Image classification method and device for solar cell

Abstract

The present application relates to a kind of solar cell piece image classification method and device. According to the method of one embodiment of the present application is executed by electronic equipment and based on the EL image (electroluminescence image) of solar cell piece to be detected whether the method for classifying solar cell piece to be detected has defect includes: primary classification step, according to whether black spot (black spot) occupies more than predetermined ratio in the EL image of solar cell piece to be detected, when black spot occupies more than predetermined ratio, it is classified as first type, when the ratio of black spot is less than predetermined ratio, it is classified as second type;And secondary classification step, in the case of being classified as first type in primary classification, the defect classification model learned is used to classify the defect type of solar cell piece to be detected based on the EL image of solar cell piece to be detected.

Description

Technical Field

[0001] This invention relates to an image classification method and apparatus for solar cells, and more specifically, to a method and apparatus for rapidly and accurately classifying the presence and type of defects in solar cells based on EL images (electrolum inescence images). Background Technology

[0002] Solar cells are devices that convert incident sunlight into electrical energy. Because they use sunlight as their energy source, they do not require a fuel supply and are therefore widely used in power supply devices for portable electronic calculators, electronic clocks, outdoor clocks, radio relay stations, communication satellites, and broadcast satellites.

[0003] Previously, the primary method for detecting defects in solar cells was to irradiate them with light and analyze their current-voltage characteristics (hereinafter referred to as "prior art"). However, this prior art can only detect a subset of the various defects closely related to the lifespan of solar cells (e.g., external cracks, internal cracks, brightness degradation, electrode defects, short circuits, junction breakdown, or hot spots). In other words, the problem with prior art is that it cannot detect defects such as internal cracks that occur during the manufacturing process, which are not visible externally.

[0004] Furthermore, because existing technologies are extremely time-consuming to inspect a single solar cell, they are difficult to apply to the majority of solar cells produced in mass production. That is, all photovoltaic cells have their own unique EL images, and existing technologies struggle to handle the analysis of over two million unique EL images per day. Summary of the Invention

[0005] Technical issues

[0006] To address the problems of the prior art, the present invention aims to provide a method and apparatus for rapidly and accurately classifying the presence and type of defects in a solar cell based on an EL image (electroluminescence image).

[0007] However, the problems to be solved by the present invention are not limited to those described above, and those skilled in the art will clearly understand other unmentioned problems from the following description.

[0008] Technical solutions

[0009] A method according to an embodiment of the present invention for solving the above-mentioned problems is a method performed by an electronic device to classify whether a solar cell to be inspected has defects based on an EL image (elec troluminescence image) of the solar cell to be inspected. The method includes: a first classification step, classifying the EL image of the solar cell to be inspected according to whether black spots occupy a predetermined percentage or more, classifying it as a first type when black spots occupy a predetermined percentage or more, and classifying it as a second type when the percentage of black spots occupies less than a predetermined percentage; and a second classification step, in the case that it is classified as a first type in the first classification, classifying the defect type of the solar cell to be inspected based on the EL image of the solar cell to be inspected using a learned defect classification model.

[0010] The defect classification model can be a model that learns using machine learning methods based on learning data. The learning data includes input data about the EL images used for learning and result data about the defect types of solar cells in the EL images used for learning.

[0011] The classification step may include classifying the EL image of the solar cell to be detected using a learned image generation model based on a machine learning method, so as to generate an image (black dot image) containing black dots in the EL image from the learned EL image.

[0012] The image generation model can be a model that learns using learning data, which includes input data about the learning EL image and result data about the black dot image based on image processing of the learning EL image.

[0013] The image processing may include histogram equalization of the learning EL image, bus-barline removal, edge-based perspective transform, and contour extraction.

[0014] The defect classification model can classify defect types that arise from various causes in multiple different manufacturing processes.

[0015] A classification apparatus according to an embodiment of the present invention includes: a memory for storing EL images of a solar cell to be inspected; and a control unit for controlling the analysis of whether the solar cell to be inspected has defects by processing the stored EL images of the solar cell to be inspected. The control unit performs primary and secondary classification. In primary classification, the EL image is classified according to whether black spots occupy a predetermined percentage or more in the EL image of the solar cell to be inspected. If black spots occupy a predetermined percentage or more, it is classified as a first type; if the percentage of black spots is less than the predetermined percentage, it is classified as a second type. In secondary classification, if the solar cell was classified as a first type in primary classification, a learned defect classification model is used to classify the defect type of the solar cell to be inspected based on the EL image of the solar cell to be inspected.

