[go: up one dir, main page]

TWI806201B - Optical inspection system and method - Google Patents

Optical inspection system and method Download PDF

Info

Publication number
TWI806201B
TWI806201B TW110139287A TW110139287A TWI806201B TW I806201 B TWI806201 B TW I806201B TW 110139287 A TW110139287 A TW 110139287A TW 110139287 A TW110139287 A TW 110139287A TW I806201 B TWI806201 B TW I806201B
Authority
TW
Taiwan
Prior art keywords
inspection
product
stations
subsequent
judges
Prior art date
Application number
TW110139287A
Other languages
Chinese (zh)
Other versions
TW202317941A (en
Inventor
蘇家男
楊衍辰
Original Assignee
大陸商業成科技（成都）有限公司
大陸商業成光電（深圳）有限公司
英特盛科技股份有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 大陸商業成科技（成都）有限公司, 大陸商業成光電（深圳）有限公司, 英特盛科技股份有限公司 filed Critical 大陸商業成科技（成都）有限公司
Publication of TW202317941A publication Critical patent/TW202317941A/en
Application granted granted Critical
Publication of TWI806201B publication Critical patent/TWI806201B/en

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
    • G01N21/84—Systems specially adapted for particular applications
    • G01N21/88—Investigating the presence of flaws or contamination
    • G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
    • G01N21/84—Systems specially adapted for particular applications
    • G01N21/88—Investigating the presence of flaws or contamination
    • G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
    • G01N2021/8887—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges based on image processing techniques

Landscapes

  • Biochemistry (AREA)
  • General Health & Medical Sciences (AREA)
  • Signal Processing (AREA)
  • Physics & Mathematics (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Chemical & Material Sciences (AREA)
  • Analytical Chemistry (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Immunology (AREA)
  • Pathology (AREA)
  • Image Analysis (AREA)
  • Investigating Materials By The Use Of Optical Means Adapted For Particular Applications (AREA)
  • Train Traffic Observation, Control, And Security (AREA)
  • Testing Of Optical Devices Or Fibers (AREA)

Abstract

An optical inspection system includes multiple inspection stations and a computer system. The inspection stations capture first inspection images of a first product in an order and determine if the first product complies with a specification requirement to generate inspection results. The computer system trains a machine learning model according to the first inspection images and the inspection results. If one of the inspections station determines that a second produce does not comply with the specification requirement and the computer determines that the second produce comply with the specification requirement in the following inspection stations, then following procedures continues and the second product is not eliminated. If one of the inspection stations determines that the second produce complies with the specification requirement and the computer determines that the second produce does not comply with the specification requirement in the following inspection stations, then the following procedures are stopped and the second product is eliminated.

Description

光學檢測系統與方法Optical detection system and method

本揭露有關於能結合多個檢測站的資訊的光學檢測系統與方法。The present disclosure relates to optical inspection systems and methods that combine information from multiple inspection stations.

一般產品的製造過程中會經過多個製程，為了確保產品的規格符合要求，會設置多個檢測站來檢查產品在各個製程之前或之後是否符合要求。然而，在習知的做法中這些檢測站之間是獨立運作的，如何結合這些檢測站之間的資料，為此領域技術人員所關心的議題。Generally, the manufacturing process of products will go through multiple processes. In order to ensure that the product specifications meet the requirements, multiple inspection stations will be set up to check whether the products meet the requirements before or after each process. However, in the conventional practice, these detection stations operate independently, how to combine the data between these detection stations is a topic of concern to those skilled in the art.

本揭露的實施例提出一種光學檢測系統，包括多個檢測站與電腦系統。每一個檢測站依序對第一產品擷取第一檢測影像並且判斷第一產品是否符合規格要求以產生檢測結果。電腦系統通訊連接至檢測站，用以根據第一檢測影像與檢測結果訓練一機器學習模型。當其中一個檢測站擷取第二產品的第二檢測影像之後，電腦系統根據機器學習模型以及第二檢測影像判斷第二產品在後續的檢測站中是否符合規格要求。如果檢測站判斷第二產品不符合規格要求但電腦系統判斷第二產品在後續的檢測站中符合規格要求，電腦系統繼續後續的製程且不淘汰第二產品。如果檢測站判斷第二產品符合規格要求但電腦系統判斷第二產品在後續的檢測站中不符合規格要求，電腦系統停止後續的製程且淘汰第二產品。Embodiments of the present disclosure provide an optical inspection system, including a plurality of inspection stations and a computer system. Each inspection station sequentially captures a first inspection image of the first product and judges whether the first product meets the specification requirements to generate inspection results. The computer system is communicatively connected to the detection station for training a machine learning model according to the first detection image and detection results. After one of the inspection stations captures the second inspection image of the second product, the computer system judges whether the second product meets the specification requirements in the subsequent inspection station according to the machine learning model and the second inspection image. If the testing station judges that the second product does not meet the specification requirements but the computer system judges that the second product meets the specification requirements at a subsequent testing station, the computer system continues the subsequent manufacturing process without eliminating the second product. If the inspection station judges that the second product meets the specification requirements but the computer system judges that the second product does not meet the specification requirements in subsequent inspection stations, the computer system stops the subsequent manufacturing process and eliminates the second product.

