WO2019117228A1

WO2019117228A1 - Image processing system, inspection system, image processing method, and program

Info

Publication number: WO2019117228A1
Application number: PCT/JP2018/045791
Authority: WO
Inventors: 俊嗣堀井
Original assignee: パナソニックＩｐマネジメント株式会社
Priority date: 2017-12-13
Filing date: 2018-12-13
Publication date: 2019-06-20

Abstract

The present disclosure addresses the problem of providing an image processing system, an inspection system, an image processing method, and a program with which it is possible to precisely assess cracks in a solar battery cell and to precisely distinguish between cracks and other elements such as unevenness in etching. The image processing system (11) according to the present disclosure is provided with a first acquisition unit (111), a second acquisition unit (112), and an image processing device (113). The first acquisition unit (111) acquires information about a transmission image, which is an image of a solar battery cell (9) created by transmitted light (Lt11) that is transmitted through the solar battery cell (9). The second acquisition unit (112) acquires information about a PL light emission image, which is an image of the solar battery cell (9) formed of emitted light (Lt21) from the solar battery cell (9) that is produced by the radiation of excitation light (Lt2). The image processing device (113) receives the information about the transmission image and the information about the PL light emission image and performs image processing to thereby determine whether a crack has formed in the solar battery cell (9).

Description

Image processing system, inspection system, image processing method, and program

The present disclosure relates to an image processing system, an inspection system, an image processing method, and a program. More particularly, the present disclosure relates to an image processing system, an inspection system, an image processing method, and a program for inspecting a crack of a solar cell based on an image of the solar cell.

Heretofore, various inspection apparatuses have been proposed to inspect a crack in a polycrystalline silicon wafer.

The inspection apparatus of Patent Document 1 captures an image by transmitting an infrared ray transmitted to the back surface of the polycrystalline silicon wafer and an image reflected by the infrared light irradiated to the front surface of the polycrystalline silicon wafer, and captures the transmitted light The image data by and the image data by the reflected light are compared. Then, the inspection apparatus determines the presence or absence of the internal crack of the polycrystalline silicon wafer based on the comparison result of the image data by the transmitted light and the image data by the reflected light.

In the conventional inspection apparatus, other elements such as etching unevenness of a silicon wafer (solar battery cell) may be erroneously determined to be a crack in some cases, and improvement in crack detection accuracy has been desired.

JP 2007-218638 A

The present disclosure is an image processing system and inspection system capable of accurately determining a crack of a solar battery cell based on a captured image of a solar battery cell and accurately distinguishing the crack and other elements such as etching unevenness, An object of the present invention is to provide an image processing method and program.

An image processing system according to an embodiment of the present disclosure is an image processing system for inspecting a crack of the solar battery cell based on an image of the solar battery cell. The image processing system includes a first acquisition unit, a second acquisition unit, and an image processing apparatus. The first acquisition unit acquires information of a transmission image which is an image of the solar battery cell by transmitted light transmitted through the solar battery cell. The second acquisition unit acquires information of a light emission image which is an image of the solar battery cell by the light emission of the solar battery cell. The image processing apparatus receives the information of the transmission image and each information of the light emission image and performs image processing to determine whether the crack is generated in the solar battery cell.

An inspection system according to an aspect of the present disclosure includes an imaging unit, a first illumination device, a light emission generation device, and the above-described image processing system. The imaging unit images a solar battery cell. The first lighting device emits illumination light to the solar battery cell to generate transmission light in which the illumination light is transmitted through the solar battery cell. The light emitting device causes the solar cell to emit light. The imaging unit receives the transmitted light, outputs information of a transmitted image to the image processing system, receives the light emission, and outputs information of a light emission image to the image processing system.

An image processing method according to an aspect according to the present disclosure is an image processing method for inspecting a crack of the solar battery cell based on an image of the solar battery cell. The image processing method includes an acquisition step and an image processing step. The acquiring step acquires information of a transmission image which is an image of the solar battery cell by transmitted light which has been transmitted through the solar battery cell, and information of a light emission image which is an image of the solar battery cell by light emission of the solar battery cell To get The image processing step receives the information of the transmission image and each information of the light emission image and performs image processing to determine whether the crack is generated in the solar battery cell.

A program according to an aspect of the present disclosure causes a computer system to execute the above-described image processing method.

FIG. 1 is a block diagram showing the configuration of an inspection system provided with the image processing system of the embodiment. FIG. 2 is a schematic view showing the inspection system of the same. FIG. 3 is a plan view showing the light receiving surface of the above solar cell. FIG. 4 is a view showing a crack area of the above solar cell. FIG. 5A is a view showing a gray-scale image (transmission image, PL light emission image) of the solar battery cell of the above. FIG. 5B is a view showing a binary image obtained by binarizing the grayscale image of the same. FIG. 6 is a view showing a corrected gray-scale image (corrected transmission image, corrected PL light emission image) after position correction of the same. FIG. 7 is a diagram showing a setting example of candidate areas and inspection areas in the corrected gray-scale image of the same. FIG. 8A is a diagram showing a setting example of an inspection area in the corrected transmission image of the same as above. FIG. 8B is a diagram showing a setting example of an inspection area in the corrected PL light emission image of the same as above. FIG. 9 is a schematic view showing the neural network of the same. FIG. 10 is a flow chart showing an image processing method of the same. FIG. 11 is a block diagram showing another configuration of the inspection system of the same. FIG. 12 is a block diagram showing another configuration of the inspection system of the same.

The inspection system of a solar battery cell described below includes an image processing system that extracts a crack occurring in an inspection target by image processing technology using a computer. The inspection object is a single crystal material formed in a plate shape, and in particular, a solar battery cell in which single crystal silicon is stacked is assumed. In addition, if the photovoltaic cell used as a test object has regularity in the direction which a crack produces, it is not essential that it is single-crystal material.

FIG. 1 shows a block configuration of an inspection system A1 of the present embodiment. FIG. 2 shows a schematic configuration of the inspection system A1. The inspection system A1 mainly includes a computer system 1, a first illumination device 2, a light emission generation device 3 (second illumination device 3A), and an imaging unit 4. The inspection system A1 preferably further includes a transport device 5, a display device 6, and a device control device 10. And inspection system A1 performs a crack inspection of photovoltaic cell 9 as a defect inspection of photovoltaic cell 9.