[0016] A classification apparatus according to another embodiment of the present invention includes: a communication unit for storing EL images of a solar cell to be inspected; and a control unit for controlling the analysis of whether the solar cell to be inspected has defects by processing the received EL images of the solar cell to be inspected, wherein the control unit performs primary classification and secondary classification. In primary classification, the EL image is classified according to whether black spots occupy a predetermined percentage or more in the EL image of the solar cell to be inspected. If black spots occupy a predetermined percentage or more, it is classified as a first type; if the percentage of black spots occupies less than the predetermined percentage, it is classified as a second type. In secondary classification, if the solar cell to be inspected is classified as a first type in primary classification, a learned defect classification model is used to classify the defect type of the solar cell to be inspected based on the EL image of the solar cell to be inspected.

[0017] The defect classification model can be a model that learns using machine learning methods based on learning data. The learning data includes input data about the EL images used for learning and result data about the defect types of solar cells in the EL images used for learning.

[0018] The control unit can control the execution of the operating procedure, which can perform the analysis based on the time, date, cell identification code, or EL image of the solar cell to be tested from the production line.

[0019] The effects of the invention

[0020] The advantage of the present invention with the above configuration is that it can quickly and accurately classify whether there are defects in the solar cell and the type of defects based on the EL image of the solar cell.

[0021] Furthermore, the advantage of this invention is that it can be applied to the full inspection of millions of EL images per day, thereby minimizing human error.

[0022] The effects that can be obtained by the present invention are not limited to those described above, and those skilled in the art will clearly understand other unmentioned effects from the following description. Attached Figure Description

[0023] Figure 1 A structural block diagram of a solar cell analysis system 10 according to an embodiment of the present invention is shown.

[0024] Figure 2 An example of a solar cell 400 is shown.

[0025] Figure 3 A structural block diagram of a sorting device 300 according to an embodiment of the present invention is shown.

[0026] Figure 4 A structural block diagram of a control unit 350 in a sorting apparatus 300 according to an embodiment of the present invention is shown.

[0027] Figure 5 A flowchart of an image classification method according to an embodiment of the present invention is shown.

[0028] Figure 6 An example of an operating procedure according to an embodiment of the present invention is shown.

[0029] Figure 7 The images shown represent the individual images used in the process of image processing of EL images to classify them.

[0030] Figure 8 Various examples of EL images and black dot images derived from a single classification of the EL image are shown.

[0031] Figure 9 An example of a defect classification model is shown.

[0032] Figure 10 shows various examples of black dot types in EL images.

[0033] Figure 11 Various examples of classification results using a defect classification model implemented in practice are shown. Detailed Implementation

[0034] The above-mentioned objectives, technical solutions, and effects of the present invention will become clearer from the following detailed description in relation to the accompanying drawings, thereby enabling those skilled in the art to readily implement the technical concept of the present invention. Furthermore, in describing the present invention, detailed descriptions of known technologies related to the present invention will be omitted when it is determined that such detailed descriptions may unnecessarily obscure the main points of the present invention.

[0035] The terminology used in this specification is for describing embodiments and is not intended to limit the invention. In this specification, the singular forms include the plural forms unless otherwise expressly indicated. In this specification, terms such as “comprising,” “possessing,” “prepared,” or “having” do not exclude the presence or addition of one or more other components different from those mentioned.

[0036] In this specification, terms such as “or” or “at least one” may refer to one of the objects listed together, or a combination of two or more. For example, “A or B” and “at least one of A and B” may include only one of A or B, or may include both A and B.

[0037] In this specification, descriptions following "for example" and the like may not perfectly match the information provided regarding the referenced characteristics, variables, or values. Therefore, implementation of the invention according to various embodiments should not be limited to the same effects as variations including other well-known factors such as tolerances, measurement errors, limits and factors of measurement accuracy.

[0038] It should be understood that in this specification, when a component is described as being "connected" or "connected" to another component, it may mean that the connection is direct or directly connected to the other component, but there may be other components in between. Conversely, it should be understood that when a component is described as being "directly connected" or "directly connected" to another component, there are no other components in between.