在一些實施例中，檢測站判斷第二產品不符合規格要求但電腦系統判斷第二產品在後續的檢測站中不符合規格要求，電腦系統停止後續的製程且淘汰第二產品。In some embodiments, the inspection station determines that the second product does not meet the specification requirements but the computer system determines that the second product does not meet the specification requirements at subsequent inspection stations, and the computer system stops the subsequent manufacturing process and eliminates the second product.

在一些實施例中，如果檢測站判斷第二產品符合規格要求但電腦系統判斷第二產品在後續的檢測站中符合規格要求，電腦系統繼續後續的製程且不淘汰第二產品。In some embodiments, if the test station determines that the second product meets the specification requirements but the computer system determines that the second product meets the specification requirements at a subsequent test station, the computer system continues the subsequent process without eliminating the second product.

在一些實施例中，上述的機器學習模型為捲積神經網路。In some embodiments, the aforementioned machine learning model is a convolutional neural network.

在一些實施例中，上述的檢測結果包括第一產品在檢測站被檢測到的一瑕疵在第一檢測影像中的位置。電腦系統用以重疊第一檢測影像並根據瑕疵在其餘檢測站中是否依然被判斷為瑕疵來產生訓練資料。In some embodiments, the above-mentioned detection result includes a position of a defect detected by the first product at the detection station in the first detection image. The computer system is used to overlap the first inspection image and generate training data according to whether the defect is still judged as a defect in other inspection stations.

以另一個角度來說，本揭露的實施例提出一種光學檢測方法，適用於多個檢測站。此光學檢測方法包括：透過每一個檢測站依序對第一產品擷取第一檢測影像並且判斷第一產品是否符合規格要求以產生檢測結果；根據第一檢測影像與檢測結果訓練一機器學習模型；當檢測站擷取第二產品的第二檢測影像之後，根據機器學習模型以及第二檢測影像判斷第二產品在後續的檢測站中是否符合規格要求；如果檢測站判斷第二產品不符合規格要求但判斷第二產品在後續的檢測站中符合規格要求，繼續後續的製程且不淘汰第二產品；以及如果檢測站判斷第二產品符合規格要求但判斷第二產品在後續的檢測站中不符合規格要求，停止後續的製程且淘汰第二產品。From another point of view, the embodiments of the present disclosure provide an optical detection method applicable to multiple detection stations. The optical inspection method includes: sequentially capturing first inspection images of the first product through each inspection station and judging whether the first product meets specification requirements to generate inspection results; training a machine learning model according to the first inspection images and inspection results ; After the inspection station captures the second inspection image of the second product, judge whether the second product meets the specification requirements in the subsequent inspection station according to the machine learning model and the second inspection image; if the inspection station judges that the second product does not meet the specification requires but judges that the second product meets the specification requirements in the subsequent testing station, continues the subsequent process without eliminating the second product; Meet the specification requirements, stop the subsequent process and eliminate the second product.

在一些實施例中，上述的光學檢測方法還包括：如果檢測站判斷第二產品不符合規格要求但判斷第二產品在後續的檢測站中不符合規格要求，停止後續的製程且淘汰第二產品。In some embodiments, the above-mentioned optical inspection method further includes: if the inspection station judges that the second product does not meet the specification requirements but determines that the second product does not meet the specification requirements in subsequent inspection stations, stopping the subsequent process and eliminating the second product .

在一些實施例中，上述的光學檢測方法還包括：如果檢測站判斷第二產品符合規格要求但判斷第二產品在後續的檢測站中符合規格要求，繼續後續的製程且不淘汰第二產品。In some embodiments, the above-mentioned optical inspection method further includes: if the inspection station judges that the second product meets the specification requirements but judges that the second product meets the specification requirements in a subsequent inspection station, continue the subsequent process without eliminating the second product.

在一些實施例中，上述的檢測結果包括第一產品在檢測站被檢測到的瑕疵在第一檢測影像中的位置。光學檢測方法還包括：重疊第一檢測影像並根據瑕疵在其餘檢測站中是否依然被判斷為瑕疵來產生訓練資料。In some embodiments, the above-mentioned detection result includes the position of the defect detected by the first product at the detection station in the first detection image. The optical inspection method further includes: superimposing the first inspection image and generating training data according to whether the defect is still judged as a defect in other inspection stations.