The solar battery cell 9 is a cell that constitutes a solar battery module. As shown in FIG. 3, the solar battery cell 9 is formed in a square plate shape (for example, a square shape having one side of about 125 mm). For the solar battery cell 9, for example, a hetero junction with intrinsic thin-layer (HIT) (registered trademark) solar cell is used. In the HIT solar cell, an amorphous silicon layer is formed on both sides of a single crystal silicon layer. The light receiving surface 91 of the solar battery cell 9 has a bluish color. Furthermore, on the light receiving surface 91 of the solar battery cell 9, a plurality of bus bar electrodes 92 are linearly formed in the vertical direction (vertical direction in FIG. 3), and a plurality of finger electrodes 93 are horizontal direction (horizontal direction in FIG. 3) It is formed linearly. The plurality of bus bar electrodes 92 formed in the vertical direction are aligned in the horizontal direction. The plurality of finger electrodes 93 formed in the horizontal direction are aligned in the vertical direction. The bus bar electrode 92 and the finger electrode 93 are wirings which flow a current generated by sunlight, and are formed of silver paste or aluminum paste or the like, and the surface has a silver color.

Generally, when a crack is generated in the solar battery cell 9, the location of the crack becomes dark. In addition, when the solar battery cell 9 includes a single crystal silicon layer, the cracks generated in the single crystal silicon layer have regularity in the direction in which the cracks are generated, and are generated linearly and only in a certain direction. Therefore, in a defect inspection that determines the presence or absence of a crack using a gray image obtained by imaging the solar battery cell 9, in general, the presence or absence of a crack is determined based on the gray value of the pixel, the direction, and the sharpness of the line. .

Since the solar battery cell 9 of the present embodiment is provided with a single crystal silicon layer, there is a high possibility that a crack is generated in the direction of cleavage of the single crystal silicon layer inside the solar battery cell 9. In particular, the single crystal silicon layer of the solar battery cell 9 is as thin as 500 μm or less in thickness dimension, and the crack inspection is effective from the viewpoint of quality assurance. And the photovoltaic cell 9 determined that the crack has arisen is classified into inferior goods. Cleavage refers to a state in which a crystal such as a single crystal silicon layer is broken along one of the crystallographic directions (plane directions of cleavage planes). Moreover, the direction of cleavage is the surface direction of the cleavage surface, and is the direction in which the crack extends. For example, a mineral having multiple cleavage planes has multiple cleavage directions, and cracks develop along at least one of the multiple cleavage directions.

FIG. 4 is an enlarged view of a part of the light receiving surface 91 of the solar battery cell 9. In FIG. 4, the crack 8 is generated in the solar battery cell 9, and the crack 8 extends along the cleavage direction of the single crystal silicon layer. The crack 8 has a linear shape, and the direction in which the crack 8 extends is approximately 45 degrees (or 225 degrees) with respect to the direction (horizontal direction) in which the finger electrode 93 extends in a plane parallel to the light receiving surface 91 (first Direction), or an angle of 135 degrees (or 315 degrees) (second direction) is likely to be made. Therefore, in the captured image, the area where the crack 8 is imaged (crack area) is also linear like the crack 8 and the direction in which the crack area extends is approximately 45 degrees (or 225 degrees) with respect to the horizontal direction Or the possibility of forming an angle of 135 degrees (or 315 degrees) is high. That is, the direction in which the gray value changes can be used as the information for determining the crack area.

First, the solar battery cell 9 is moved by the transport device 5 below the second lighting device 3A and the imaging unit 4 and above the first lighting device 2 in the forward direction X1 from the rear to the first lighting, as shown in FIG. Each lighting process by the device 2 and the second lighting device 3A and an imaging process by the imaging unit 4 are performed.

The conveying device 5 is configured of, for example, a belt conveyor device. The conveying device 5 includes a motor, a pulley, a conveyor belt, and the like, and the rotational driving force of the motor is transmitted to the conveyor belt via the pulley and the like. The solar cells 9 placed on the conveyor belt advance in the traveling direction X1 and temporarily stop at a predetermined inspection position (immediately below the imaging unit 4), and after a crack inspection is performed, the traveling direction X1 And the next photovoltaic cell 9 is temporarily stopped at the inspection position. The light receiving surface 91 of the solar cells 9 mounted on the conveyor belt is directed upward. Further, the upper and lower sides of the transfer device 5 are arranged such that illumination light of the first illumination device 2 disposed below the transfer device 5 is irradiated from below to the solar battery cells 9 stopped at the inspection position. An opening communicating with the direction is provided.

Alternatively, the solar cell 9 may be fixed, and the first lighting device 2, the second lighting device 3A, and the imaging unit 4 may move in the traveling direction X1 (or in the opposite direction to the traveling direction X1).

And inspection system A1 performs the following inspection processing to photovoltaic cell 9 which stopped at the inspection position.

First, the first lighting device 2 is disposed below the transport device 5. The first lighting device 2 emits illumination light Lt1 from below to the solar battery cell 9 stopped at the inspection position. And illumination light Lt1 which injected into the lower surface of the photovoltaic cell 9 permeate | transmits the inside of the photovoltaic cell 9, and radiate | emits upwards from the light-receiving surface 91 of the photovoltaic cell 9 as the transmitted light Lt11. That is, the first lighting device 2 is a light source for generating the transmitted light Lt11 which is the light transmitted through the solar battery cell 9. The wavelength of the illumination light Lt1 is preferably in the range of 900-1300 nm at which the efficiency of transmission through the solar battery cell 9 is relatively high. In the present embodiment, the wavelength of the illumination light Lt1 is 950 nm, and the illumination light Lt1 corresponds to near infrared light. The wavelength of the transmitted light Lt11 is equal to the wavelength of the illumination light Lt1.

The second lighting device 3 </ b> A is an embodiment of the light emission generation device 3 that causes the solar cell 9 to emit light. The second lighting device 3 </ b> A is disposed above the transport device 5. The second lighting device 3A emits excitation light Lt2 from above to the solar battery cell 9 stopped at the inspection position. Then, the solar battery cell 9 irradiated with the excitation light Lt2 emits light by photoluminescence (PL) to emit a light emission Lt21. That is, the second lighting device 3A is a light source for causing the solar battery cell 9 to emit PL light to generate the light emission Lt21. In the present embodiment, the wavelength of the excitation light Lt2 is 630 nm, and the excitation light Lt2 corresponds to red light. The wavelength of light emission Lt21 due to PL light emission of the solar battery cell 9 substantially falls within the range of 1000-1100 nm.

In addition, although inspection system A1 of FIG. 1, FIG. 2 is equipped with one 1st illuminating device 2 and two 2nd illuminating devices 3A, each number of the 1st illuminating devices 2 and 2nd illuminating devices 3A is Not limited to a specific number.