[0039] It should be understood that in this specification, when a component is described as being "on" or "in contact" with another component, it may be in direct contact or connection with the other component, but there may be another component in between. Conversely, it should be understood that when a component is described as being "directly on" or "directly in contact with" another component, there are no other components in between. Other expressions describing the relationship between components, such as "between with" and "directly between," can be interpreted in the same way.

[0040] In this specification, terms such as "first" and "second" may be used to describe various components, but these components should not be limited by the aforementioned terms. Furthermore, the aforementioned terms are used to distinguish one component from another and should not be construed as limiting the order of the components. For example, "first element" may be called "second element," and similarly, "second element" may be called "first element."

[0041] Unless otherwise defined, all terms used in this specification may be used in the meaning commonly understood by one of ordinary skill in the art to which this invention pertains. Furthermore, unless explicitly defined, terms as defined in common dictionaries will not be interpreted ideally or excessively.

[0042] In the following, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings.

[0043] Figure 1 A structural block diagram of a solar cell analysis system 10 according to an embodiment of the present invention is shown. Figure 2 An example of a solar cell 400 is shown.

[0044] A solar cell analysis system 10 according to an embodiment of the present invention is a system for analyzing and managing whether solar cells manufactured through various processes have defects and the types of defects. Figure 1 As shown, this solar cell analysis system 10 may include an image detection device 100, a database device 200, and a classification device 300. The image detection device 100, the database device 200, and the classification device 300 can transmit / receive data via various wired / wireless communication methods.

[0045] Reference Figure 2 The solar cell 400 includes multiple finger-bars 401 and bus-bars 402. The finger-bars 401 are thin strips with a thin metallization treatment on their front surface. These finger-bars 401 and bus-bars 402 serve as channels for electrons generated by sunlight (photoelectric effect) to move through electromotive force (photovoltaic effect), corresponding to electrodes.

[0046] Multiple fine grids 401 may be spaced apart from each other on the front surface of the solar cell 400. Furthermore, main grids 402 may be configured to extend through the front surface of the solar cell 400 in a horizontal or vertical direction. That is, multiple main grids 402 may extend in a direction intersecting or orthogonal to the extending direction of the fine grids 401 and be spaced apart from each other, connecting to the fine grids 401 formed on the front surface of the solar cell 400. In this case, the multiple fine grids 401 are connected to the main grids 402, and the generated current is collected and transmitted to the main grids 402. The main grids 402 may be formed continuously with a constant width or discontinuously.

[0047] For example, the extremely fine grid 401 can be arranged horizontally, and the main grid 402 can be arranged vertically. Furthermore, in a solar cell module including multiple solar cells 400, the solar cells can be connected to each other through the main grid.

[0048] The image inspection device 100 is an apparatus for detecting images of solar cells that are the objects of inspection. In particular, the image inspection device 100 can detect the electroluminescence (EL) image of the solar cell using an electroluminescence camera. For example, the image inspection device 100 can generate an EL image by capturing the EL light emitted from the solar cell when a forward current or reverse voltage is applied. Furthermore, when power is applied on a per-solar-cell basis, an EL image based on the EL light emitted from each solar cell can be detected.

[0049] Database device 200 stores and manages a database containing various data processed for the operation of solar cell analysis system 10. The database may include data of EL images detected by image detection device 100. That is, database device 200 can receive data of EL images detected by image detection device 100 and store and manage this data in the database. Furthermore, the database may include classification result data of EL images analyzed by classification device 300. That is, database device 200 can receive classification result data from classification device 300 and store and manage this data in the database.

[0050] Figure 3 A structural block diagram of a sorting device 300 according to an embodiment of the present invention is shown.

[0051] The classification device 300 is a device that classifies whether a solar cell to be inspected has defects based on EL image data. It can be an electronic device with computing capabilities or a computing network. In this case, the EL image can be detected by the inspection device 100 or received separately from other devices.

[0052] For example, electronic devices can be desktop personal computers, laptop personal computers, tablet personal computers, netbook computers, workstations, personal digital assistants (PDs), smartphones, smartpads, or mobile phones, but are not limited to these.

[0053] like Figure 3 As shown, this sorting device 300 may include an input unit 310, a communication unit 320, a display 330, a memory 340, and a control unit 350.

