WO2020007096A1 - 显示屏质量检测方法、装置、电子设备及存储介质 - Google Patents

显示屏质量检测方法、装置、电子设备及存储介质 Download PDF

Info

Publication number
WO2020007096A1
WO2020007096A1 PCT/CN2019/083164 CN2019083164W WO2020007096A1 WO 2020007096 A1 WO2020007096 A1 WO 2020007096A1 CN 2019083164 W CN2019083164 W CN 2019083164W WO 2020007096 A1 WO2020007096 A1 WO 2020007096A1
Authority
WO
WIPO (PCT)
Prior art keywords
display screen
defect
defect detection
loss
image
Prior art date
Legal status (The legal status 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 status listed.)
Ceased
Application number
PCT/CN2019/083164
Other languages
English (en)
French (fr)
Inventor
文亚伟
冷家冰
刘明浩
徐玉林
郭江亮
李旭
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing Baidu Netcom Science and Technology Co Ltd
Original Assignee
Beijing Baidu Netcom Science and Technology Co Ltd
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 Beijing Baidu Netcom Science and Technology Co Ltd filed Critical Beijing Baidu Netcom Science and Technology Co Ltd
Priority to KR1020197034323A priority Critical patent/KR102229594B1/ko
Priority to JP2019563779A priority patent/JP6921241B2/ja
Publication of WO2020007096A1 publication Critical patent/WO2020007096A1/zh
Priority to US16/939,277 priority patent/US11488294B2/en
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

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
    • G01N21/95—Investigating the presence of flaws or contamination characterised by the material or shape of the object to be examined
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G06T7/0004—Industrial image inspection
    • G06T7/0008—Industrial image inspection checking presence/absence
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G06T7/0004—Industrial image inspection
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • 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
    • G02—OPTICS
    • G02F—OPTICAL DEVICES OR ARRANGEMENTS FOR THE CONTROL OF LIGHT BY MODIFICATION OF THE OPTICAL PROPERTIES OF THE MEDIA OF THE ELEMENTS INVOLVED THEREIN; NON-LINEAR OPTICS; FREQUENCY-CHANGING OF LIGHT; OPTICAL LOGIC ELEMENTS; OPTICAL ANALOGUE/DIGITAL CONVERTERS
    • G02F1/00—Devices or arrangements for the control of the intensity, colour, phase, polarisation or direction of light arriving from an independent light source, e.g. switching, gating or modulating; Non-linear optics
    • G02F1/01—Devices or arrangements for the control of the intensity, colour, phase, polarisation or direction of light arriving from an independent light source, e.g. switching, gating or modulating; Non-linear optics for the control of the intensity, phase, polarisation or colour 
    • G02F1/13—Devices or arrangements for the control of the intensity, colour, phase, polarisation or direction of light arriving from an independent light source, e.g. switching, gating or modulating; Non-linear optics for the control of the intensity, phase, polarisation or colour  based on liquid crystals, e.g. single liquid crystal display cells
    • G02F1/1306—Details
    • G02F1/1309—Repairing; Testing
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/09—Supervised learning
    • H—ELECTRICITY
    • H10—SEMICONDUCTOR DEVICES; ELECTRIC SOLID-STATE DEVICES NOT OTHERWISE PROVIDED FOR
    • H10K—ORGANIC ELECTRIC SOLID-STATE DEVICES
    • H10K71/00—Manufacture or treatment specially adapted for the organic devices covered by this subclass
    • H10K71/70—Testing, e.g. accelerated lifetime tests
    • 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
    • G01N21/95—Investigating the presence of flaws or contamination characterised by the material or shape of the object to be examined
    • G01N2021/9513—Liquid crystal panels
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30108—Industrial image inspection
    • G06T2207/30121—CRT, LCD or plasma display