The imaging unit 4 includes a first imaging device 41, a second imaging device 42, and a filter 42a. The first imaging device 41 and the second imaging device 42 are respectively disposed above the transport device 5. The first imaging device 41 and the second imaging device 42 respectively image the light receiving surface 91 of the solar battery cell 9 stopped at the inspection position from above. The filter 42 a is provided in the second imaging device 42. The filter 42 a attenuates the excitation light Lt 2 and passes the light emission Lt 21. For example, the filter 42 a according to the present embodiment attenuates light having a wavelength of less than 900 nm and transmits light having a wavelength of 900 nm or more. And the 2nd imaging device 42 light-receives the light which passed the filter 42a, and images it. The second imaging device 42 can increase the light reception amount of the light emission Lt 21 by the filter 42 a compared to the light reception amount of the excitation light Lt 2.

And each of the 1st imaging device 41 and the 2nd imaging device 42 is a camera which images a black-and-white still picture, and is arrange | positioned so that the whole of one photovoltaic cell 9 may be stored in an imaging range. Each of the images captured by the first imaging device 41 and the second imaging device 42 is a grayscale image in which the grayscale level is set to, for example, 256 levels. The first imaging device 41 outputs the information of the gray-scale image to the computer system 1 as the information of the transmission image. The second imaging device 42 outputs the gray scale image information to the computer system 1 as the PL light emission image information. Each of the first imaging device 41 and the second imaging device 42 may be a camera that captures a color image.

The operations of the first lighting device 2, the second lighting device 3A, the first imaging device 41, the second imaging device 42, and the transport device 5 are controlled by the device control device 10.

Specifically, by controlling the transport device 5, the device control device 10 moves the solar battery cell 9 to be inspected in the X1 direction to the predetermined inspection position. Then, the device control device 10 causes the first imaging device 41 to image the light receiving surface 91 of the solar battery cell 9 when the first lighting device 2 is turned on and the second lighting device 3A is turned off. In this case, the first imaging device 41 receives the transmitted light Lt11 to generate information of a gray image, and outputs the information of the gray image by the transmitted light Lt11 to the computer system 1 as information of a transmitted image.

The device control device 10 causes the second imaging device 42 to image the light receiving surface 91 of the solar battery cell 9 when the first lighting device 2 is turned off and the second lighting device 3A is turned on. In this case, the second imaging device 42 receives the light emission Lt21 to generate information of the gray scale image, and outputs the information of the gray scale image by the light emission Lt21 to the computer system 1 as the information of the PL light emission image (light emission image) Do.

The computer system 1 monitors each operation of the first lighting device 2, the second lighting device 3A, the first imaging device 41, and the second imaging device 42 by the device control device 10, and transmits information of a transmission image and PL light emission image. Receive information The computer system 1 may have the function of the device control apparatus 10.

The computer system 1 executes an image processing program to perform image processing on information on a transmission image and information on a PL emission image. Furthermore, it is preferable that the computer system 1 executes a program for image processing to perform a crack inspection of the solar battery cell 9 using the information of the transmission image and the information of the PL light emission image.

The computer system 1 mainly includes a processor and memory as hardware. The processor executes the program (program for image processing) recorded in the memory (non-transitory recording medium) of the computer system 1 to realize the function of the image processing system 11 described later in the present disclosure. The program may be pre-recorded in the memory of the computer system 1, but may be provided through a telecommunication line, or may be non-transitory recording such as a memory card, an optical disc, a hard disk drive, etc. readable by the computer system 1. It may be recorded on a medium and provided. The processor of the computer system 1 is configured of one or more electronic circuits including a semiconductor integrated circuit (IC) or a large scale integrated circuit (LSI). The plurality of electronic circuits may be integrated into one chip or may be distributed to a plurality of chips. The plurality of chips may be integrated into one device or may be distributed to a plurality of devices.

The computer system 1 functions as an image processing system 11 by executing a program for image processing.

The image processing system 11 includes a first acquisition unit 111, a second acquisition unit 112, an image processing device 113, and an output unit 114, and determines whether or not the crack 8 is generated in the solar battery cell 9 to be inspected. Do.

First, the first acquisition unit 111 receives the information of the transmission image from the first imaging device 41. The second acquisition unit 112 receives the information of the PL light emission image from the second imaging device 42. Then, the image processing apparatus 113 receives the information of the transmission image from the first acquisition unit 111, and receives the information of the PL light emission image from the second acquisition unit 112.

The image processing apparatus 113 includes a position correction unit 113a, an area setting unit 113b, and an inspection unit 113c.

The position correction unit 113a performs position correction processing on each of the transmission image and the PL light emission image. The position correction process is respectively performed on the transmission image and the PL light emission image in the virtual two-dimensional space generated by the computer system 1. In the position correction processing of the present embodiment, coordinate conversion processing is performed so that each direction of the region where the solar battery cells 9 appear in the transmission image and the region where the solar battery cells 9 appear in the PL light emission image becomes the reference direction. Is done. In addition, since the content of each position correction | amendment processing of a permeation | transmission image and PL light emission image is the same, in the following description, position correction | amendment of the grayscale image G1 shown to FIG. The process will be described. That is, reading gray-scale image G1 in the following description as a transmission image corresponds to position correction processing of a transmission image, and reading gray-scale image G1 as a PL light emission image corresponds to position correction processing of a PL light emission image.

The solar battery cell 9 is shown in the gray image G1, and in the gray image G1, the region where the solar battery cell 9 is shown is taken as a cell region 71. The cell area 71 has a substantially rectangular shape as in the solar battery cell 9.

The position correction unit 113 a extracts the positions of the four corners of the cell area 71 from the gray-scale image G1 of the solar battery cell 9. The position correction unit 113a performs differentiation processing or binarization processing on the grayscale image G1 to extract the cell area 71 in the grayscale image G1. Then, the position correction unit 113a arranges four points P1, P2, P3, and P4 for position correction at positions of four corners of the cell area 71, respectively.

For example, the position correction unit 113a uses a binary image G2 (see FIG. 5B) obtained by binarizing the gray value of each pixel of the gray image G1 so that the four sides surrounding the outer periphery of the cell area 71 can be easily extracted. The positions of four points P1, P2, P3 and P4 are determined. Points P1, P2, P3 and P4 are two sides parallel to finger electrode 93 among four sides surrounding the outer periphery of cell area 72 (an area corresponding to cell area 71 of gray image G1) in binary image G2. These are intersections obtained by extending two sides orthogonal to the finger electrode 93. When the position correction unit 113a extracts the coordinates of four points P1, P2, P3, and P4 in the binary image G2, the four points P1, P2, P3, and P4 in the grayscale image G1 are the same as the binary image G2. Arrange at each coordinate. Then, the position correction unit 113a performs coordinate conversion of each pixel in the grayscale image G1 such that a quadrangle having four points P1, P2, P3, and P4 of the grayscale image G1 as the apex points in the reference direction. At this time, the position correction unit 113a matches one or more predetermined ones of the four points P1, P2, P3, and P4 of the grayscale image G1 with the reference coordinates.

