, the smaller the generation frequency of each of the set intervals of 1000 mm, 750 mm, and 500 mm becomes.

Further, as is known from Fig. 5B, if the density of the noise component reaches a certain level or more, the frequency of occurrence is lowered. That is, the generated frequency has a peak with respect to the density of generation of the noise component. The reason for this is that the noise component becomes larger between the set intervals of 1000 mm, 750 mm, and 500 mm due to the increase in the density of the noise component, and as a result, the frequency of occurrence of the set intervals of 1000 mm, 750 mm, and 500 mm is lowered.

Therefore, the target generation density of the defect candidate is defined in such a manner that it is determined that the interval at which the periodicity is noted is smaller than the value of the density at which the peak position is generated.

Further, in the examples shown in FIGS. 5A and 5B, the frequency threshold is determined using the interval of the defect candidates in the analog image formed based on the noise component and the frequency of the interval generation, but the present invention does not It is limited to the case where an analog image formed based on a noise component is used.

For example, it is also possible to assume that the noise components are randomly distributed in the region, and the noise components in the width direction at the same position in the region are used as defect candidates, and the frequency of generation of the noise components is analytically determined, and The frequency is determined by the frequency of the determination.

The fact that the generation frequency is obtained analytically means that the generation frequency is calculated using the equation. For example, in the one-dimensional region, the probability pp for finding the n-group interval is expressed by the following equation. At this time, the probability pp is calculated by changing n from 1 to N/P (P is a length between the unit units), and the expected value is added, whereby the frequency of occurrence with respect to the pitch P can be obtained.

Here, p is the probability that a noise component is generated in one unit, the total number of N-type units, and P is long in the unit indicated by the unit. _{(Nn, P) The} C of C _n represents a combination of the mergers.

Further, as another aspect, it is possible to use the defect inspection unit 10 in the glass plate G in which the periodic defect is not caused by the conveyance roller, and to determine the frequency of occurrence of the noise component by actual measurement, and use the The frequency threshold is determined by the measured result. The frequency of occurrence of defect candidates caused by the noise component is actually uneven in each region of the image, and there is also a case where the noise component is not generated by the same density in the entire region.

Therefore, the frequency threshold after the above unevenness can be determined in consideration of the actual measurement using the glass plate G. In this respect, it is preferable to obtain the relationship between the interval between the actual measurement results and the frequency of occurrence of the defect candidates in advance, and use the relationship to determine the frequency threshold. Furthermore, even in this case, the frequency threshold is set so as to be substantially larger as the interval is smaller.

FIG. 6A is a graph showing the frequency of occurrence of the interval at the time of defect inspection (measured) with respect to the defect inspection unit 10. The symbol ■ in the graph indicates the result of actual measurement using the defect inspection unit 10. On the other hand, the result obtained by synchronizing the generation density (average production density) of the defect candidates to generate a noise image of the noise component as described above is represented by a symbol ◇.

As can be seen from Fig. 6A, when the interval between the defect candidates is 500 mm or less, there is a deviation between the frequency of the actual measurement and the frequency of the simulation. The reason for this is considered to be that even if the average density of the noise components in the simulated image is the same as the average density of the actual measurement, the density of the candidate candidates in the image to be inspected which is actually measured varies depending on the region. Actually, the probability density function has a distribution as shown in FIG. 6B when the image of the inspection object obtained by the actual measurement is divided into small areas and the number of defect candidates is counted, thereby checking the probability density function for generating the defect candidate.

Therefore, in order to conform to the measured probability density function, by making the same probability density function in the simulated image, as shown in FIG. 6C, the symbol ◇ of the simulated image is used to indicate the frequency of generation of the symbol close to the actual measurement. Therefore, it is also possible to determine the relationship between the interval and the generation frequency based on the probability density function used for the generation of the noise component, the analog image obtained by the distribution in accordance with the measured method, and use the relationship to determine the frequency threshold.

As described above, the relationship between the interval and the frequency of occurrence of the defect candidate can be determined in advance by using the glass plate G having no periodic flaws, and the frequency threshold can be determined using the relationship.

Next, a defect check is performed (step S160). In the defect inspection, the image of the inspection target sent from the camera 14 is first cut out to the area of the inspection unit length, and the image of the area of the inspection unit length is imaged by the unit size as shown in FIG. Regionalization.

The zoning of the image is performed based on the cell size determined by the image of the search target of the search defect candidate, and a plurality of cell regions having the same cell size are formed. When the plurality of unit regions including the plurality of defect candidates are located at the same position in the width direction, the defect candidates are positioned at the same position in the width direction, and the interval between the defect candidates and the generation frequency are obtained.

Next, using the determined signal threshold (first signal threshold), the image of the region of the unit length is binarized, and a plurality of dark regions are extracted as defect candidates, and the end of the transfer direction from the image is extracted. The part repeats the search along the transport direction and detects the defect candidate. The detection of the defect candidate is performed in the unit of the unit size. If there is a defect candidate in the division of the unit size, the position of the representative point of the division (the center point of the division or the vertex of the rectangular division) and the conveyance direction are used. The position is stored in a memory (not shown) of the processing unit 16.

Further, it is searched for whether or not there is a defect candidate along the transport direction. When the defect candidate is detected, the position in the transport direction of the defect candidate having the same position in the width direction stored in the memory is called, and the difference between the position of the transport direction of the found defect candidate and the position of the transferred transport direction is obtained. And set the difference to interval.

Further, the count value which is determined in each of the positions in the width direction, which is set in the memory (not shown) of the processing unit 16, and which is determined by the frequency of each interval, is advanced by one. The position in the transport direction and the position in the width direction are expressed by the value of the representative point of the division divided into the unit sizes. In this manner, the search and detection of the defect candidates are performed until the entire image of the inspection target is searched. Finally, using the count values memorized in the memory, the positions in the respective width directions and the frequency of occurrence of the defect candidates for each interval are obtained.