Definitions

  • the present application relates to the field of computer technology, and in particular, to a method, a device, an electronic device, and a storage medium for detecting the quality of a display screen.
  • Display screens are also widely used because of their small size, light weight, low power, high resolution, high brightness, and no geometric distortion.
  • display defects such as point defects, line defects, and surface defects may exist in the produced display screens due to various reasons. Therefore, the display quality inspection is an important part of the production process.
  • the display quality inspection mainly uses manual inspection or machine-assisted manual inspection methods.
  • the manual detection method refers to relying on the naked eyes of industry experts to observe the pictures collected from the production environment to give a judgment;
  • the machine-assisted manual detection method refers to the use of a quality inspection system that has solidified the experience of industry experts to perform the inspection of the display screen image. Inspection, followed by industry experts to detect and judge the suspected pictures.
  • the present application provides a method, a device, an electronic device, and a storage medium for detecting the quality of a display screen, in order to overcome the existing subjective influencing factors of the display screen defect detection method, which results in low detection accuracy, poor system performance, and service.
  • a first aspect of the present application provides a display screen quality detection method, including:
  • a defect on the display screen corresponding to the display screen image, a defect type corresponding to the defect, and a position corresponding to the defect are determined according to the defect detection result.
  • a defect detection model obtained by training a historical defect display screen image based on a deep convolutional neural network structure and an object detection algorithm is used to perform defect detection on the display screen image to determine a defect on the display screen corresponding to the display screen image.
  • the defect type corresponding to the defect, and the position corresponding to the defect are used to perform defect detection on the display screen image to determine a defect on the display screen corresponding to the display screen image.
  • the defect detection results obtained based on the defect detection model have high accuracy, strong intelligence, high system performance, and high business scalability.
  • the defect detection model is obtained by training a historical defect display image using a deep convolutional neural network structure and an object detection algorithm, including:
  • the defect detection model is a combination training of candidate region loss, region category loss, and region boundary loss of the historical defect display screen image, so that the candidate region loss, the region category loss, and the region boundary loss are combined.
  • the candidate area loss refers to a loss value between a selected defect area and an actual defect area in the historical defect display screen image
  • the area category loss refers to a predicted defect category and an actual defect category in the selected defect area.
  • the value of the loss between the regions is the loss between the predicted defect boundary and the actual defect boundary in the selected defect region.
  • the defect detection model obtained through combined training of candidate region loss, region category loss, and region boundary loss of the display screen image has high classification accuracy, which improves the performance of the detection system.
  • the method before the inputting the display image into a defect detection model to obtain a defect detection result, the method further includes:
  • image preprocessing on the display screen image, wherein the image preprocessing includes one or more of the following processing:
  • the step of inputting the display screen image into a defect detection model to obtain a defect detection result includes:
  • the display screen image is input to the defect detection model running on the detection model server to obtain a defect detection result.
  • image preprocessing is performed on the obtained display screen image so that the processed display screen image meets the detection standard, which lays a foundation for defect detection in subsequent display screen images and improves the accuracy of the display screen quality detection. degree.
  • the method further includes:
  • the production stage information is used to indicate a manufacturer, a production environment, and the display corresponding to the display screen Screen type.
  • the quality of the display screens is determined by combining the production stage information of each display screen, which can improve the quality of the display screens. Get the accuracy of your display quality.
  • the method further includes:
  • the display screen image and the defect detection result are input into the defect detection model in order to optimize the defect detection model.
  • the accuracy of the defect detection and defect positioning can be reviewed manually by using information in the production database, and then the training database is updated to retrain the defect detection Model to improve the accuracy of defect detection.
  • a second aspect of the present application provides a display screen quality detection device, which is characterized by including:
  • a receiving module configured to receive a quality inspection request sent by a console deployed on a display screen production line, where the quality inspection request includes a display screen image collected by an image acquisition device on the display screen production line;
  • a processing module for inputting the display screen image into a defect detection model to obtain a defect detection result, wherein the defect detection model is obtained by training a historical defect display screen image using a deep convolutional neural network structure and an object detection algorithm of;
  • a determining module configured to determine a defect on a display screen corresponding to the display screen image, a defect type corresponding to the defect, and a position corresponding to the defect according to the defect detection result.
  • the technical effect of the display screen quality detection device provided in the present application may be specifically the same as the technical effect of the foregoing display screen quality detection method.
  • the defect detection model is obtained by training a historical defect display image using a deep convolutional neural network structure and an instance segmentation algorithm, including:
  • the defect detection model is a combination training of candidate area loss, area category loss, and area boundary loss of the historical defect display screen image, so that the candidate area loss, the area category loss, and the area boundary loss are combined.