[0054] Input unit 310 generates input data in response to input from various users (e.g., supervisors) and may include various input tools. For example, input unit 310 may send input from the user to control unit 350 for various selections made by the user for the execution of an operating procedure. For example, input unit 310 may include, but is not limited to, a keyboard, keypad, dome switch, touch panel, touch key, touch pad, mouse, menu button, etc.

[0055] The communication unit 320 is a component that performs communication with other devices. The communication unit 320 receives EL image data from the image detection device 100, the database device 200, or other devices. Furthermore, the communication unit 320 can send data representing the classification results of the EL images, performed under the control of the control unit 350, to the database device 200 or other devices. For example, the communication unit 320 can perform wireless communication such as 5G, LTE-A, LTE, Bluetooth, Bluetooth Low Energy (BLE), NFC, Wi-Fi, or wired communication such as cable communication, but is not limited to these.

[0056] The display 330 displays various image data on a screen and can be composed of a non-emissive panel or an emissive panel. For example, the display 330 can display EL images, blackspot images, and execution screens for operating programs. The display 330 may include, but is not limited to, a liquid crystal display (LCD), a light-emitting diode (LED) display, an organic light-emitting diode (OLED) display, a microelectromechanical system (MEMS) display, or an electronic paper display. Furthermore, the display 330 can be combined with an input unit 310 and implemented as a touch screen.

[0057] The memory 340 stores various information required for the operation of the classification device 300. For example, the stored information may include EL images, black spot images, image generation models, defect classification models, and program information related to the image classification method described later. Furthermore, the memory 340 may store the aforementioned database. For example, the memory 340 may be of various types, including, but not limited to, hard disk type, magnetic media type, CD-ROM (compact disc read-only memory), optical media type, magneto-optical media type, multimedia card micro type, flash memory type, read-only memory type, or random access memory type. Moreover, depending on its purpose / location, the memory 340 may be a cache, buffer, main memory, or auxiliary memory, or a separately prepared storage system, but is not limited to these.

[0058] The control unit 350 can perform various control operations on the classification device 300. That is, the control unit 350 can control the execution of the image classification method, which will be described later, and can control the operation of other configurations of the classification device 300, namely the input unit 310, communication unit 320, display 330, and memory 340. For example, the control unit 350 may include a processor as hardware or a process as software executing on that processor. The processor may be a processing unit such as a graphics processing unit (GPU), a central processing unit (CPU), a microprocessor unit (MPU), or a microcontroller unit (MCU), but is not limited thereto.

[0059] Figure 4 A structural block diagram of a control unit 350 in a sorting apparatus 300 according to an embodiment of the present invention is shown.

[0060] The control unit 350 controls the execution of an image classification method according to an embodiment of the present invention, such as... Figure 4As shown, it may include a learning unit 351, a primary classification unit 352, and a secondary classification unit 353. For example, the learning unit 351, the primary classification unit 352, and the secondary classification unit 353 may be hardware components of the control unit 350, or they may be software programs executed by the control unit 350, but are not limited thereto.

[0061] The image classification method according to the present invention will now be described in more detail.

[0062] Figure 5 A flowchart of an image classification method according to an embodiment of the present invention is shown.

[0063] An image classification method according to an embodiment of the present invention is a classification method for detecting whether defects exist in a solar cell under test based on EL images, such as... Figure 5 As shown, steps S101 to S104 may be included.

[0064] Figure 6 An example of an operating procedure according to an embodiment of the present invention is shown.

[0065] S101 is the step of selecting the EL image (hereinafter referred to as the "target EL image") of the solar cell to be tested. That is, in S101, as... Figure 6 As shown, the control unit 350 can control the execution of the operating procedure and select the target EL image according to the input commands selected and entered by the user in the executed operating procedure, so as to perform analysis such as classification of the target EL image.

[0066] That is, in addition to the EL image data, the unique identification information (ID) of the solar cell to be inspected in the target EL image, as well as the production date, time, production line, etc. of the solar cell to be inspected, can be stored together in the database of the database device 200 or the memory 340. Accordingly, in S101, the user can select the desired target EL image according to the corresponding time, date, cell ID, or production line through the operating procedure.

[0067] For example, refer to Figure 6 For target EL images, batch analysis can be performed based on the selected date unit (DAILY-SCALE), time unit (SELECT ED-SCALE), or cell ID selection. Furthermore, analysis can be performed according to the selected production line (Select Production Line). Additionally, for target EL image analysis, you can choose between automatic / manual execution over 1 hour (1-hour automatic analysis / 1-hour unit analysis).