透過上述的光學檢測方法與系統，可以結合多個檢測站的資訊，預測瑕疵是否會被放大或是消失，藉此降低過檢率以及漏檢率，提升瑕疵分類檢測率。Through the above-mentioned optical inspection method and system, the information of multiple inspection stations can be combined to predict whether the defect will be enlarged or disappeared, thereby reducing the over-inspection rate and missed-inspection rate, and improving the defect classification and detection rate.

關於本文中所使用之「第一」、「第二」等，並非特別指次序或順位的意思，其僅為了區別以相同技術用語描述的元件或操作。The terms "first", "second" and the like used herein do not specifically refer to a sequence or sequence, but are only used to distinguish elements or operations described with the same technical terms.

圖1是根據一實施例繪示光學檢測系統的示意圖。請參照圖1，光學檢測系統100包括多個檢測站121~123以及一電腦系統140。檢測站121~123可以包含影像感測器以及其他合適的感測器，例如溫度感測器、壓力感測器等，本揭露並不在此限。在這些檢測站121~123的之前或之後，產品110可進行產線上的一或多道製程。這些檢測站121~123依照產線的順序依序對產品110進行檢查，拍攝產品110的檢測影像以判斷產品110是否符合規格要求。舉例來說，檢測站121擷取了檢測影像131，檢測站122擷取了檢測影像132，而檢測站123擷取了檢測影像133。每個檢測站121~123會產生對應的檢測結果，這些檢測結果包含了產品110是否符合規格要求的資訊，也可包含所檢測到的瑕疵在對應檢測影像中的位置資訊。舉例來說，在檢測影像131可發現瑕疵151，在檢測影像132中可發現瑕疵152，而檢測影像133中發現了瑕疵153。也就是說，上述的檢測結果可以包含瑕疵151~153的位置。FIG. 1 is a schematic diagram illustrating an optical detection system according to an embodiment. Please refer to FIG. 1 , the optical inspection system 100 includes a plurality of inspection stations 121 - 123 and a computer system 140 . The detection stations 121 - 123 may include image sensors and other suitable sensors, such as temperature sensors, pressure sensors, etc., and the present disclosure is not limited thereto. Before or after these inspection stations 121 - 123 , the product 110 can undergo one or more processes on the production line. These inspection stations 121-123 inspect the product 110 sequentially according to the production line sequence, and take inspection images of the product 110 to determine whether the product 110 meets the specification requirements. For example, the detection station 121 captures the detection image 131 , the detection station 122 captures the detection image 132 , and the detection station 123 captures the detection image 133 . Each inspection station 121-123 will generate a corresponding inspection result, which includes information on whether the product 110 meets the specification requirements, and may also include location information of the detected defect in the corresponding inspection image. For example, a defect 151 can be found in the inspection image 131 , a defect 152 can be found in the inspection image 132 , and a defect 153 can be found in the inspection image 133 . That is to say, the above detection results may include the positions of the blemishes 151 - 153 .

電腦系統140透過任意有線或無線的方式通訊連接至檢測站121~123。檢測站121~123會將檢測影像131~133以及上述的檢測結果傳送至電腦系統140，電腦系統140據此可以訓練一機器學習模型，此機器學習模型例如為捲積神經網路，在其他實施例中機器學習模型也可以是支持向量機或其他合適的機器學習模型。電腦系統140會重疊這些檢測影像151~153並根據某一個瑕疵在其餘檢測站中是否依然被判斷為瑕疵來產生訓練資料。舉例來說，請參照圖2，在檢測影像131中具有瑕疵151，在檢測影像132中對應的位置上也具有瑕疵152，但是在檢測影像133中對應的位置上並沒有瑕疵，這是因為有一些瑕疵可能會因為後續的製程而消失，例如在產品上的一個刮痕可能因為後續的鍍膜製程而消失。另一方面，在檢測影像133中具有瑕疵153，但在檢測影像131~132中對應的位置上卻沒有瑕疵，這可能是檢測影像151、152中的瑕疵很小(符合規格要求)，但是在後續的製程中此瑕疵被放大了，導致了在後續的製程中產生了不符合規格要求瑕疵153。The computer system 140 is communicatively connected to the testing stations 121-123 through any wired or wireless means. The detection stations 121-123 will transmit the detection images 131-133 and the above-mentioned detection results to the computer system 140, and the computer system 140 can train a machine learning model based on this. This machine learning model is, for example, a convolutional neural network. In the example, the machine learning model may also be a support vector machine or other suitable machine learning models. The computer system 140 will overlap these inspection images 151 - 153 and generate training data according to whether a certain defect is still judged as a defect in other inspection stations. For example, please refer to FIG. 2 , there is a flaw 151 in the detection image 131, and there is a flaw 152 at the corresponding position in the detection image 132, but there is no flaw at the corresponding position in the detection image 133, because there are Some defects may disappear due to the subsequent process, for example, a scratch on the product may disappear due to the subsequent coating process. On the other hand, there is a flaw 153 in the detection image 133, but there is no flaw at the corresponding position in the detection images 131-132, which may be that the flaws in the detection images 151 and 152 are very small (meeting the specification requirements), but in This defect is amplified in the subsequent process, resulting in the defect 153 that does not meet the specification requirement in the subsequent process.