Next, the processing unit 16 creates a histogram of the defect candidates whose horizontal axis indicates the interval and the vertical axis indicates the generation frequency as shown in FIG. 4 based on the obtained generation frequency (step S170). Specifically, the frequency of generation of the frequency of the memory and the position of the width direction stored in the memory are accumulated at each interval, and a histogram indicating the frequency of generation of each interval is created. By summing the frequency of occurrence of each of the intervals noted in the histogram, it is checked whether the frequency of occurrence of the interval of interest is higher than the determined frequency threshold (frequency threshold A in FIG. 4). When the generation frequency is higher than the frequency threshold, the interval corresponding to the generation frequency is determined as the interval formed by the defects having the periodicity, and it is determined that the glass sheet G has the periodic defect (step S180).

Since the glass sheet G is formed in a continuous shape in the transport direction, the image of the unit length to be determined is subjected to the above-described defect inspection as an image for a plurality of units, and is time-series The production frequency is produced for each inspection unit length. Of course, the images of the plurality of units produced are respectively used as the object of defect inspection.

In this way, the defect inspection of the long glass sheet G conveyed on the conveyance path is performed.

As a result of the above defect inspection, a defect is generated at the interval between the judgments. Therefore, it is possible to estimate the diameter of the roller in the conveyance roller 11 by using the information of the pitch interval.

In the present embodiment, after step S180, the cause of the periodic defect can be inferred in accordance with the flow shown in FIG.

First, the frequency distribution which is determined in step S180 and the predetermined distance between the width directions is calculated, and the feature quantity of the distribution is calculated (step S181). The characteristic quantity of the distribution includes, for example, a position in the width direction of the maximum generation frequency and a standard deviation of the generation frequency in the frequency distribution in the width direction.

Since the frequency of each of the intervals and the positions in the width direction is stored in the memory of the processing unit 16, the interval can be fixed to the predetermined interval, and the frequency distribution of the position in the width direction can be obtained. Three examples of the distribution of the specific interval in the width direction are shown in Figs. 8A, 8B, and 8C. The distribution of the width direction described above is preferably shown on the screen of the display 18a (see Fig. 1A). The example of Fig. 8A produces a pattern in which the frequency is projected at a position in the width direction. The example of Fig. 8B is a generation pattern in which a frequency is generated in a fixed range in the width direction and a distribution is formed in the range. The example of Fig. 8C is a generation pattern in which a frequency is uneven in a wide range in the width direction.

In this way, the frequency distribution of the position in the width direction is classified into a plurality of generation patterns in addition to the standard deviation (unevenness) of the distribution, and the feature amount such as the position in the width direction of the maximum generation frequency is used.

Next, the defect candidate which is determined to have a periodic defect (hereinafter referred to as a candidate candidate having a periodicity) is subjected to the distribution of the time series of the generation frequency and the generation density, and the feature quantity of the distribution is calculated (step S182). As described above, the determined inspection unit length is used as the inspection target of one time series unit, and the defect inspection is performed on a plurality of time series units in order, so that the generation density of the periodic defect candidates can be created (/m ² And the time series distribution of the generated frequencies.

Fig. 9 is a diagram showing the time in which the generation density (generation frequency) of the specific interval interval, the generation density of all defect candidates, and the frequency of occurrence of the width direction of the specific interval interval (standard deviation) are overwritten in one graph. An example of a sequence distribution.

Further, as the feature quantity of the distribution, a value set for each time series distribution is set, and the duration in which the density is higher than the value is determined. Or find the correlation coefficient of each time series distribution. Alternatively, as the feature quantity of the distribution, the standard deviation of each time series distribution is obtained. Furthermore, the range of the value of the vertical axis of the time series distribution of Fig. 9 differs from the respective examples. The above time series distribution is preferably displayed on the screen of the display 18a.

Further, the time-series distribution (time-series data) of the frequency of occurrence of the defect candidate may be a time-series distribution (time series data) of a plurality of generation frequencies in which at least one of the positions between the intervals of interest and the width direction is changed. Overwritten in the same chart and displayed on the screen of display 18a.

Further, a method of expressing the frequency of occurrence of the defect candidate or the like, a method of expressing the time-series distribution such as the frequency of occurrence of the defect candidate, and a time-series representation method of the frequency of occurrence of the defect candidate, etc., will be described below.

Then, with respect to the defect candidates having periodicity which are determined in step S182 among the plurality of pitch intervals which are determined to have periodicity, the frequency distribution of the width direction of the defect candidates of the other pitch intervals is obtained, and the frequency is further created. The time series distribution corresponding to the defect candidate of the pitch interval and corresponding to FIG. 9 is displayed on the screen of the display 18a. At this time, the correlation between the frequency distribution of the width direction and the time series distribution of the obtained width direction, the frequency distribution of the width direction obtained in steps S181 and 182, and the time series distribution are evaluated (step S183). ).

Specifically, the correlation coefficient is obtained for each of the frequency distribution and the time series distribution in the width direction. Further, the feature quantity of the frequency distribution of the width direction of the defect candidate in the other pitch interval and the feature quantity of the time series distribution are calculated, and compared with the feature quantity calculated in steps 181 and 182.

Further, a plurality of feature quantities α (image size, shape, value of image data, etc. of the defect candidate) that are determined to have an image of a plurality of periodic defect candidates are calculated, and an average value including the feature amount α is calculated. The feature amount β such as the value and the standard deviation (step S184).