[0068] S102 is a step of classifying the target EL image based on the darker parts of the selected target EL image, specifically, based on whether black spots occupy a predetermined percentage or higher. That is, the efficiency of a solar cell is related to the black spot rate; when the black spot rate reaches a predetermined level or higher, the efficiency of the solar cell may deteriorate rapidly. S102 reflects this and first performs the classification step based on the black spot rate.

[0069] In S102, when the black spots account for a predetermined percentage (e.g., more than 1.5%) in the target EL image, the primary classification unit 352 can classify the target EL image into a first type (e.g., black spots are True). Conversely, when the black spots account for a predetermined percentage (e.g., less than 1.5%) in the target EL image, the primary classification unit 352 can classify the target EL image into a second type (e.g., black spots are False).

[0070] Figure 7 The images shown represent the individual images used in the process of image processing of EL images to perform a single classification of the EL images. Figure 8 Various examples of EL images and black dot images derived from a single classification of the EL image are shown.

[0071] In particular, in S102, the primary classification unit 352 can perform a primary classification of the target EL image using an image generation model. That is, the image generation model is learned using a machine learning method based on unsupervised learning to generate an image (black dot image) containing black dots in the EL image from the learned EL image.

[0072] For example, unsupervised learning methods can be auto-encoders, variational auto-encoders, or generative adversarial networks, but are not limited to these.

[0073] For example, an image generation model learns from a pair of input and output data, including multiple layers, and has a function relating the input and output data. In this case, the input data may include data from the learning EL image, and the output data may include black dot image data obtained through image processing of the learning EL image. Therefore, when the input is data from a target EL image, the image generation model can output black dot image data of that target EL image.

[0074] Specifically, refer to Figure 7 To perform image processing on EL images, the following can be performed on the learning EL images: histogram equalization, bus-bar line removal, edge extraction, perspective transform based on extracted edges, contour extraction, etc.

[0075] At this point, histogram equalization is an image processing technique that normalizes the brightness value range of the EL image to a predetermined range. In other words, this is to perform various image processing operations under the same brightness range conditions by scaling the brightness value range of EL images with various ranges according to a predetermined standard.

[0076] Main grid line removal is an image processing step that removes the main grid lines corresponding to the main grid lines in the EL image. Since the main grid lines correspond to areas where no black dots appear, removing the corresponding main grid lines from the EL image enables more accurate black dot detection.

[0077] Edge extraction processing is image processing that extracts edges from an EL image as individual corner points. Because of the various image angles (e.g., the shape of a solar cell in an EL image is not perfectly rectangular and has distorted angles, etc.), this processing is performed to extract edges that will be used as reference points for matching.

[0078] Perspective transformation processing is used to match EL images from various angles to an EL image at a constant angle. It is an image processing technique that projects and transforms EL images based on edges extracted through edge extraction processing. For example, projection transformation processing can project and transform an EL image with a curved solar cell shape into an EL image with a rectangular solar cell shape.

[0079] Contour extraction processing is a method of extracting the contours of various regions existing in an EL image in order to extract black points from an EL image that has already been processed in various ways.

[0080] For example, the image processing of the EL image described above can be performed in the following order: histogram equalization, main grid line removal, edge extraction, projection transformation based on the extracted edges, and contour extraction. However, it is not limited to this, and the image processing can be performed in a different order. However, the contour extraction process can correspond to the last step in the image processing described above.

[0081] On the other hand, images processed according to these image processing methods can be generated separately using models learned through unsupervised learning methods. That is, histogram equalization, main grid line removal, edge extraction, projection transformation, and contour extraction models can be learned and prepared separately. In this case, the image generation model can be referred to as a model that includes all of these models. Alternatively, the image generation model can include multiple layers, each performing these image processing methods sequentially.

[0082] This image generation model can be pre-learned by the learning unit 351 or by other devices and transmitted to the classification device 300 via the communication unit 320.

[0083] In S102, the classification unit 352 can calculate the proportion of black dots generated by the image generation model in the corresponding EL image, and classify the corresponding EL image as either the first type or the second type based on the calculation results.