By subjecting the transmission image to the position correction processing applied to the gray image G1, the cell area 71 in the transmission image can be oriented in the reference direction. In addition, the position correction process performed on the gray image G1 described above is performed on the PL light emission image, so that the cell region 71 in the PL light emission image can be oriented in the reference direction. In other words, when the position correction unit 113a generates the transmission image and the PL light emission image in the virtual two-dimensional space, the imaging region of the finger electrode 93 extends along the 0 ° direction or the 180 ° direction of the two-dimensional space. In this way, coordinate conversion processing is performed. As a result, the cell regions 71 of the transmission image and the cell regions 71 of the PL light emission image face the same predetermined direction, which will be described later by the region setting unit 113 b as compared with the case where the position correction processing is not performed. The accuracy of the extraction process of the candidate area 80 and the inspection area 81 is improved.

Further, even if the relative position of the first imaging device 41 to the solar battery cell 9 and the relative position of the second imaging device 42 to the solar battery cell 9 are different, the cell region in the transmission image in the virtual two-dimensional space 71 and the coordinates of the cell area 71 in the PL light emission image match (almost match).

The position correction unit 113a generates the corrected gray-scale image G3 shown in FIG. 6 by performing the above-described position correction process on the gray-scale image G1. When the gray image G1 is a transmission image, the corrected gray image G3 corresponds to a corrected transmission image. When the grayscale image G1 is a PL light emission image, the corrected grayscale image G3 corresponds to a corrected PL light emission image.

The area setting unit 113b performs a sharpening process on the corrected gray-scale image G3 (corrected transmission image and corrected PL light emission image) to determine a candidate area 80 in which a crack may occur, and further surrounds the candidate area 80. The inspection area 81 is determined.

When cracks 8 occur in the solar battery cell 9, the direction in which the candidate region 80 extends in the corrected gray-scale image G3 is the direction of 45 degrees (225 degrees) or 135 degrees (315 degrees) in a virtual two-dimensional space . Therefore, the area setting unit 113b evaluates the direction in which the gray value in the corrected gray-scale image G3 changes, and extracts the candidate area 80 extending in the direction of 45 degrees or 135 degrees in the virtual two-dimensional space. The region setting unit 113 b uses a two-dimensional wavelet filter, preferably a two-dimensional Gabor filter, in order to evaluate the direction in which the gray value changes.

The two-dimensional Gabor filter is a spatial filter that uses a kernel represented by a Gabor function. The Gabor function is a product of a harmonic function represented by a sine function and a cosine function and a Gaussian function. Therefore, the Gabor filter includes an angle (angle parameter) in the parameter. The kernel of the Gabor filter is n pixels × n pixels, and n is selected in an appropriate range of about 5 or more and 256 or less.

When the Gabor filter is applied to the corrected gray-scale image G3, a two-dimensional Gabor filter is superimposed on part of the corrected gray-scale image G3. Next, in a region where the Gabor filter and the corrected gray-scale image G3 overlap, the gray-scale value of the corrected gray-scale image G3 is multiplied by the signal value of the Gabor filter for each pixel, and the Gabor filter and the corrected gray-scale image G3 overlap. The sum of the multiplied values for all the pixels in the area is determined. The sum of the multiplication values obtained in this manner becomes the pixel value at the center of the kernel. That is, convolution of the gray-scale image by the Gabor filter is performed. Therefore, it is desirable that the value of n that determines the size of the kernel is an odd number. When the change in the lightness value of the area where the Gabor filters overlap in the corrected light and dark image G3 and the parameter of the Gabor filter have a predetermined relationship, the signal value after the convolution operation becomes a large value. If the signal value after the convolution operation is regarded as the response of the Gabor filter, it can be said that the response of the Gabor filter becomes strong when the change of the gray value in the corrected gray-scale image G3 and the parameter of the Gabor filter have a predetermined relationship.

The area setting unit 113b scans the Gabor filter whose angle parameter is set at 45 degrees and 135 degrees in the corrected gray-scale image G3 after position correction, and performs convolution calculation by the Gabor function for each position where the Gabor filter is overlapped. . That is, the area setting unit 113b performs Gabor wavelet transform on the corrected gray-scale image G3. In the corrected gray-scale image G3 on which Gabor wavelet transformation has been performed, a region representing extension in the direction of 45 degrees or 135 degrees is emphasized by the change in gray value. Here, the kernel of the Gabor filter used by the region setting unit 113b is smaller than the corrected gray-scale image G3. The area setting unit 113 b uses, for example, a Gabor filter having a size of about 41 pixels × 41 pixels.

Note that the cleavage direction of single crystal silicon is 45 degrees or 135 degrees with respect to the horizontal direction. Therefore, the Gabor filter angle parameters are set to 45 degrees and 135 degrees. As a result, the change of the gray value in the corrected gray image G3 becomes large in the direction orthogonal to the direction of cleavage.

When scanning the Gabor filter, the distance for moving the Gabor filter once is set smaller than the number of pixels on one side of the Gabor filter. The distance for moving the Gabor filter once may be at least one pixel. As the distance for moving the Gabor filter once increases, the time required for scanning decreases, but the probability of missing a candidate area increases. Therefore, the distance by which the Gabor filter is moved once is appropriately determined by the relationship between time and accuracy.

Then, the region setting unit 113b further performs at least one of binarization processing and differentiation processing on the corrected gray-scale image G3 subjected to Gabor wavelet transform to generate a sharpened image. In a sharpened image, a region representing extension in the direction of 45 degrees or 135 degrees is further emphasized as a candidate region of a crack due to a change in gray value.

FIG. 7 shows a part of a sharpened image G4 (binary image) in which the corrected gray-scale image G3 is subjected to Gabor wavelet transformation and then binarized. In FIG. 7, in the cell area 73 of the sharpened image G4 (area corresponding to the cell area 71 of the gray image G1), a pixel of a dark pixel or a bright pixel which is a candidate for a crack extending in the 45.degree. Groups are highlighted as candidate regions 80. In the sharpened image G4, a plurality of candidate areas 80 may exist.

Then, the region setting unit 113b sets a circumscribed rectangle D1 (see FIG. 7) circumscribing the candidate region 80 in the sharpened image G4. The circumscribed rectangle D1 circumscribing the candidate area 80 is set to have sides parallel to the 0 degree (or 180 degree) direction and the 90 degree (or 270 degree) direction of the two-dimensional space in the sharpened image G4. . Not all the pixels constituting the candidate area 80 are connected, but there may be a gap of several pixels between the pixels.

Then, the region setting unit 113b applies the circumscribed rectangle D1 in the sharpened image G4 to the corrected gray-scale image G3 (corrected transmission image and corrected PL light emission image). That is, the area setting unit 113b sets the coordinates at which the circumscribed rectangle D1 is located in the sharpened image G4 and the coordinates at which the circumscribed rectangle D1 is located in the corrected grayscale image G3 as the same coordinates. Then, the area setting unit 113b sets an area within the circumscribed rectangle D1 in the corrected gray-scale image G3 as the inspection area 81.