The feature amount α and the feature amount β calculated in this manner are compared with the feature quantities calculated in steps S181 and 182 together with the feature amounts stored in the database as past data in advance, and the comparison result is allowed When the range is the same, the cause of the defect registered in association with the feature amount is called, and it is estimated that the cause of the periodic defect is generated (step S185). Furthermore, it is judged that the defect candidate system which is evaluated as the other pitch interval having high correlation by the step S183 is caused by the same reason as the inferred cause.

For example, when a plurality of foreign matters of the same conveying roller and the same material are attached to different positions on the circumference of the same conveying roller in the same period of time, a defect occurs, and two defects appear between the defect candidates. Spacing interval. However, the feature quantities of the frequency distribution and the time series distribution in the width direction of the defect candidate are calculated and compared, thereby having the same generated frequency distribution and having the same time series distribution.

Therefore, it can be inferred that the defect candidates evaluated as the other pitch intervals having higher correlation by the step S183 are generated by the same transfer roller as the defect candidates set in the step 181.

Furthermore, the comparison of the above feature quantities (consistent, inconsistent) can be compared by setting the conditional differences of each feature quantity, and the Mahalanobis space and the Mahalanobis distance with respect to the feature quantity can be used. For comparison, it is also possible to compare by constructing a neural network.

Assuming that two pitch intervals are formed between the defect candidates, in the present invention, in addition to finding the interval between adjacent defect candidates (previous defect candidates), the interval between the candidate candidates can be obtained. The interval between the defect candidates adjacent to the defect candidates (the two preceding defect candidates), and the interval between the defect candidates (three preceding defect candidates) adjacent to the two preceding defect candidates, ...., the interval between the defect candidates (N former defect candidates) adjacent to the defect candidate of (N-1) (N is an integer of 4 or more) is obtained. For example, in the defect inspection in step 160, the adjacent defect candidates are set as the target to obtain the interval, and in step S181, when two or more intervals are generated, the interval memorized in the memory may be used. In addition to the interval between the adjacent defect candidates (previous defect candidates), the information is obtained by determining the interval between the defect candidates adjacent to the plurality of previous defect candidates.

In this case, the upper limit of the interval to be determined is the maximum circumference of the conveyance roller 11. In this case, the frequency of occurrence of the interval associated with all combinations of one to N preceding defect candidates may be determined to be a histogram, and the frequency threshold to be set may be used to determine whether or not the frequency is set. With this configuration, when the circumference of the conveying roller 11 is 1000 mm, when the frequency of generation of the candidate candidates is 300 mm and 700 mm exceeds the frequency threshold, 300 mm is added to the frequency of 1000 mm. The sum of the generated frequencies of 700 mm, so the frequency of occurrence exceeding the set frequency threshold occurs. Thereby, it can be more accurately estimated that the defect candidate system is generated by the same conveying roller.

Fig. 10 is a view for explaining a flow of an example of the execution of the defect inspection shown in Fig. 2, and can also be carried out as follows.

In step S160 shown in FIG. 2, as described above, in the defect inspection, the image of the inspection object transmitted from the camera 14 and produced is divided into the area of the inspection unit length, and the image of the area of the unit length is as shown in FIG. The image is zoned by the cell size (step S161). Next, the image of the area of the unit length is binarized using the determined first signal threshold, and the area on the dark side is used as a defect candidate, and the defect is searched for from the end of the transport direction in the image along the transport direction. The candidate is extracted, and the position in the width direction of the detected defect candidate is extracted (step S162).

At this time, if the defect candidate is detected, the attribute of the defect candidate is determined (step S163). As the attribute, for example, whether or not the value of the image signal of the defect candidate is less than a specific value, the area of the image of the defect candidate, or the shape of the defect candidate satisfies the set condition. Moreover, whether or not the defect candidate is located on the back surface of the glass plate G (the surface on the side of the conveyance roller 11) is mentioned. The attribute of whether the defect candidate is located on the back side or on the surface can be discriminated using the reflection image obtained by the defect inspection unit 26.

That is, the reflected image is irradiated onto the back surface of the glass sheet G and is imaged by the camera 24, so that when there is a defect on the back surface as described above, there is no positional deviation between the real image and the mirror image in the reflected image. On the other hand, there are real images and mirror images of defects existing on the surface, and the amount of positional deviation coincides with a fixed value depending on the thickness of the glass sheet G. In this regard, it is possible to determine whether or not the defect candidate of the reflected image located at the position corresponding to the defect candidate is located on the back side. The attribute is discriminated using the result of the discrimination.

Further, as the attribute, the defect type can also be listed. The type of defect can be identified by the shape of the image of the corresponding defect candidate obtained from the reflected image.

Then, the position in the transport direction and the position in the width direction of the region to which the defect candidate corresponding to the desired attribute belongs are proposed (step S164), and the position in the width direction of the representative point of the region, the position in the transport direction, and the candidate candidate map. The information of the image area is stored in a memory (not shown) of the processing unit 16. Further, it is searched for whether or not there is a defect candidate of a desired attribute along the transport direction.

At this time, correlation between the size and the correlation coefficient of the shape of the image region is obtained between the image region of the defect candidate detected as the desired attribute and the image region of the defect candidate that has been memorized in the memory. The sex is evaluated as the similarity (step S165). For example, when the correlation coefficient is used for evaluation, when the value of the correlation coefficient exceeds a specific value, it is judged that the degree of similarity is high. The defect candidates detected as the evaluation target of the similarity and the candidate candidates stored in the memory are, for example, defect candidates generated at specific intervals, or defect candidates detected and detected and stored before the detection. The defect candidate is a defect candidate that is averaged as a parent group, or a defect candidate detected, and a candidate candidate model determined in advance. The related objects include the feature amount of the defect candidate, the image data of the defect candidate, or the image feature amount (the feature amount of the shape, etc.) of the defect candidate.