[0084] In S103, the control unit 350 determines whether the EL image classified in S102 belongs to the first type. If it belongs to the first type, it corresponds to black spots indicating low solar cell efficiency, so S104 is executed to classify the corresponding solar cell's black spot type (i.e., defect type) in more detail. Conversely, if it does not belong to the first type, it corresponds to black spots indicating excellent solar cell efficiency, so there is no need to classify the corresponding solar cell's black spot type in detail. In this case, the process can return to S101 and receive selection input for other target EL images.

[0085] S104 is a step of secondary classification of the defect type (i.e., its black dot type) of the target EL image of the first type. That is, solar cells can be classified into various defect types according to the shape of the black dots in the EL image. S104 is a step of classifying defect types by reflecting these contents.

[0086] In particular, in S104, the secondary classification unit 353 can use the learned defect classification model to perform secondary classification of the defect type of the solar cell to be detected based on the target EL image.

[0087] For example, a defect classification model learns using supervised learning machine learning techniques by learning a pair of input and output data, and includes multiple layers, thus having a function relating the input and output data.

[0088] For example, a defect classification model can be a model trained using deep learning techniques, which may include, but are not limited to, deep neural networks (DNNs), convolutional neural networks (CNNs), recurrent neural networks (RNNs), restricted Boltzmann machines (RBMs), deep belief networks (DBNs), and deep Q-networks.

[0089] Figure 9 An example of a defect classification model is shown.

[0090] Reference Figure 9 The defect classification model can express the relationship between input and output data through multiple hidden layers between the input and output layers. Each hidden layer consists of at least one filter, and each filter has a weight matrix. That is, in the matrix of the corresponding filter, each element (pixel) corresponds to a weight value.

[0091] In other words, the input data for the defect classification model can include data from training EL images. Furthermore, the output data can include defect type data of the solar cell from the training EL images, which are then labeled. When data from the target EL image is input, the defect classification model can output data on the defect type (i.e., black dot type) of the target EL image. In this case, the training EL image can be a first-type image, i.e., an EL image where black dots occupy a predetermined percentage or more.

[0092] Figure 10 shows various examples of black dot types in EL images. Figure 11 Various examples of classification results using a defect classification model implemented in practice are shown.

[0093] Referring to Figure 10, the defect classification model can classify the input target EL image into defect types (i.e., black spot types) that are generated under various manufacturing processes (i.e., solar cell manufacturing processes) that differ from each other due to multiple reasons. For example, black spot types may include, but are not limited to, the following types.

[0094] - Type 1: Types generated by scratches in the ARC Robot Pad

[0095] - Second type: Contamination caused by automation belts in the WB process.

[0096] - Type 3: Type resulting from finger breakage during the printing process.

[0097] - Fourth type: Type resulting from fingerprints left by operators when handling wafers.

[0098] - Fifth type: When classifying existing OpenCV black dot types, the type that reclassifies over-rejection.

[0099] - Type 6: Type arising from uneven film thickness caused by Talox in the MAIA process.

[0100] - Type 7: Types arising from aluminum deficiency in the LCO process.

[0101] - Type 8: Types arising from reflectivity issues in the WB process.

[0102] - Ninth type: A type arising from problems with the wafer raw materials themselves.

[0103] - Type 10: Types arising from null values ​​and hotspots

[0104] - Type 11: Types arising from the lower part of the boat in the POCL process.

[0105] - Type 12: Type arising from damaged wafers on the POCL process boat.

[0106] - Type 13: Types arising from performing the process in the POCL process when the wafer is reversed.

[0107] - Type Fourteen: Types arising from instability in uniformity during the POCL process.

[0108] - Type 15: A type arising from the absence of reverse-oriented wafers on TALOX process boats.

[0109] - Sixteenth type: A type that arises because the predictions of the defect classification model are unstable due to insufficient images of the new type or insufficient learning.

[0110] - Types 17 and 18: Types arising from contamination in WB or POCL processes.

[0111] - Type 19: Types arising from laser pattern align defects in the By-facial process.

[0112] - Type 20: Types of defects caused by contamination due to WB roller scratches.

[0113] - Type 21: Types arising from defects in the pattern (pattern) of the Oxidation and LDSE processes.

[0114] - Type 22: Type arising from defects in the uniformity of the firing process temperature.

[0115] - Type 23: Type arising from defects in the aluminum paste coating process during PRINTER.

[0116] - Type 24: Type arising from contamination of the automated robot pick-up pad.