That is, when the corrected gray-scale image G3 is a corrected transmission image, an inspection area 81 is set in the corrected transmission image. When the corrected gray-scale image G3 is a corrected PL light emission image, an inspection area 81 is set in the corrected PL light emission image. Then, the corrected transmission image in which the inspection area 81 is set and the corrected PL light emission image in which the inspection area 81 is set are delivered to the inspection unit 113c.

The inspection unit 113c executes a recognition algorithm on the inspection area 81 of the corrected transmission image and the inspection area 81 of the correction PL light emission image to determine whether or not the crack 8 is generated in the solar cell 9 to be inspected. Determine

FIG. 8A shows a part of the corrected transmission image G3a as the corrected gray-scale image G3. The candidate area 80a is shown in the cell area 71a (an area corresponding to the cell area 71 of the gray-scale image G1). An inspection area 81a to be included is set. FIG. 8B shows a part of the corrected PL light-emitting image G3b as the corrected gray-scale image G3. The candidate area 80b is shown in the cell area 71b (an area corresponding to the cell area 71 of the gray-scale image G1). An inspection area 81 b including the above is set. In the corrected transmission image G3a and the corrected PL light emission image G3b, etching unevenness of the solar battery cell 9, etching pattern, abrasion marks of the surface film, dirt, and the like are respectively reflected as noise. However, when the corrected transmission image G3a and the correction PL light emission image G3b are compared, the noise concentration and the noise density such as the distribution are different from each other. For example, in the corrected transmission image G3a, etching unevenness of the solar battery cell 9 and noise due to the etching pattern are likely to be captured. In addition, in the corrected PL light emission image G3b, noise due to abrasion marks on the surface film is likely to appear.

Therefore, the inspection unit 113c according to the present embodiment determines whether or not the crack 8 is generated in the solar battery cell 9 using the corrected transmission image G3a and the correction PL light emission image G3b having different noise densities. As a result, the inspection unit 113c can improve the determination accuracy of the crack inspection as compared to the prior art by using or complementing the features of the corrected transmission image G3a and the corrected PL light emission image G3b. That is, based on the captured image of the solar battery cell 9, the inspection unit 113c can accurately determine the crack 8 of the solar battery cell 9, and can accurately distinguish the crack 8 from other elements (noises). Other factors include uneven etching, etched patterns, scratches on the surface film, and dirt.

In the present embodiment, the inspection unit 113c preferably executes a recognition algorithm constructed by machine learning, in particular, deep learning. In the recognition algorithm constructed by deep learning, a transmission image and a PL light emission image of the solar battery cell 9 in which a crack 8 is generated using a neural network of a multilayer structure in advance, the solar battery cell 9 in which the crack 8 is not generated. It is constructed by deep learning in which each data such as transmission image and PL emission image is used as a learning model. Then, whether or not the

candidate regions

80a and 80b respectively shown in the corrected transmission image G3a and the corrected PL emission image G3b correspond to the cracks 8 by deep learning using a huge number of learning models (ie, solar cells) A feature amount for determining whether or not a crack 8 has occurred in 9 is extracted, and a recognition algorithm is constructed. The inspection unit 113 c further improves the accuracy of the crack inspection process by executing the deep learning recognition algorithm.

For example, the inspection unit 113c is implemented with a neural network NN1 shown in FIG. In this case, information of the corrected transmission image G3a in which the inspection area 81a is set and the corrected PL light emission image G3b in which the inspection area 81b is set is input to the neural network NN1. Therefore, the inspection unit 113c has a simpler configuration than when a plurality of neural networks are mounted.

The neural network NN1 is a convolutional neural network (Convolutional Neural Network) including

convolutional layers

101 and 102, pooling

layers

103 and 104, a coupling layer 105, an inception module 106, a coupling layer 107, and a total coupling layer 108.

The convolution layer 101 is a convolution filter which receives the information of the corrected transmission image G3a in which the inspection region 81a is set, and performs a convolution operation of the inspection region 81a of the correction transmission image G3a and the image patch representing the feature of the crack 8. . The result of the convolution operation by the convolution layer 101 is output from the convolution layer 101 as data of the same size as the inspection area 81 a. The convolution layer 102 is a convolution filter that receives the corrected PL light emission image G3b in which the inspection area 81b is set, and performs a convolution operation of the inspection area 81b of the corrected PL light emission image G3b and the image patch representing the feature of the crack 8. . The result of the convolution operation by the convolution layer 102 is output from the convolution layer 102 as data of the same size as the inspection area 81 b.

The pooling layer 103 reduces the data output by the convolutional layer 101, and the pooling layer 104 reduces the data output by the convolutional layer 102. The bonding layer 105 delivers each output of the pooling layers 103 and 104 to the inception module 106. In the inception module 106, a plurality of convolution filters are provided in parallel, and each of the plurality of convolution filters performs a convolution operation. Bonding layer 107 delivers each output of inception module 106 to all bonding layers 108.

In the total bonding layer 108, whether or not the

candidate areas

80a and 80b reflected in the

inspection areas

81a and 81b correspond to the cracks 8 of the solar cells 9 based on the respective outputs of the inception module 106 (ie, solar cells It is determined whether or not the crack 8 is generated in the cell 9).

The image processing system 11 includes an output unit 114. The output unit 114 outputs image data including an image of the solar battery cell 9 in which the location of the candidate area 80 corresponding to the crack 8 is indicated, and each information such as the determination process, as a result of the crack inspection. Are output to the display device 6. The display device 6 displays the result of the crack inspection based on the received image data. The display device 6 is any one of a liquid crystal display, an organic EL display EL: Electroluminescence), a CRT (Cathode Ray Tube), and the like.

In addition, the inspection unit 113c receives each information of the first neural network to which the information of the corrected transmission image G3a in which the inspection area 81a is set is input and the correction PL light emission image G3b in which the inspection area 81b is set. 2 A neural network may be provided. In this case, the inspection unit 113c has a final determination unit that creates and outputs one determination result based on the determination result of the first neural network and the determination result of the second neural network. The final determination unit may be a neural network or an arithmetic unit that performs a logical operation of two inputs.

In addition, the inspection unit 113c may be implemented with a support vector machine (SVM). The support vector machine receives the information of the corrected transmission image G3a in which the inspection area 81a is set and the correction transmission image G3b in which the inspection area 81b is set. Then, the support vector machine determines whether the

candidate areas

80 a and 80 b correspond to the cracks 8 of the solar battery cell 9. The inspection unit 113c obtains in advance each feature amount of the corrected transmission image G3a in which the inspection region 81a is set and the correction transmission image G3b in which the inspection region 81b is set, and inputs these feature amounts to the support vector machine. Is preferred.