Furthermore, the evaluation of similarity can be evaluated using the Mahalanobis space and the Mahalanobis distance of the feature quantity in addition to the correlation, or by constructing a neural network. .

The difference between the position in the transport direction of the defect candidate at the position in the same width direction that is stored in the memory and in the memory is determined with respect to the position of the transport direction of the defect candidate that is determined to have a high degree of similarity. . The count value indicating the frequency of occurrence defined by each position and interval in the width direction is advanced by one with respect to the interval as the difference (step S166).

It is determined whether or not the search and detection of the defect candidate have been performed for the entire image of the inspection target (step S167). If the determination is negative, the process returns to step S162. If the above determination is affirmative, the process proceeds to step S170.

As described above, when the defect candidate is detected, the condition of the desired defect candidate and the similarity of the image area of the defect candidate are strictly set, and the defect candidate to be detected can be restricted. Of course, either one of the attribute of the defect candidate and the similarity of the image area of the defect candidate may be set as a condition.

Further, as a step subsequent to step S180 shown in FIG. 2, the cause of the defect may be removed in accordance with the flow shown in FIG.

If it is determined in step S180 that there is a candidate candidate having a periodicity, the frequency distribution and the time-series distribution in the width direction of the interval between the candidate candidates are obtained as described above, and the generated frequency distribution is obtained based on the obtained The feature quantity of the time series distribution is extracted to extract features of the pattern, thereby evaluating candidate candidates having periodicity. Alternatively, the defect type of the defect candidate is identified (step S191). The type of the defect refers to the shape of the image of the defect candidate in the reflection image captured by the defect inspection unit 26, and the type of the defect determined by using the value of the image data indicating the degree of the light intensity, for example, the glass plate G. A flaw generated on the surface or an attachment on the surface of the glass sheet G.

Next, based on the evaluation result or the recognition result, the cause of the occurrence, that is, which of the conveyance rollers is generated, is generated (step S192). For example, a conveying roller having a circumference conforming to the interval interval is inferred as a cause. Further, it is estimated which of the conveyance rollers causes a defect depending on the type of the defect. These inferences can be made by pre-constructing a database that correlates the cause of the defect with the type of defect or pattern.

Next, the cause of the occurrence is removed based on the estimation of the cause (step S193). For example, when the cause of the transfer is a specific transfer roller, the transport path is controlled such that the transfer roller is automatically removed from the transport path by the transfer path and the transfer roller is automatically removed from the transport path. And the drive device operates. Or, perform automatic maintenance on a specific conveying roller. As an automatic maintenance, an example of a protective film which thickens the surface of a conveyance roller is mentioned.

The cause of the result thus obtained is additionally registered in the above-described database in association with the pattern or the type of the defect (step S194). Of course, when the reason for the inference is incorrect, the correction is performed by the input of the operator and then registered in the database. This database is used for the inference of the cause in step S182.

Alternatively, in the cutting step of the glass sheet G after the conveyance, the cutting machine may receive the command so as to avoid the position in the width direction of the defect having the periodic defect, and cut the glass sheet G to a specific size. Cut and cut out.

Further, in the present invention, it is possible to use the data of the frequency distribution in the width direction of the defect inspection and the width direction of the defect candidate, and to easily detect the existence of the periodic defect by the method described below.

In other words, the processing unit 16 determines the region of interest including the position where the defect candidate having the pitch interval is located in the width direction. In the region of interest, detailed defect candidates are extracted from the beginning of the image using the second signal threshold from the image captured by the camera 14.

The second signal threshold is used to search for detailed defect candidates using the second signal threshold as a central search area with the position of the detailed defect candidate obtained by the extraction in the transport direction as the center, and the detailed defect candidates and extractions obtained by the search are evaluated. The similarity of the images between the detailed defect candidates obtained. Based on the evaluation result of the similarity, it is determined that there is a periodic defect in the transport direction in the region of interest. The second signal threshold may also be set to a value lower than the first signal threshold determined in step S150 of FIG.

Fig. 12 shows an example of the flow of this inspection. First, as shown in FIG. 13A, a region of interest AX which is obtained by the defect inspection of the flow shown in FIG. 2 and which is determined to have a position in the width direction of the periodicity of the defect candidate is defined, and is used in the region of interest AX. The second signal threshold detects a detailed defect candidate from the beginning of the image (step S195). In FIG. 13B, the defect candidate D ₁ is detected as the detected detailed defect candidate.

Secondly, based on the detected the defect candidate D ₁ of the transport position and direction of, as shown in FIG. 13B, as the center is set a fixed range of the search area AY (step from the position away from the location to the pitch spacer portion is to the conveyance direction S196).

In the search area AY that is set, the detailed defect candidate is extracted using the second signal threshold (step S197).

Next, the detailed defect candidates detected in step S195 and the attributes of the detailed defect candidates are evaluated (step S198). For example, the position (surface side or back surface) where the defect is generated is determined as an attribute. As described above, the discrimination is determined based on the difference between the positional deviation of the real image and the mirror image of the defect candidate located on the surface and the defect candidate located on the back surface, based on the image data supplied to the defect inspection unit 26 of the processing unit 16.

Next, when the above attributes are identical, it is determined that the periodic defect exists in the region of the glass sheet G of the inspection object (step S199).

In this manner, the region of interest AX is determined using the previously obtained interval interval, and an accurate defect candidate is searched for in the region, and a check for determining whether or not there is a periodic defect is performed.

Further, the inspection method can be applied to a sheet glass G of a specific size, in addition to being applicable to the long glass sheet G in the conveyance. In particular, in the case of the sheet-shaped glass sheet G, the presence or absence of periodic defects can be individually determined using the pitch interval.