上述的訓練資料包含了機器學習模型的輸入與輸出。電腦系統140可以將某一個檢測站之前(包含目前檢測站)產生的檢測影像當作機器學習模型的輸入，而將後續檢測站產生的檢測結果當作機器學習模型的輸出。機器學習模型可以輸出瑕疵的位置，也可以輸出一個數值來表示有沒有不符合規格要求的瑕疵，或者機器學習模型也可以輸出一張影像，其中每個像素的值可用來表示該像素是否為瑕疵。據此，訓練好的機器學習模型便可以根據先前的檢測影像來預測在後續的檢測站中是否會有不符合規格要求的瑕疵(以及位置)。在一些實施例中，這樣的機器學習模型可以有多個，例如一個機器學習模型是根據前n-1個檢測站的檢測影像來預測第n個檢測站的檢測結果，此n可以是任意的正整數，若總共有N個檢測站，則在這樣的實施例中可以有(N-1)個機器學習模型。作者，機器學習模型可以根據第i個檢測站的檢測影像來預測第j個檢測站的檢測結果，其中i、j為正整數且j大於i，例如在一些實施例中j=i+1。或者，在一些實施例中機器學習模型也可以是遞歸神經網路(recurrent neural network，RNN)，因此可以把檢測影像依序輸入至遞歸神經網路，每次輸入一個檢測影像都可以對應一個輸出(檢測結果)，在這樣的例子中機器學習模型的個數也可以是1。本領域具有通常知識者當可根據上述揭示內容而設計出任意合適的機器學習模型。 The above training data includes the input and output of the machine learning model. The computer system 140 can use the detection images generated by a certain detection station (including the current detection station) as the input of the machine learning model, and use the detection results generated by the subsequent detection stations as the output of the machine learning model. The machine learning model can output the location of the defect, or it can output a value to indicate whether there is a defect that does not meet the specification requirements, or the machine learning model can also output an image, where the value of each pixel can be used to indicate whether the pixel is a defect . Based on this, the trained machine learning model can predict whether there will be defects (and locations) that do not meet the specification requirements in subsequent inspection stations based on previous inspection images. In some embodiments, there may be multiple such machine learning models. For example, a machine learning model predicts the detection result of the nth detection station based on the detection images of the first n-1 detection stations, and this n can be arbitrary. Positive integer, if there are N detection stations in total, there may be (N-1) machine learning models in such an embodiment. The author, the machine learning model can predict the detection result of the j-th detection station based on the detection image of the i-th detection station, where i and j are positive integers and j is greater than i, for example, j=i+1 in some embodiments. Or, in some embodiments, the machine learning model can also be a recurrent neural network (RNN), so the detection images can be sequentially input to the recurrent neural network, and each input detection image can correspond to an output (Detection result), in such an example, the number of machine learning models can also be 1. Those skilled in the art can design any suitable machine learning model based on the above disclosure.