The image processing method by the above-described image processing system 11 is summarized as a flowchart of FIG.

First, the first acquisition unit 111 receives the information of the transmission image from the first imaging device 41. The second acquisition unit 112 receives the information of the PL light emission image from the second imaging device 42. That is, the image processing system 11 receives the information of the transmission image and the information of the PL emission image (step S1). Then, the image processing system 11 executes image processing by the above-described image processing apparatus 113 (the position correction unit 113a, the area setting unit 113b, and the inspection unit 113c) (step S2).

Further, as in the inspection system A2 shown in FIG. 11, the imaging unit 4 may be configured to include the second imaging device 42 and the filter 42 a and not include the first imaging device 41. In this case, the filter 42a transmits the transmitted light Lt11 and the light emission Lt21 and attenuates the excitation light Lt2. The inspection system A2 can reduce the number of imaging devices as compared with the inspection system A1, and can achieve cost reduction and simplification of the system configuration.

Moreover, in test | inspection system A3 shown in FIG. 12, the power supply device 3B is provided as another form of the light emission generator 3 which makes light emission of the photovoltaic cell 9 occur. The power supply device 3 </ b> B supplies power to the solar battery cells 9 to flow forward current in the solar battery cells 9. The solar battery cell 9 to which the forward current is supplied emits light by electroluminescence (EL) to emit light emission Lt 31. That is, the power supply device 3B causes the solar battery cell 9 to emit light to generate the light emission Lt31.

Then, the device control device 10 controls the transport device 5 to move the solar battery cell 9 to be inspected in the X1 direction to the predetermined inspection position. Then, when the first lighting device 2 is turned off, the device control device 10 causes the power supply device 3B to supply power to the solar battery cell 9, and causes the second imaging device 42 to receive the light receiving surface 91 of the solar battery cell 9. Take an image. In this case, the second imaging device 42 receives the light emission Lt31 and generates information of a gray scale image, and outputs the information of the gray scale image by the light emission Lt31 to the computer system 1 as information of an EL light emission image (light emission image) Do. The inspection system A3 does not have to include the filter 42a (see FIG. 1).

The computer system 1 performs image processing on the information on the EL light emission image as in the case of the information on the PL light emission image described above. The content of the image processing performed on the information of the EL light emission image is the same as the content of the image processing performed on the information of the PL light emission image, and thus the description thereof is omitted.

Therefore, inspection system A3 can also distinguish crack 8 of photovoltaic cell 9 accurately based on the image pick-up picture of photovoltaic cell 9. Then, the inspection system A3 can accurately distinguish the crack 8 from other elements (noise) such as etching unevenness, etching patterns, abrasion marks of the surface film, and dirt.

The image processing system (11) according to the first aspect of the present invention is an image processing system for inspecting a crack (8) of a solar cell (9) based on an image of the solar cell (9). is there. The image processing system (11) includes a first acquisition unit (111), a second acquisition unit (112), and an image processing apparatus (113). The first acquisition unit (111) acquires information of a transmission image (grayscale image G1) which is an image of the solar battery cell (9) by the transmitted light (Lt11) transmitted through the solar battery cell (9). The second acquisition unit (112) acquires information of a light emission image (grayscale image G1) that is an image of the solar battery cell (9) by the light emission (Lt21 or Lt31) of the solar battery cell (9). The image processing apparatus (113) receives the information of the transmission image and the respective information of the light emission image and performs image processing to determine whether or not a crack (8) has occurred in the solar battery cell (9).

Therefore, the image processing system (11) can accurately determine the crack (8) of the solar battery cell (9) based on the captured image of the solar battery cell (9). Then, the image processing system (11) can accurately distinguish the crack (8) from other elements (noises) such as etching unevenness, etching patterns, abrasion marks of the surface film, dirt and the like.

In the image processing system (11) according to the second aspect of the embodiment, in the first aspect, the light emission image is captured by the solar battery cell (9) emitting light by irradiation of the excitation light (Lt2). It is preferable that the obtained PL light emission image (grayscale image G1).

Therefore, the image processing system (11) can accurately determine the crack (8) of the solar battery cell (9) based on the captured image of the solar battery cell (9). And an image processing system (11) can distinguish a crack (8) and other elements with sufficient accuracy.

In the image processing system (11) of the third aspect according to the embodiment, in the first aspect, the light emitting image is an EL from which a solar battery cell (9) emitting light is supplied by being supplied with current. It is preferable that it is a luminescent image (grayscale image G1).

In the image processing system (11) of the fourth aspect according to the embodiment, in any one of the first to third aspects, the transmission image and the light emission image are gray-scale images, respectively. The image processing apparatus (113) preferably includes a position correction unit (113a), an area setting unit (113b), and an inspection unit (113c). In the position correction unit (113a), the cell region (71) in which the solar battery cell (9) is shown in the transmission image, and the cell region (71) in which the solar battery cell (9) is shown in the light emission image are each specified. A position correction process is performed to correct the transmission image and the light emission image so as to face in the direction. The position correction unit (113a) corrects the corrected transmission image (G3a), which is a transmission image subjected to position correction processing, and the corrected light emission image (for example, a correction PL light emission image G3b), which is a light emission image subjected to position correction processing. Generate The area setting unit (113b) generates information of an inspection area (81a, 81b) in which it is estimated that the crack (8) appears in the corrected transmission image (G3a) and the correction emission image. The inspection unit (113c) determines whether or not a crack (8) has occurred in the solar battery cell (9) based on the corrected transmission image (G3a) and the correction light emission image in which the inspection regions (81a, 81b) are respectively set Determine

The predetermined direction in the position correction process is a reference direction. The reference direction is one predetermined direction in a virtual two-dimensional space in which a transmission image and a light emission image are generated.

Therefore, in the image processing system (11), since the position correction processing is performed by the position correction unit (113a), the inspection area (81a, 81b) by the area setting unit (113b) is compared with the case where the position correction processing is not performed. The accuracy of the extraction process of 81 b) is improved. As a result, the inspection accuracy by the inspection unit (113c) is also improved.

In the image processing system (11) according to the fifth aspect of the embodiment, in the fourth aspect, the inspection unit (113c) corrects the transmission image (G3a) in which the inspection areas (81a, 81b) are set. It is preferable to determine whether or not a crack (8) has occurred in the solar battery cell (9) by executing a recognition algorithm constructed by machine learning on the corrected light emission image.

Therefore, the image processing system (11) further improves the accuracy of the crack inspection process by executing the recognition algorithm constructed by machine learning.