In this type, the region of interest is set for detecting the interval between the defects by the defect inspection, but the present invention is not limited thereto. In the present invention, the case where the circumference of the conveyance roller can be limited and the case where the circumference of the specific conveyance roller immediately after the maintenance is concerned can be considered. Therefore, the region of interest can be set based on the circumference of the conveyance roller. Further, in the present invention, when there is a shoulder roller or a step roller or the like, there is a case where the relationship between the position in the width direction and the conveying roller can be determined, and the region of interest can be set based on this.

In the present embodiment, as shown in FIG. 9, a graph in which a plurality of time series distributions such as a time series distribution of frequency of occurrence of defect candidates are overwritten is displayed on the screen of the display 18a. However, the present invention is not limited to the present invention. this.

In the present invention, as shown in Fig. 14A, the time-series distribution (time-series data) of the frequency of occurrence of the defect candidate may be displayed on the screen of the display 18a as a two-dimensional density image indicating the frequency of occurrence of the density. The two-dimensional density image is represented by one axis, for example, the vertical axis represents time, the other axis, for example, the horizontal axis represents the position in the width direction, and the color or density (shading) indicates the pitch of the eye corresponding to each inspection condition (for example, equivalent) A two-dimensional density image of the frequency of occurrence of defect candidates for the circumference of the transfer roller concerned. Further, instead of the two-dimensional density image, as shown in FIG. 14B, the height in the direction orthogonal to the two-dimensional coordinate in which the position in the width direction is the axis on the other side and the axis in the width direction may be orthogonal to each other. The three-dimensional graph shown is displayed on the screen of the display 18a.

The frequency of occurrence of the defect candidate in the time and width direction indicated by the black dot of the two-dimensional density image shown in FIG. 14A is high, and it is understood that the black dot corresponds to the peak of the frequency of the three-dimensional graph shown in FIG. 14B.

Further, as shown in FIG. 15A, the other axis (horizontal axis) may be changed to a pitch, and as in FIG. 14A, the time-series distribution of the frequency of occurrence of the defect candidate is taken as a two-dimensional image indicating the frequency of generation by density. Displayed on the screen of the display 18a. In the two-dimensional density image, one axis (vertical axis) is set to time, the other axis (horizontal axis) is set as a pitch, and the color or density (light and dark) indicates the width direction of each of the inspection conditions. A two-dimensional density image of the frequency at which a position (e.g., a position where a flaw of unknown cause is likely to occur) is generated. Further, instead of the two-dimensional density image, as shown in FIG. 15B, the height in the direction orthogonal to the two-dimensional coordinate in which the time is set to one axis and the pitch is the other axis may be used. The represented three-dimensional graph is displayed on the screen of the display 18a.

As described above, in the present embodiment, the parameters (variables) of the frequency of occurrence of the defect candidate include three items of time, width direction, and pitch. Therefore, as shown in FIGS. 14A to 15B, instead of the two-dimensional density image or the three-dimensional graph which is the frequency of occurrence of the inter- and inter-width position, the time-series distribution of the frequency of occurrence of the defect candidate can be expressed as shown in FIG. 16A. 16C and FIGS. 17A to 17C show a three-dimensional graph or a two-dimensional density image which sequentially expresses the frequency of generation of a specific unit time.

16A to 16C are three-dimensional graphs in which one axis is a width direction position and the other axis is a pitch, and the frequency of generation of defect candidates per unit time corresponding to each inspection condition is indicated by height, and is a time series, respectively. Sexually indicate the frequency of occurrence data per day.

Of course, instead of the three-dimensional graphs of FIGS. 16A to 16C, as shown in FIGS. 17A to 17C, one axis (vertical axis) is set as a pitch, and the other axis (horizontal axis) is set to a width direction position. The color or density (shading) indicates that the two-dimensional density image of the frequency per unit time corresponding to each inspection condition is temporally changed and displayed.

Here, FIGS. 16A to 16C and FIGS. 17A to 17C are time-seriesly representing the generation frequency data per day, but may be continuously switched as shown in the display 18a as a dynamic picture. The time interval for generating the frequency data is not particularly limited, and may be longer than one day or less than one day, or may be continuous as a so-called dynamic picture.

As described above, in the present embodiment, the distance between the defect candidates and the position in the width direction of the defect candidate having the pitch interval can be defined by the defect inspection shown in FIG. 2, thereby adding FIGS. 7, 10, and 11 and The process shown in 12 effectively detects defects with periodicity. Further, the cause of the defect can be inferred.

The above defect inspection method can be preferably used for a method of manufacturing a glass sheet G or the like. In other words, the defect inspection method is used to perform the defect inspection during the conveyance of the sheet-like body such as the glass sheet G, and the cause of the sheet-shaped body transport path is estimated based on the result of the inspection. The result of the inference is preferably displayed on the screen of the display 18a (see Fig. 1A).

Alternatively, it is also possible to obtain a countermeasure against the cause of the occurrence of the transport path based on the result of the estimation. For example, when a defect candidate having a periodicity is formed due to foreign matter adhering to the conveying roller, the conveying roller is configured to be detached from the conveying path. Alternatively, in this way, the conveyance roller that causes defects such as scratches on the glass is maintained, or it is replaced with another conveyance roller. Further, in the step of cutting out the glass sheet G after the conveyance, the width direction position of the defect which is determined to have a periodic defect is avoided, and the glass sheet G is cut by a specific size and cut out. deal with.

In this type, the defect inspection of the present invention is used to perform feedback for eliminating defects of the glass by detachment, maintenance, replacement, and the like of the transfer roller. However, the present invention is not limited thereto, and evaluation of the glass manufacturing environment may be utilized. Feedback. Specifically, in the manufacture of glass, there is also a case where the contamination in the gas does not periodically adhere to the back surface of the glass, and the liquid is attached to the lower surface of the glass such as tin (slag) or The semi-liquid object is transferred onto the conveying roller 11 (see FIG. 1A) and re-transferred onto the glass G.