[0117] - Type 25: Type arising from scratches on the front side of the wafer and WB etching defects.

[0118] On the other hand, defect classification models are image classification models, which have distinctly different characteristics from image augmentation models. For example, image augmentation models can identify the [defect location], [size], and [type] of an EL image of a solar cell module. However, image classification models, due to the inherently complex configuration of their hidden layers, suffer from very high computational cost, time commitment, and storage requirements. Consequently, when determining the [defect type], the conclusions regarding the vector values ​​(i.e., representative values) of the EL image are often ambiguous. In contrast, defect classification models can examine the [defect type] of a solar cell's EL image based on the simple and explicit function of image classification. Therefore, compared to image augmentation models, they have significantly lower computational cost, time commitment, and storage requirements, and offer the advantage of very clear conclusions regarding representative values.

[0119] Furthermore, in S104, the secondary classification unit 353 can control the storage of the data generated after performing the above-mentioned secondary classification in the memory 340 or the database of the database device 200. For example, the result data stored in the database may include the names of the EL images that have undergone primary and secondary classification, the shooting time, and detailed information of the classification results (the black spot ratio in the primary classification, the defect type as a classification result in the secondary classification, etc.).

[0120] The image classification method described above can be executed by loading a program into memory 340 and executing it under the control of control unit 350. This program can be stored in memory 340 on various types of non-transitory computer-readable medium. Non-transitory computer-readable medium includes tangible storage media having various types of entities.

[0121] For example, non-transitory computer-readable media include, but are not limited to, magnetic recording media (e.g., floppy disks, magnetic tapes, hard disk drives), magneto-optical recording media (e.g., magneto-optical disks), read-only memory (CD-ROM), CD-R, CD-R / W, semiconductor memory (e.g., mask ROM, programmable ROM (PROM), erasable PROM (EPROM), flash ROM, random access memory (RAM)).

[0122] Furthermore, the program can be provided by various types of transient computer-readable media. For example, transient computer-readable media may include, but are not limited to, electrical signals, optical signals, and electromagnetic waves. That is, the transient computer-readable media can provide the program to the control unit 350 via a wired communication path such as a power line or optical fiber, or a wireless communication path.

[0123] The advantage of the present invention with the above configuration is that it can quickly and accurately classify whether defects exist in the solar cell and the type of defects based on the EL image of the solar cell. For example, the advantage of the present invention is that it can be applied to the full inspection of more than millions (e.g., 2 million) EL images per day, thereby minimizing human error.

[0124] Specific embodiments have been described in the Summary of the Invention section of this invention; however, various modifications may be made without departing from the scope of the invention. Therefore, the scope of the invention is not limited to the described embodiments, but should be defined by the appended claims and their equivalents.

[0125] Industrial availability

[0126] This invention provides a method and apparatus for rapidly and accurately classifying solar cells based on EL (Elasticity Image) images to determine whether defects exist and the types of defects, thus possessing industrial applicability.

Claims

1. A method performed by an electronic device to classify whether a solar cell to be inspected has defects based on an EL image of the solar cell, the method comprising: In one classification step, the EL image is classified according to whether the black dots occupy a predetermined percentage or more in the EL image of the solar cell to be tested. When the black dots occupy a predetermined percentage or more, it is classified as the first type, and when the black dots occupy a predetermined percentage or less, it is classified as the second type. as well as The secondary classification step involves classifying the solar cell into the first type during the first classification. Then, the learned defect classification model is used to further classify the defect type of the solar cell based on the EL image of the cell under inspection. The defect classification model is an image classification model that learns using machine learning methods based on learning data. It outputs the defect type of the input EL image of the solar cell to be inspected. The learning data includes input data about the learning EL image and result data about the defect type of the solar cell in the learning EL image. The learning EL image is a first type image. The input EL image is also of the first type. The defect classification model categorizes defect types arising from various causes in multiple different manufacturing processes. The classification process includes the following steps: generating black dot images for black dots in the EL image using a learned image generation model based on a machine learning method, so as to generate black dot images containing black dots in the EL image from the learned EL image; calculating the proportion of black dots in the corresponding EL image based on the black dot images generated by the image generation model; and classifying the EL image of the solar cell to be detected into a first type or a second type. The image generation model is trained using learning data to output a black dot image from an input EL image of the solar cell to be detected. The learning data includes input data about the EL image being used for training and result data about the black dot image based on image processing of the EL image. The image generation model uses the EL image of the solar cell to be detected to normalize the brightness value range to a predetermined range, removes the main grid lines from the EL image, extracts the corner points of the EL image, converts the EL image into a solar cell with a preset shape based on the extracted corner points, and processes the contours of the regions present in the EL image.