In the image processing system (11) according to the sixth aspect of the present invention, in the fifth aspect, the machine learning is preferably deep learning. And an inspection part (113c) has a neural network (NN1) which performs a recognition algorithm constructed by deep learning.

Therefore, the image processing system (11) further improves the accuracy of the crack inspection process by executing the recognition algorithm constructed by deep learning.

In the image processing system (11) of the seventh aspect according to the embodiment, in the sixth aspect, it is preferable that the neural network (NN1) has two convolutional layers (101, 102). Two convolutional layers (101, 102) are input with the respective information of the corrected transmission image (G3a) and the correction emission image in which the inspection regions (81a, 81b) are respectively set.

Therefore, in the image processing system (11), the information of the corrected transmission image (G3a) and the corrected light emission image is input to the same neural network (NN1). As a result, the image processing system (11) has a simple configuration as compared with the case of using two neural networks to which the corrected transmission image (G3a) and the corrected light emission image are respectively input.

In the image processing system (11) of the eighth aspect according to the embodiment, in any one of the fourth to seventh aspects, the region setting unit (113b) performs correction by performing a sharpening process. It is preferable to extract an inspection area (81a, 81b) from the transmission image (G3a) and the corrected light emission image. The sharpening process emphasizes an edge extending along the cleavage direction of the solar battery cell (9) in each of the corrected transmission image (G3a) and the corrected light emission image.

The direction of cleavage is the surface direction of the cleavage plane of the solar battery cell (9). Alternatively, the direction of cleavage is the direction in which the cracks of the solar cell (9) extend. Alternatively, the direction of cleavage is one or more edges (two in the present embodiment) of one or more cleavage planes of the solar battery 9 on one surface (the light receiving surface in the present embodiment) of the solar battery cell (9). (Edge) is the direction in which it extends. For example, the direction of cleavage includes at least one direction of 45 degrees, 225 degrees, 135 degrees, and 315 degrees with respect to one direction.

Therefore, the image processing system (11) can accurately detect the edge extending along the cleavage direction of the solar battery cell (9) by the sharpening process, and the extraction accuracy of the inspection area (81a, 81b) is also improved. .

In the image processing system (11) of the ninth aspect according to the embodiment, in the eighth aspect, the cleavage directions of the solar battery cell (9) are first directions (0 degree direction or 180 degrees) orthogonal to each other It is preferable that the direction is the second direction (90 degrees direction or 270 degrees direction).

Therefore, the image processing system (11) can inspect the crack (8) of the solar battery cell (9) using a single crystal material such as single crystal silicon with high accuracy.

In the image processing system (11) according to the tenth aspect, the area setting unit (113b) preferably performs the sharpening process using a wavelet filter in the eighth or ninth aspect. .

Therefore, the image processing system (11) can detect the candidate area (80) with high accuracy.

An inspection system (A1, A2, A3) according to an eleventh aspect of the present invention includes an imaging unit (4), a first illumination device (2), a light emission generation device (3), and And an image processing system (11) according to any one of the ten aspects. An imaging unit (4) images a solar cell (9). The first lighting device (2) applies illumination light (Lt1) to the solar battery cell (9) to generate transmission light (Lt11) in which the illumination light (Lt1) is transmitted through the solar battery cell (9). The light emission generation device (3) (the second lighting device 3A or the power supply device 3B) generates the light emission (Lt21 or Lt31) of the solar battery cell (9). Then, the imaging unit (4) receives the transmitted light (Lt11), outputs the information of the transmitted image to the image processing system (11), receives the light emission, and receives the information of the light emission image to the image processing system (11). Output.

Accordingly, the inspection system (A1, A2, A3) can accurately determine the crack (8) of the solar cell (9) based on the captured image of the solar cell (9). The inspection system (A1, A2, A3) can accurately distinguish the crack (8) from other elements (noises) such as etching unevenness, etching patterns, abrasion marks on the surface film, dirt, and the like.

In the inspection system (A1, A2) of the twelfth aspect according to the embodiment, in the eleventh aspect, the light emission generation device (3) irradiates the solar battery cell (9) with excitation light (Lt2). It is preferable that it is a 2nd illuminating device (3A) which produces light emission (Lt21) of a photovoltaic cell by photoluminescence. The luminescence image is a PL luminescence image.

Accordingly, the inspection system (A1, A2) can accurately determine the crack (8) of the solar cell (9) based on the captured image of the solar cell (9). And an inspection system (A1, A2) can distinguish a crack (8) and other elements with sufficient accuracy.

In the inspection system (A1) according to the thirteenth aspect, the imaging unit (4) includes the first imaging device (41), the filter (42a), and the second imaging device according to the twelfth aspect. It is preferable to have (42). The first imaging device (41) receives the transmitted light (Lt11) and outputs information of the transmitted image to the image processing system (11). The filter (42 a) passes the light emission (Lt 21) and attenuates the excitation light (Lt 2). The second imaging device (42) receives the light emission (Lt21) through the filter (42a) and outputs the information of the PL light emission image to the image processing system (11).

Therefore, since the inspection system (A1) can be provided with the first imaging device (41) suitable for capturing a transmission image and the second imaging device (42) suitable for capturing a PL light emission image, the inspection system (A1) can achieve high image quality. A captured image can be generated.

In the inspection system (A2) according to the fourteenth aspect of the present invention, in the twelfth aspect, the imaging unit (4) includes one imaging device (42), transmitted light (Lt11), and light emission (Lt21). And a filter (42a) for attenuating the excitation light (Lt2). The imaging device (42) receives the transmitted light (Lt11) through the filter (42a), outputs information of the transmitted image to the image processing system (11), and emits light (Lt21) through the filter (42a). The light is received and information of the PL emission image is output to the image processing system (11).

Therefore, the inspection system (A2) can achieve cost reduction and simplification of the system configuration.

Moreover, in the inspection system (A3) according to the fifteenth aspect of the present invention, in the eleventh aspect, the light emission generating device (3) supplies power to the solar battery cell (9), and the solar battery is electroluminescent. It is preferable that it is a power supply device (3B) which produces light emission (Lt31) of a cell (9). The luminescence image is an EL luminescence image.

Therefore, the inspection system (A3) can accurately determine the crack (8) of the solar battery cell (9) based on the captured image of the solar battery cell (9). And an inspection system (A3) can distinguish a crack (8) and other elements with sufficient accuracy.

Moreover, the image processing method of the 16th aspect which concerns on embodiment is an image processing method for test | inspecting the crack (8) of a photovoltaic cell (9) based on the image of a photovoltaic cell (9). The image processing method includes an acquisition step (S1) and an image processing step (S2). An acquisition step (S1) acquires the information of the transmitted image (grayscale image G1) which is an image of the photovoltaic cell (9) by the transmitted light (Lt11) which permeate | transmitted the photovoltaic cell (9). Moreover, an acquisition step (S1) acquires the information of the light emission image (grayscale image G1) which is an image of the photovoltaic cell (9) by light emission (Lt21 or Lt31) of a photovoltaic cell (9). The image processing step (S2) receives the information of the transmission image and the respective information of the light emission image and performs image processing to determine whether or not a crack (8) is generated in the solar battery cell (9).