At this time, the contaminated area Y shown in FIG. 1B is periodically generated on the lower surface of the glass G. Therefore, the glass manufacturing environment of the floatation or the like can be evaluated based on the periodic defect points (contamination) detected on the surface of the glass G.

As described above, by applying the defect inspection method and apparatus of the present invention to the inspection of defects caused by a manufacturing environment such as dross, the manufacturing environment such as glass can be evaluated, and the evaluation result can be fed back to the manufacture of glass.

The processing method of the image data for defect inspection described above can be processed on a computer by executing a program.

For example, the processing program for image data for defect inspection according to the present invention has the steps of processing the image data for defect inspection by a computer, specifically, a CPU (central processing unit) The order of the person. A program containing the sequences may also be constructed as one or a plurality of program modules.

The processing program for image data for defect inspection including the order of execution by the computer can be memorized in the memory (memory device) of the computer or the server, or can be memorized in the recording medium, and when executed, the computer ( The CPU or other computer reads from the memory or the recording medium and executes it. Therefore, the present invention can also be a computer-readable memory or recording medium for storing a processing program for image data for defect inspection which is useful for causing a computer to execute the image data for defect inspection of the above-described type 14.

In the above, the processing device and the processing method for the image data for defect inspection according to the present invention, the defect inspection device and the defect inspection method using the same, the method for manufacturing the plate-shaped body using the same, and the program for recording the execution processing method The present invention is not limited to the above-described embodiments and examples, and various modifications and changes can be made without departing from the spirit and scope of the invention.

The present invention has been described in detail with reference to the specific embodiments thereof. It is understood that various changes and modifications may be added without departing from the spirit and scope of the invention. The present application is based on Japanese Patent Application No. 2008-187450, filed on Jan.

1. . . Defect inspection device

10, 26. . . Defect inspection unit

11. . . Transfer roller

12, 22. . . light source

14, 24. . . camera

16. . . Processing department

18. . . Output system

18a. . . monitor

18b. . . Printer

20. . . Input operating system

Fig. 1A is a view showing a schematic configuration of a defect inspection device according to an embodiment of the defect inspection device of the present invention;

1B is a view for explaining a glass plate to be inspected of a defect inspection device according to an embodiment of the defect inspection device of the present invention;

Figure 2 is a flow chart showing an example of a flow of an embodiment of the defect inspection method of the present invention;

Figure 3 is a view for explaining a part of the processing of the defect inspection method of the present invention;

Figure 4 is a view for explaining a part of the processing of the defect inspection method of the present invention;

5A is a view for explaining an example of a frequency threshold used in the defect inspection method of the present invention;

5B is a view for explaining an example of a frequency threshold used in the defect inspection method of the present invention;

6A is a view showing another example of a frequency threshold used in the defect inspection method of the present invention;

6B is a view showing another example of the frequency threshold used in the defect inspection method of the present invention;

6C is a view showing another example of the frequency threshold used in the defect inspection method of the present invention;

Figure 7 is a flow chart showing an example of the flow of another embodiment of the defect inspection method of the present invention;

Figure 8A is a view showing an example of a frequency distribution of the width direction obtained by the defect inspection method of the present invention;

Figure 8B is a view showing an example of a frequency distribution of the width direction obtained by the defect inspection method of the present invention;

Figure 8C is a view showing an example of the frequency distribution of the width direction obtained by the defect inspection method of the present invention;

Figure 9 is a view showing an example of a time series distribution obtained by the defect inspection method of the present invention;

Figure 10 is a flow chart showing an example of the flow of another embodiment of the defect inspection method of the present invention;

Figure 11 is a flow chart showing an example of the flow of another embodiment of the defect inspection method of the present invention;

Figure 12 is a flow chart showing an example of the flow of another embodiment of the defect inspection method of the present invention;

FIG. 13A is a view for explaining a defect inspection method shown in FIG. 12; FIG.

Figure 13B is a view for explaining the defect inspection method shown in Figure 12;

Figure 14A is a two-dimensional density image showing another example of the time-series distribution of the frequency of occurrence obtained by the defect inspection method of the present invention;

Figure 14B is a three-dimensional diagram showing another example of the time-series distribution of the frequency of occurrence obtained by the defect inspection method of the present invention;

Figure 15A is a two-dimensional density image showing another example of the time-series distribution of the frequency of occurrence obtained by the defect inspection method of the present invention;

Figure 15B is a three-dimensional diagram showing another example of the time-series distribution of the frequency of occurrence obtained by the defect inspection method of the present invention;

Figure 16A is a three-dimensional diagram showing, in time series, other examples of the frequency of occurrence obtained by the defect inspection method of the present invention;

Figure 16B is a three-dimensional graph showing, in time series, other examples of the frequency of occurrence obtained by the defect inspection method of the present invention;

Figure 16C is a three-dimensional diagram showing, in time series, other examples of the frequency of occurrence obtained by the defect inspection method of the present invention;

Figure 17A is a two-dimensional density image showing other examples of the frequency of occurrence obtained by the defect inspection method of the present invention in a time series;

17B is a two-dimensional density image showing other examples of the frequency of occurrence obtained by the defect inspection method of the present invention in a time series; and

Fig. 17C is a two-dimensional density image showing, in time series, another example of the frequency of generation obtained by the defect inspection method of the present invention.