2. An apparatus, the apparatus comprising: Memory for storing EL images of the solar cell to be tested; as well as The control unit processes stored EL images of the solar cells under test to analyze whether the solar cells are defective. The control unit performs primary and secondary classification. In one classification, the EL image is classified according to whether the black dots occupy a predetermined percentage or more in the EL image of the solar cell to be inspected. If the black dots occupy a predetermined percentage or more, it is classified as type one; and if the percentage is less than the predetermined percentage, it is classified as type two. In the secondary classification, if the solar cell was classified as type 1 in the primary classification, the learned defect classification model is used to classify the defect type of the solar cell based on the EL image of the solar cell to be inspected. The defect classification model is an image classification model that learns using machine learning methods based on learning data. It outputs the defect type of the input EL image of the solar cell to be inspected. The learning data includes input data about the learning EL image and result data about the defect type of the solar cell in the learning EL image. The learning EL image is a first type image. The input EL image is also of the first type. The defect classification model categorizes defect types arising from various causes in multiple different manufacturing processes. The classification process includes the following steps: the control unit uses a learned image generation model based on a machine learning method to generate black dot images for the black dots in the EL image, so as to generate black dot images containing black dots in the EL image from the learned EL image; the proportion of black dots in the corresponding EL image is calculated based on the black dot images generated by the image generation model; and the EL image of the solar cell to be detected is classified into a first type or a second type. The image generation model is trained using learning data to output a black dot image from an input EL image of the solar cell to be detected. The learning data includes input data about the EL image being used for training and result data about the black dot image based on image processing of the EL image. The image generation model uses the EL image of the solar cell to be detected to normalize the brightness value range to a predetermined range, removes the main grid lines from the EL image, extracts the corner points of the EL image, converts the EL image into a solar cell with a preset shape based on the extracted corner points, and processes the contours of the regions present in the EL image.

3. An apparatus, the apparatus comprising: The communication unit stores the EL image of the solar cell to be tested; as well as The control unit processes the received EL images of the solar cells under test to analyze whether the solar cells are defective. The control unit performs primary and secondary classification. In one classification, the EL image is classified according to whether the black dots occupy a predetermined percentage or more in the EL image of the solar cell to be inspected. If the black dots occupy a predetermined percentage or more, it is classified as type one; and if the percentage is less than the predetermined percentage, it is classified as type two. In the secondary classification, if the solar cell was classified as type 1 in the primary classification, the learned defect classification model is used to classify the defect type of the solar cell based on the EL image of the solar cell to be inspected. The defect classification model is an image classification model that learns using machine learning methods based on learning data. It outputs the defect type of the input EL image of the solar cell to be inspected. The learning data includes input data about the learning EL image and result data about the defect type of the solar cell in the learning EL image. The learning EL image is a first type image. The input EL image is also of the first type. The defect classification model categorizes defect types arising from various causes in multiple different manufacturing processes. The classification process includes the following steps: the control unit uses a learned image generation model based on a machine learning method to generate black dot images for the black dots in the EL image, so as to generate black dot images containing black dots in the EL image from the learned EL image; the proportion of black dots in the corresponding EL image is calculated based on the black dot images generated by the image generation model; and the EL image of the solar cell to be detected is classified into a first type or a second type. The image generation model is trained using learning data to output a black dot image from an input EL image of the solar cell to be detected. The learning data includes input data about the EL image being used for training and result data about the black dot image based on image processing of the EL image. The image generation model uses the EL image of the solar cell to be detected to normalize the brightness value range to a predetermined range, removes the main grid lines from the EL image, extracts the corner points of the EL image, converts the EL image into a solar cell with a preset shape based on the extracted corner points, and processes the contours of the regions present in the EL image.

4. The apparatus according to claim 2 or 3, wherein, The control unit controls the execution of the operating procedure. The operating procedure is an apparatus that performs the analysis on the EL image of the solar cell to be tested based on the time, date, cell identification code, or production line.

Images