圖3是根據一實施例繪示光學檢測方法的流程圖。請參照圖3，在步驟301中，產品經過一或多個製程以後進入一檢測站。在步驟302中，由目前的檢測站判斷產品是否符合規格要求。在一些實施例中，此規格要求是由客戶所決定，但本揭露並不在此限。如果步驟302的結果為是，則在步驟303中由電腦系統根據訓練好的機器學習模型以及目前檢測站的檢測影像(也可包含先前檢測站的檢測影像)判斷在後續的檢測站中是否符合規格要求。如果步驟303的結果為是，則表示沒有瑕疵，在步驟305中會繼續後續的製程且不淘汰產品。如果步驟303的結果為否，則表示雖然產品目前符合規格要求(瑕疵可能很微小)，但此瑕疵會在後續的製程中被放大，因此在步驟304會停止後續的製程且淘汰該產品。另外，如果步驟302的結果為否，則進行步驟306，此步驟306相同於步驟303。如果步驟306的結果為是，這表示雖然目前的檢測站判斷產品不符合規格要求，但對應的瑕疵可能在後續的製程中消失不見，因此可以進行步驟305。如果步驟306的結果為否，則在步驟307中停止後續的製程且淘汰該產品。值得注意的是，圖3中各步驟可以實作為多個程式碼或是電路，本發明並不在此限。FIG. 3 is a flowchart illustrating an optical detection method according to an embodiment. Please refer to FIG. 3 , in step 301 , the product enters a testing station after passing through one or more processes. In step 302, the current inspection station judges whether the product meets the specification requirements. In some embodiments, the specification requirement is determined by the customer, but the disclosure is not limited thereto. If the result of step 302 is yes, then in step 303, the computer system judges in step 303 whether the following inspection stations meet specification requirements. If the result of step 303 is yes, it means that there is no defect, and in step 305 the subsequent process will continue and the product will not be eliminated. If the result of step 303 is no, it means that although the product currently meets the specification requirements (the defect may be very small), the defect will be amplified in the subsequent process, so the subsequent process will be stopped and the product will be eliminated in step 304 . In addition, if the result of step 302 is no, go to step 306 , which is the same as step 303 . If the result of step 306 is yes, it means that although the current inspection station judges that the product does not meet the specification requirements, the corresponding defect may disappear in the subsequent manufacturing process, so step 305 can be performed. If the result of step 306 is no, then in step 307 the subsequent process is stopped and the product is eliminated. It should be noted that each step in FIG. 3 can be implemented as a plurality of program codes or circuits, and the present invention is not limited thereto.

在上述的方法與系統中，可以結合多個檢測站的資訊以訓練一機器學習模型，此機器學習模型可以用來預測產品在後續檢測站的檢測結果，藉此可以提早淘汰產品，或者是保留符合規格要求的產品，可以降低過檢率以及漏檢率，提升瑕疵分類檢測率。In the above method and system, the information of multiple testing stations can be combined to train a machine learning model. This machine learning model can be used to predict the testing results of products at subsequent testing stations, thereby eliminating products early or keeping them. Products that meet the specification requirements can reduce the over-inspection rate and missed-inspection rate, and increase the rate of defect classification and detection.

雖然本發明已以實施例揭露如上，然其並非用以限定本發明，任何所屬技術領域中具有通常知識者，在不脫離本發明的精神和範圍內，當可作些許的更動與潤飾，故本發明的保護範圍當視後附的申請專利範圍所界定者為準。Although the present invention has been disclosed above with the embodiments, it is not intended to limit the present invention. Anyone with ordinary knowledge in the technical field may make some changes and modifications without departing from the spirit and scope of the present invention. The scope of protection of the present invention should be defined by the scope of the appended patent application.

100:光學檢測系統 110:產品 121~123:檢測站 131~133:檢測影像 140:電腦系統 151~153:瑕疵 301~307:步驟 100: Optical detection system 110: product 121~123: Testing station 131~133: Detection image 140: Computer system 151~153: Defects 301~307: Steps

為讓本發明的上述特徵和優點能更明顯易懂，下文特舉實施例，並配合所附圖式作詳細說明如下。 圖1是根據一實施例繪示光學檢測系統的示意圖。 圖2是根據一實施例繪示重疊多張檢測影像的示意圖。 圖3是根據一實施例繪示光學檢測方法的流程圖。 In order to make the above-mentioned features and advantages of the present invention more comprehensible, the following specific embodiments are described in detail together with the accompanying drawings. FIG. 1 is a schematic diagram illustrating an optical detection system according to an embodiment. FIG. 2 is a schematic diagram illustrating overlapping multiple detection images according to an embodiment. FIG. 3 is a flowchart illustrating an optical detection method according to an embodiment.

100:光學檢測系統 110:產品 121~123:檢測站 131~133:檢測影像 140:電腦系統 151~153:瑕疵 100: Optical detection system 110: product 121~123: Testing station 131~133: Detection image 140: Computer system 151~153: Defects

Claims (10)