S100~S180. . . step

Claims (28)

A processing apparatus which is characterized in that a defect inspection image data for inspecting a defect existing in a plate-like body by using an image obtained by relatively moving a plate-like body in a specific direction and facing the plate-like body is used. The processing device includes a processing unit that extracts a plurality of defect candidates from the image using the first signal threshold, and searches for the specific direction from the extracted plurality of defect candidates in the moving direction. The defect candidate having the same direction in the width direction of the orthogonal direction is obtained, and the position of the defect candidate detected by the search in the moving direction of the plate-shaped body and the candidate candidate in the moving direction and the detected defect candidate are obtained. The interval between the positions of the adjacent defect candidates in the moving direction is obtained by repeating the above-described processing to obtain a plurality of intervals, and the frequency of generation of the plurality of intervals is obtained, and when the frequency of the interval of attention exceeds the set frequency When the frequency threshold is used, it is determined that the plate-like body has a periodic defect in the moving direction; the frequency used in the processing unit The threshold value is defined according to the interval between the above-mentioned points of interest. When the two frequency threshold values are set to different values, the above-mentioned attention interval of the frequency threshold value of the larger one is smaller than the frequency threshold of the predetermined smaller one. In the manner of the interval, the above frequency threshold is set. The processing device according to claim 1, wherein the processing unit includes a reference table indicating a relationship between a density of occurrence of the defect candidate and the first signal threshold, so that a density of occurrence of the defect candidate is set to a target target density. The reference table sets the first signal threshold value, and the frequency threshold value is a value that varies according to the value of the target generation density in addition to the change in the interval of interest. The processing device of claim 1, wherein the frequency threshold used in the processing unit is specified in such a manner that noise components are randomly distributed in the region, and the positions in the width direction are at the same position. As the defect candidate, the signal component analytically obtains a frequency of generation of the noise component with respect to the interval, or obtains an image of the noise component in the analog image formed by the noise component as a defect candidate. The frequency threshold is determined based on the obtained generation frequency with respect to the frequency of generation of the above-described noise component with respect to the interval. The processing device of claim 3, wherein the analog image is created in such a manner that a density of noise components in the image differs depending on an image region. The processing device according to claim 1, wherein the processing unit divides the search target image for searching for the defect candidate into a plurality of portions in the width direction and the moving direction to form a plurality of unit regions having the same size, and includes When a plurality of unit regions of the plurality of defect candidates are at the same position in the width direction, the defect candidates are set to have the same position in the width direction, and the interval and the generation frequency are obtained. The processing device according to claim 1, wherein the processing unit further obtains a width direction indicating a distribution of the generation frequency at the position in the width direction with respect to the interval of the target of the defect candidate determined to have a periodic defect A frequency distribution is generated, and the width direction is used to generate the unevenness of the above-described generation frequency in the width direction in the frequency distribution, and the defect generation pattern is classified. The processing apparatus of claim 1, wherein the plate-shaped system has a long shape continuous in the moving direction; and the processing unit divides the plate-shaped body into a region having a plate-shaped body of a predetermined length, and the region is The above discrimination is performed on a plurality of time series units as an inspection target of one time series unit. The processing device according to claim 7, wherein the processing unit records the generation frequency distribution of the interval defined by the interval and the position in the width direction for the plurality of time series units, and specifies the attention frequency according to the recorded frequency distribution. The generation frequency is obtained by the interval and the position in the width direction, and the generation frequency is expressed as time-series data, whereby the defect generation information is displayed on the screen. The processing device of claim 8, wherein the processing unit changes the time series data of the plurality of generated frequencies obtained by changing at least one of the positions of the intervals and the width directions of the time series data of the generated frequency, Written in the same chart and displayed on the screen. The processing device of claim 1, wherein the processing unit searches for and detects a defect candidate having the same position in the width direction in the moving direction, and obtains the moving direction by using the adjacent defect candidate as a previous defect candidate. In addition to the upper interval, the interval between the plurality of previous defect candidates in the moving direction is also obtained, and by repeating the above-described processing, a plurality of intervals are obtained, and the frequency at which the plurality of intervals are generated is obtained. When the frequency of occurrence of the interval of the attention exceeds the set frequency threshold, it is determined that the plate-like body has a periodic defect in the moving direction. The processing device of claim 1, wherein the processing unit further defines a defect candidate having the pitch interval in the width direction when the interval between the intervals determined to have a periodic defect in the moving direction is referred to as a pitch interval. The attention area of the position where it is located, and using the second signal threshold, extracts the detailed defect candidate from the beginning of the image from the image of the attention area, and specifies the detailed defect obtained from the extraction in the moving direction. The candidate position is a search area centered on the position of the pitch interval, and in the search area, the detailed defect candidate is searched using the second signal threshold, and the detailed defect candidate detected by the search and the detailed result of the extraction are respectively evaluated. The attribute of the defect candidate determines whether or not the region of interest includes the periodic detailed defect candidate in the moving direction based on the evaluation result. The processing device of claim 1, wherein when the interval between the intervals determined to have a periodic defect in the moving direction is referred to as a pitch interval, the processing unit further defines a defect candidate having the pitch interval in the width direction In the region of interest at the position where the position is located, the detailed defect candidate is extracted from the image of the region of interest using the second signal threshold, and the detailed defect candidate obtained from the extraction in the moving direction is specified. The position is a search area centered on a position corresponding to the distance of the circumference of the transport roller that transports the plate-shaped body, and the detailed defect candidate is searched for using the second signal threshold value in the search area, and the detailed defect candidate detected by the search is evaluated. And the attribute of the detailed defect candidate obtained by the extraction, and determining whether the region of interest includes the periodic detailed defect candidate in the moving direction based on the evaluation result. The processing device according to claim 1, wherein the processing unit searches for and detects the defect candidate having the same position in the width direction in the moving direction, and obtains the interval, and evaluates the attribute of the detected