Therefore, the above-mentioned image processing method can discriminate crack (8) of a photovoltaic cell (9) accurately based on an image pick-up picture of a photovoltaic cell (9). Then, the image processing method can accurately distinguish the crack (8) from other elements (noises) such as etching unevenness, etching patterns, abrasion marks of the surface film, dirt and the like.

The program according to the seventeenth aspect relates to the computer system (1) to execute the image processing method according to the sixteenth aspect.

Therefore, the above-mentioned program can judge crack (8) of a photovoltaic cell (9) accurately based on an image pick-up picture of a photovoltaic cell (9). Then, the program can accurately distinguish the crack (8) from other elements (noises) such as etching unevenness, etching patterns, abrasion marks of the surface film, dirt, and the like.

The above-described embodiment is an example of the present disclosure. Therefore, the present disclosure is not limited to the above-described embodiment, and various other embodiments may be used according to the design and the like without departing from the technical concept of the present disclosure. Of course it is possible to change.

DESCRIPTION OF SYMBOLS 1 Computer system 11 Image processing system 111 1st acquisition part 112 2nd acquisition part 113 Image processing apparatus 113a Position correction part 113b Area setting part 113c Inspection part 2 1st illuminating device 3 luminescence generator 3A 2nd illuminating device 3B power supply 4 Imaging unit 41 first imaging device 42 second imaging device (imaging device)
42a filter 71 cell area 8

crack

81a, 81b inspection area 9 photovoltaic cell A1, A2, A3 inspection system Lt11 transmitted light Lt2 excitation light Lt21, Lt31 emission G1 gray-scale image (transmission image, emission image, PL emission image. EL emission image )
G3a corrected transmission image G3b corrected PL emission image NN1 neural network S1 acquisition step S2 image processing step

Claims

An image processing system for inspecting a crack of the solar battery cell based on an image of the solar battery cell,
A first acquisition unit configured to acquire information of a transmission image which is an image of the solar battery cell by transmitted light transmitted through the solar battery cell;
A second acquisition unit that acquires information of a light emission image that is an image of the solar battery cell by light emission of the solar battery cell;
An image processing apparatus that receives the information of the transmission image and each information of the light emission image and performs image processing to determine whether the crack is generated in the solar battery cell; Image processing system.
The image processing system according to claim 1, wherein the light emission image is a PL light emission image obtained by capturing the solar battery cell emitting light by irradiation of excitation light.
The image processing system according to claim 1, wherein the light emission image is an EL light emission image obtained by capturing an image of the solar battery cell which is emitting a current by being supplied with an electric current.
Each of the transmission image and the light emission image is a gray-scale image,
The image processing apparatus is
The transmission image and the light emission image are corrected so that each of the cell region in which the solar battery cell is photographed in the transmission image and the cell region in which the solar battery cell is photographed in the light emission image are directed in a predetermined direction. A position correction unit that performs position correction processing to generate a corrected transmission image that is the transmission image subjected to the position correction processing, and a correction light emission image that is the light emission image that is subjected to the position correction processing;
An area setting unit configured to respectively generate information on an inspection area in which the crack is estimated to be captured in the corrected transmission image and the corrected light emission image;
An inspection unit that determines whether the crack is generated in the solar battery cell based on the corrected transmission image and the correction light emission image in which the inspection area is set, respectively. The image processing system according to any one of 1 to 3.
The inspection unit generates the crack in the solar battery cell by executing a recognition algorithm constructed by machine learning on the corrected transmission image and the correction emission image in which the inspection region is set, respectively. 5. The image processing system according to claim 4, wherein the image processing system determines whether or not there is an error.
The machine learning is deep learning,
The image processing system according to claim 5, wherein the inspection unit includes a neural network that executes a recognition algorithm constructed by the deep learning.
7. The image processing system according to claim 6, wherein the neural network has two convolutional layers to which each information of the corrected transmission image and the correction light emission image in which the inspection area is set respectively is input.
The area setting unit performs a sharpening process that emphasizes an edge extending along the cleavage direction of the solar battery cell in each of the corrected transmission image and the correction light emission image, to thereby correct the correction transmission image and the correction light emission. The image processing system according to any one of claims 4 to 7, wherein each of the inspection areas is extracted from an image.
The image processing system according to claim 8, wherein a cleavage direction of the solar battery cell is a first direction and a second direction orthogonal to each other.
10. The image processing system according to claim 8, wherein the area setting unit performs the sharpening process using a wavelet filter.
An imaging unit for imaging a solar battery cell;
A first illumination device that emits illumination light to the solar battery cell to generate transmitted light in which the illumination light is transmitted through the solar battery cell;
A light emission generator for causing light emission of the solar battery cell;
An image processing system according to any one of the preceding claims.
The inspection unit receives the transmitted light, outputs information of a transmitted image to the image processing system, receives the light emission, and outputs information of a light emission image to the image processing system. .
The light emission generation device is a second illumination device that emits excitation light to the solar battery cell to cause light emission of the solar battery cell by photoluminescence,
The inspection system according to claim 11, wherein the luminescence image is a PL luminescence image.
The imaging unit is
A first imaging device that receives the transmitted light and outputs information of the transmitted image to the image processing system;
A filter that passes the emitted light and attenuates the excitation light;
13. The inspection system according to claim 12, further comprising: a second imaging device that receives the light emission through the filter and outputs information of the PL light emission image to the image processing system.
The imaging unit is
One imaging device,
A filter for transmitting the transmitted light and the light emission and for attenuating the excitation light;
The imaging device receives the transmitted light through the filter, outputs information of the transmitted image to the image processing system, receives the light emission through the filter, and transmits the information of the PL emission image. The inspection system according to claim 12, wherein the inspection system outputs the image to an image processing system.
The light emission generation device is a power supply device that supplies power to the solar battery cell to cause light emission of the solar battery cell by electroluminescence,
The inspection system according to claim 11, wherein the luminescence image is an EL luminescence image.
An image processing method for inspecting a crack of the solar battery cell based on an image of the solar battery cell,
Acquisition step of acquiring information of a transmission image which is an image of the solar cell by the transmission light transmitted through the solar cell, and acquiring information of a light emission image which is an image of the solar cell by light emission of the solar cell When,
An image processing step of receiving information on the transmission image and each information on the light emission image and performing image processing to determine whether the crack is generated in the solar battery cell. Image processing method.
A program causing a computer system to execute the image processing method according to claim 16.