一種光學檢測系統，適於一鍍膜製程，包括：多個檢測站，其中每一該些檢測站依序對一第一產品擷取一第一檢測影像並且判斷該第一產品是否符合一規格要求以產生一檢測結果；一電腦系統，通訊連接至該些檢測站，用以根據該些第一檢測影像與該些檢測結果訓練一機器學習模型，其中當該些檢測站的其中之一擷取一第二產品的第二檢測影像之後，該電腦系統根據該機器學習模型以及該第二檢測影像判斷該第二產品在後續的檢測站中是否符合該規格要求，如果該些檢測站的該其中之一判斷該第二產品不符合該規格要求但該電腦系統判斷該第二產品在該後續的檢測站中符合該規格要求，該電腦系統繼續後續的該鍍膜製程且不淘汰該第二產品，如果該些檢測站的該其中之一判斷該第二產品符合該規格要求但該電腦系統判斷該第二產品在該後續的檢測站中不符合該規格要求，該電腦系統停止該後續的該鍍膜製程且淘汰該第二產品；其中，該電腦系統將該至少一個檢測站之前產生的檢測影像或目前檢測站產生的檢測影像當作該機器學習模型的輸入，而將後續檢測站產生的檢測結果當作機器學習模型的輸出。 An optical inspection system suitable for a coating process, comprising: a plurality of inspection stations, wherein each of the inspection stations sequentially captures a first inspection image of a first product and judges whether the first product meets a specification requirement to generate a test result; a computer system, communicatively connected to the test stations, for training a machine learning model according to the first test images and the test results, wherein when one of the test stations captures After the second inspection image of a second product, the computer system judges whether the second product meets the specification requirements in subsequent inspection stations according to the machine learning model and the second inspection image. One judges that the second product does not meet the specification requirements but the computer system judges that the second product meets the specification requirements in the subsequent testing station, the computer system continues the subsequent coating process without eliminating the second product, If one of the testing stations judges that the second product meets the specification but the computer system judges that the second product does not meet the specification at the subsequent testing station, the computer system stops the subsequent coating manufacturing process and eliminate the second product; wherein, the computer system takes the inspection image generated by at least one inspection station before or the inspection image generated by the current inspection station as the input of the machine learning model, and uses the inspection results generated by subsequent inspection stations as the output of the machine learning model. 如請求項1所述之光學檢測系統，其中如果該些檢測站的該其中之一判斷該第二產品不符合該規格要求但該電腦系統判斷該第二產品在該後續的檢測站中不符合該規格要求，該電腦系統停止該後續的該鍍膜製程且淘汰該第二產品。 The optical inspection system as claimed in claim 1, wherein if the one of the inspection stations judges that the second product does not meet the specification but the computer system judges that the second product does not meet the specification at the subsequent inspection station The specification requires that the computer system stop the subsequent coating process and eliminate the second product. 如請求項1所述之光學檢測系統，其中如果該些檢測站的該其中之一判斷該第二產品符合該規格要求但該電腦系統判斷該第二產品在該後續的檢測站中符合該規格要求，該電腦系統繼續該後續的該鍍膜製程且不淘汰該第二產品。 The optical inspection system as claimed in claim 1, wherein if the one of the inspection stations judges that the second product meets the specification but the computer system judges that the second product meets the specification at the subsequent inspection station Requirement, the computer system continues the subsequent coating process without eliminating the second product. 如請求項1所述之光學檢測系統，其中該機器學習模型為捲積神經網路。 The optical inspection system as claimed in claim 1, wherein the machine learning model is a convolutional neural network. 如請求項1所述之光學檢測系統，其中該些檢測結果包括該第一產品在該些檢測站的其中之一被檢測到的一瑕疵在該些第一檢測影像中的位置，該電腦系統用以重疊該些第一檢測影像並根據該瑕疵在其餘檢測站中是否依然被判斷為瑕疵來產生訓練資料。 The optical inspection system as claimed in claim 1, wherein the inspection results include the position of a defect detected by the first product at one of the inspection stations in the first inspection images, the computer system The method is used to overlap the first inspection images and generate training data according to whether the defect is still judged as a defect in other inspection stations. 