defect candidate or the defect candidate. The similarity with the specific defect candidate is obtained when at least one of the attributes and the similarity satisfies the set condition. A defect inspection device characterized in that it is an inspection of a defect existing in a plate-like body, comprising: a light source that illuminates a surface of the plate-like body; and a camera whose one surface is opposite to the light source An image of a plate-like body that is irradiated with light by the light source while the plate-like body is relatively moved; and the processing device of claim 1; wherein the processing unit of the processing device uses the first signal threshold value from the camera Extracting the plurality of defect candidates from the captured image, and searching for the direction of relative movement of the camera relative to the plate-shaped body from the plurality of extracted defect candidates in the moving direction The moving direction is orthogonal to the defect candidate having the same position in the width direction. A method for producing a plate-like body, which is characterized in that it is a plate-like body which is conveyed by a conveying roller as a belt-like continuous body; and the above-mentioned board is inspected during the movement using the defect inspection device of claim 14. According to the result of the inspection, a conveying roller that causes a defect in the plate-like body on the moving path of the plate-like body is determined; and the determined conveying roller is removed or maintained. A method for producing a plate-like body, which is characterized in that it is a plate-like body which is conveyed by a conveying roller as a belt-like continuous body; and the above-mentioned board is inspected during the movement using the defect inspection device of claim 14. The body is cut and taken out of the plate-like body while avoiding the position in the width direction of the defect determined to have the periodic defect. A processing method for processing a defect inspection image data for inspecting a defect existing in the plate-like body by using an image obtained by relatively moving a plate-like body in a specific direction while detecting a relative movement in a specific direction; Using the first signal threshold value, a plurality of defect candidates are extracted from the captured image; and in the moving direction, the plurality of extracted defect candidates are searched for in the width direction orthogonal to the specific direction, that is, the moving direction. The defect candidate having the same position is obtained by finding a position between the position of the defect candidate detected by the search in the moving direction and the position of the defect candidate adjacent to the defect candidate in the moving direction in the moving direction. Obtaining a plurality of intervals by repeating the above processing; determining a frequency of generation of the plurality of intervals; and determining that the plate has a period in a moving direction when a frequency of occurrence of the interval of the attention exceeds a set frequency threshold Sexual defect; the above frequency threshold is specified according to the above-mentioned interval of attention, when the two frequency thresholds are different, The frequency threshold is set such that the above-mentioned frequency threshold of the larger one is smaller than the above-mentioned interval of the frequency threshold of the smaller one. The processing method of claim 17, wherein the checking condition is set before the determining, and in the step of setting the checking condition, the reference table is used to set the density of the defect candidate to the set target generating density. A signal threshold value; the frequency threshold value is a value that varies in accordance with the value of the target generation density in addition to the change in the interval noted above. The processing method of claim 17, wherein the frequency threshold is defined as follows: an image of the noise component of the analog image formed by the noise component is used as a defect candidate, and the above-mentioned interval is determined The frequency at which the noise component is generated is defined by the frequency threshold. The processing method of claim 19, wherein the analog image is created in such a manner that a density of noise components in the image differs depending on an image region. The processing method of claim 17, wherein the interval between the above-mentioned attention intervals determined to have a periodic defect candidate in the moving direction is referred to as a pitch interval, and after the step of performing the determining, further including The defect of the pitch interval is candidate for the region of interest at the position in the width direction; and the second signal threshold is used to extract detailed defect candidates from the image of the region of interest; The location of the detailed defect candidate obtained from the extraction is a search area centered on the position of the pitch interval; in the search area, the detailed defect candidate is searched using the second signal threshold; and the detailed defect candidate detected by the search is separately evaluated And an attribute of the detailed defect candidate obtained by the extraction; and determining whether the region of interest includes the periodic defect candidate in the moving direction based on the evaluation result. The processing method of claim 21, wherein the image used in the determination of the periodic defect candidate in the region of interest is an image of a panel obtained by cutting the plate-like body at a fixed size. The processing method of claim 17, wherein the interval between the intervals determined to have a periodic defect in the moving direction is referred to as a pitch interval, and after the step of determining, further including the defect having the pitch interval a region of interest that is located at a position in the width direction; a second signal threshold is used to extract a detailed defect candidate from the beginning of the image from the image of the region of interest; and the extraction is performed in the moving direction The position of the detailed defect candidate obtained is a search area centered on a position corresponding to the distance of the circumferential length of the transport roller that transports the plate-shaped body, and the detailed defect candidate is searched for using the second signal threshold value in the search area; The detailed defect candidate detected and the attribute of the detailed defect candidate obtained by the extraction are determined, and based on the evaluation result, it is determined whether or not the region of interest includes the periodic detailed defect candidate in the moving direction. The processing method of claim 17, wherein when the defect candidate having the same position in the width direction is searched for and detected in the moving direction and the interval is obtained, the attribute of the detected defect candidate, or the defect candidate and the specific defect candidate are evaluated. The degree of similarity is obtained when at least one of the attribute and the similarity satisfies the set condition. A defect inspection method for inspecting a defect existing in a plate-like body; and irradiating light to the surface of the plate-like body and moving the plate-like body relatively while photographing the irradiated light An image of the plate-like body; the processing method of claim 17 is performed using the above-described image obtained by photographing. A method for producing a plate-like body, which is characterized in that it is a plate-like body which is conveyed by a conveyance roller as a belt-like continuous body; and the above-mentioned board is inspected during the movement using the defect inspection method of claim 25. According to the result of the inspection, a conveying roller that causes a defect in the plate-like body on the moving path of the plate-like body is determined; and the determined conveying roller is removed or maintained. A method for producing a plate-like body, which is characterized in that it is a plate-like body which is conveyed by a conveyance roller as a belt-like continuous body; and the above-mentioned board is inspected during the movement using the defect inspection method of claim 25. The body is cut and taken out of the plate-like body while avoiding the position in the width direction of the defect determined to have the periodic defect. A recording medium readable by a computer and recorded with a computer executable program for executing a method of processing image data for defect inspection as in claim 17.