一種光學檢測方法，適於一鍍膜製程，適用於多個檢測站，該光學檢測方法包括：透過每一該些檢測站依序對一第一產品擷取一第一檢測 影像並且判斷該第一產品是否符合一規格要求以產生一檢測結果；根據該些第一檢測影像與該些檢測結果訓練一機器學習模型；當該些檢測站的其中之一擷取一第二產品的第二檢測影像之後，根據該機器學習模型以及該第二檢測影像判斷該第二產品在後續的檢測站中是否符合該規格要求；如果該些檢測站的該其中之一判斷該第二產品不符合該規格要求但判斷該第二產品在該後續的檢測站中符合該規格要求，繼續後續的該鍍膜製程且不淘汰該第二產品；以及如果該些檢測站的該其中之一判斷該第二產品符合該規格要求但判斷該第二產品在該後續的檢測站中不符合該規格要求，停止該後續的該鍍膜製程且淘汰該第二產品；其中，該電腦系統將該至少一個檢測站之前產生的檢測影像或目前檢測站產生的檢測影像當作該機器學習模型的輸入，而將後續檢測站產生的檢測結果當作機器學習模型的輸出。 An optical inspection method, suitable for a coating process, applicable to multiple inspection stations, the optical inspection method includes: sequentially capturing a first product through each of the inspection stations for a first inspection image and determine whether the first product meets a specification requirement to generate a test result; train a machine learning model based on the first test images and the test results; when one of the test stations captures a second After the second inspection image of the product, according to the machine learning model and the second inspection image, it is judged whether the second product meets the specification requirements in the subsequent inspection station; if one of the inspection stations judges that the second product The product does not meet the specification requirements but it is judged that the second product meets the specification requirements in the subsequent inspection station, continue the subsequent coating process without eliminating the second product; and if one of the inspection stations judges The second product conforms to the specification but judges that the second product does not meet the specification in the subsequent inspection station, stops the subsequent coating process and eliminates the second product; wherein the computer system uses at least one The detection image generated by the previous detection station or the detection image generated by the current detection station is used as the input of the machine learning model, and the detection results generated by the subsequent detection station are used as the output of the machine learning model. 如請求項6所述之光學檢測方法，還包括：如果該些檢測站的該其中之一判斷該第二產品不符合該規格要求但判斷該第二產品在該後續的檢測站中不符合該規格要求，停止該後續的該鍍膜製程且淘汰該第二產品。 The optical inspection method as described in claim 6, further comprising: if one of the inspection stations judges that the second product does not meet the specification requirements but judges that the second product does not meet the specification in the subsequent inspection station According to the specification, the subsequent coating process is stopped and the second product is eliminated. 如請求項6所述之光學檢測方法，還包括：如果該些檢測站的該其中之一判斷該第二產品符合該規格要求但判斷該第二產品在該後續的檢測站中符合該規格要求，繼續該後續的該鍍膜製程且不淘汰該第二產品。 The optical inspection method as described in claim 6, further comprising: if one of the inspection stations judges that the second product meets the specification requirements but judges that the second product meets the specification requirements in the subsequent inspection station , continue the subsequent coating process without eliminating the second product. 如請求項6所述之光學檢測方法，其中該機器學習模型為捲積神經網路。 The optical inspection method as claimed in claim 6, wherein the machine learning model is a convolutional neural network. 如請求項6所述之光學檢測方法，其中該些檢測結果包括該第一產品在該些檢測站的其中之一被檢測到的一瑕疵在該些第一檢測影像中的位置，該光學檢測方法還包括：重疊該些第一檢測影像並根據該瑕疵在其餘檢測站中是否依然被判斷為瑕疵來產生訓練資料。 The optical inspection method as described in claim 6, wherein the inspection results include the position of a defect detected by the first product at one of the inspection stations in the first inspection images, the optical inspection The method further includes: overlapping the first inspection images and generating training data according to whether the defect is still judged as a defect in other inspection stations.
TW110139287A 2021-10-19 2021-10-22 Optical inspection system and method TWI806201B (en)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202111216747.7A CN114002225B (en) 2021-10-19 2021-10-19 Optical detection system and method
CN202111216747.7 2021-10-19

Publications (2)

Publication Number Publication Date
TW202317941A TW202317941A (en) 2023-05-01
TWI806201B true TWI806201B (en) 2023-06-21

Family

ID=79923231

Family Applications (1)

Application Number Title Priority Date Filing Date
TW110139287A TWI806201B (en) 2021-10-19 2021-10-22 Optical inspection system and method

Country Status (2)

Country Link
CN (1) CN114002225B (en)
TW (1) TWI806201B (en)

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180293722A1 (en) * 2017-04-10 2018-10-11 Dpix, Llc Manufacturing Quality Improvement Through Statistical Root Cause Analysis Using Convolution Neural Networks
TWI667575B (en) * 2018-06-29 2019-08-01 由田新技股份有限公司 Defect inspection system and method using artificil intelligence
TW202105317A (en) * 2019-07-23 2021-02-01 緯創資通股份有限公司 Image recognition apparatus, image recognition method, and computer program product thereof
US20210142456A1 (en) * 2019-11-12 2021-05-13 Bright Machines, Inc. Image Analysis System for Testing in Manufacturing

Family Cites Families (16)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2000206053A (en) * 1999-01-18 2000-07-28 Nkk Corp Surface defect inspection method using multiple surface defect meters
JP4453503B2 (en) * 2004-09-28 2010-04-21 オムロン株式会社 Substrate inspection device, substrate inspection method, inspection logic generation device and inspection logic generation method for substrate inspection device
JP5299046B2 (en) * 2009-04-16 2013-09-25 新日鐵住金株式会社 Wrinkle detection device, wrinkle detection method and program
TW201413258A (en) * 2012-09-19 2014-04-01 Shen Cheng Technology Co Ltd Central control type automatic recognition and marking method of defective articles
TWI544213B (en) * 2014-03-04 2016-08-01 All Ring Tech Co Ltd Object detection method and device
TWI531787B (en) * 2015-01-20 2016-05-01 An automatic optical detection method and an automatic optical detection system for carrying out the method
US10783629B2 (en) * 2017-09-29 2020-09-22 Align Technology, Inc. Aligner image based quality control system
CN108362702A (en) * 2017-12-14 2018-08-03 北京木业邦科技有限公司 A kind of defect of veneer detection method, system and equipment based on artificial intelligence
CN111492401B (en) * 2017-12-19 2022-04-05 利乐拉瓦尔集团及财务有限公司 Method for defect detection in packaging containers
TWI653605B (en) * 2017-12-25 2019-03-11 Utechzone Co., Ltd. Automatic optical detection method, device, computer program, computer readable recording medium and deep learning system using deep learning
IT201800006680A1 (en) * 2018-06-26 2019-12-26 METHOD FOR PREDICTING THE PRESENCE OF PRODUCT DEFECTS DURING AN INTERMEDIATE PROCESSING PHASE OF A THIN PRODUCT WRAPPED IN COIL
CN110335270B (en) * 2019-07-09 2022-09-13 华北电力大学（保定） Power transmission line defect detection method based on hierarchical regional feature fusion learning
TWI707299B (en) * 2019-10-18 2020-10-11 汎思數據股份有限公司 Optical inspection secondary image classification method
TWI722861B (en) * 2020-04-08 2021-03-21 晶碩光學股份有限公司 Classification method and a classification system
TWM604031U (en) * 2020-07-03 2020-11-11 兆米智慧檢測股份有限公司 Artificial intelligence-based optical inspection system
CN112272293A (en) * 2020-10-28 2021-01-26 业成科技（成都）有限公司 Image processing method

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180293722A1 (en) * 2017-04-10 2018-10-11 Dpix, Llc Manufacturing Quality Improvement Through Statistical Root Cause Analysis Using Convolution Neural Networks
TWI667575B (en) * 2018-06-29 2019-08-01 由田新技股份有限公司 Defect inspection system and method using artificil intelligence
TW202105317A (en) * 2019-07-23 2021-02-01 緯創資通股份有限公司 Image recognition apparatus, image recognition method, and computer program product thereof
US20210142456A1 (en) * 2019-11-12 2021-05-13 Bright Machines, Inc. Image Analysis System for Testing in Manufacturing

Also Published As

Publication number Publication date
TW202317941A (en) 2023-05-01
CN114002225A (en) 2022-02-01
CN114002225B (en) 2023-05-12

Similar Documents

Publication Publication Date Title
TWI689875B (en) Defect inspection and classification apparatus and training apparatus using deep learning system
Putera et al. Printed circuit board defect detection using mathematical morphology and MATLAB image processing tools
CN109060817B (en) Artificial intelligence review system and method
JP7426261B2 (en) Learning device, image inspection device, learned parameters, learning method, and image inspection method
CN110658198A (en) Optical detection method, optical detection device and optical detection system
KR20210038143A (en) System and method of inspecting object based on artificial intelligence
TWI826207B (en) Pcb defect detection system and pcb defect detection method
CN108242054A (en) A steel plate defect detection method, device, equipment and server
CN102522350B (en) Method and device for detecting failed production machine platforms
CN119762891B (en) Industrial image anomaly detection method based on multimodal large model
CN113077430B (en) Laser Chip Defect Detection and Classification Method and System Based on SSD Algorithm
CN108334902A (en) A kind of track train equipment room smog fireproof monitoring method based on deep learning
KR20190111644A (en) Apparatus and method of classifying defect type of repetition pattern image
CN118822996B (en) Prefabricated bridge welding quality inspection method based on MIU-Transformer
TWI806201B (en) Optical inspection system and method
JP2020077158A (en) Image processing device and image processing method
CN114120109B (en) Belt longitudinal tear detection method based on neural network
TW202117664A (en) Optical inspection secondary image classification method which can effectively improve the accuracy of image recognition and classification
JP2021105758A (en) Article inspection device
CN110263630A (en) A kind of aluminium flaw identification equipment of the convolutional neural networks based on Inception model
CN116977241A (en) Method, apparatus, computer readable storage medium and computer program product for detecting defects in a vehicle component
CN115953385A (en) An interactive industrial cold start defect detection system and method
TWI742710B (en) Method and system for improving machine learning to use data for training
CN108763063B (en) Software defect detection method without defect labeling data
CN117740960B (en) TOFD spectrum feature detection method