  • the candidate area loss refers to a loss value between a selected defect area and an actual defect area in the historical defect display screen image
  • the area category loss refers to a predicted defect category and an actual defect category in the selected defect area.
  • the value of the loss between the regions is the loss between the predicted defect boundary and the actual defect boundary in the selected defect region.
  • the processing module is further configured to perform image preprocessing on the display screen image before inputting the display image into a defect detection model to obtain a defect detection result;
  • the image preprocessing includes one or more of the following processes:
  • the processing module is specifically configured to determine a detection model server carrying processing resources according to a load balancing policy, and input the display screen image to the defect detection model running on the detection model server to obtain Defect inspection results.
  • the determining module is further configured to determine the quality of the display screen corresponding to the display screen image according to the production stage information and the defect detection result; the production stage information is used to indicate the display screen The corresponding manufacturer, production environment, and type of the display screen.
  • the processing module is further configured to, after the determining module determines the quality of the display screen corresponding to the display screen image according to the defect detection result, if it is determined that the display screen is a bad screen, then Do one or more of the following:
  • the display screen image and the defect detection result are input into the defect detection model in order to optimize the defect detection model.
  • a third aspect of the present application provides an electronic device including a processor, a memory, and a computer program stored on the memory and executable on the processor.
  • the processor executes the program, the first aspect as described above and The method according to any one of the various possible implementations of the first aspect.
  • a fourth aspect of the present application provides a storage medium, where the storage medium stores instructions, and when it runs on a computer, causes the computer to execute as described in the first aspect and any one of the various possible implementation manners of the first aspect Methods.
  • the present application provides a display screen quality detection method, device, electronic device and storage medium.
  • the method includes: receiving a quality detection request sent by a console deployed on a display screen production line, and the quality detection request includes a display screen production line
  • the display screen image collected by the image acquisition device is input to the defect detection model to obtain the defect detection result.
  • the defect detection model is obtained by training the historical defect display screen image using a deep convolution neural network structure and an object detection algorithm. According to the defect detection result, the defect on the display screen corresponding to the display screen image, the defect type corresponding to the defect, and the position corresponding to the defect are determined.
  • the technical scheme has high defect detection accuracy, good system performance, and high service expansion capability.
  • FIG. 1 is a schematic structural diagram of a display screen quality detection system according to an embodiment of the present application.
  • FIG. 2 is a schematic flowchart of a first embodiment of a display screen quality detection method according to an embodiment of the present application
  • FIG. 3 is a schematic flowchart of a second embodiment of a display screen quality detection method according to an embodiment of the present application.
  • FIG. 4 is a schematic flowchart of a third embodiment of a display screen quality detection method according to an embodiment of the present application.
  • FIG. 5 is a schematic flowchart of a third embodiment of a display screen quality detection method according to an embodiment of the present application.
  • FIG. 6 is a first schematic structural diagram of an embodiment of a display screen quality detection device according to an embodiment of the present application.
  • FIG. 7 is a second schematic structural diagram of an embodiment of a display screen quality detection device according to an embodiment of the present application.
  • FIG. 1 is a schematic structural diagram of a display screen quality detection system provided in an embodiment of the present application.
  • the display screen quality inspection system includes a plurality of different devices, such as an image acquisition device 11, a console 12, a server group 13, a controller 14, a database 15, and a trainer 16, which are deployed on a display screen production line. .
  • the image acquisition device 11 collects the display screen image on the display screen production line
  • the console 12 receives the display screen image collected by the image acquisition device 11 and sends the display screen image to the detection model server 130 in the server group 13 to detect the model.
  • the server 130 inputs the received display screen image into the defect detection model that it runs to obtain the defect detection result.
  • the controller 14 receives the defect detection result of the detection model server 130 and gives a business response based on the production stage information.
  • the controller 14 also
  • the defect detection result may be stored in the database 15 as a log.
  • the display screen image collected by the image acquisition device 11 can also be directly stored in the database 15 as raw data for training the defect detection model.
  • the trainer 16 extracts the historical defect display image in the database, it uses a deep convolutional neural network structure and an instance segmentation algorithm to train the historical defect display image to obtain a defect detection model.
  • the above database 15 may include a production database 151 and a training database 152.
  • the production database 151 may receive and save the defect detection results sent by the controller 14 and the display screen images collected by the image acquisition device 11.
  • the training database 152 may store data from The historical defect display screen image and the corresponding original display screen image extracted by the production database 151 are used to train the trainer 16 to obtain a defect detection model with high detection accuracy.
  • the trainer 16 in the embodiment of the present application may be a training engine implemented by hardware and / or software functions, which serves as a training tool for a defect detection model.
  • the display screen quality detection system in the embodiment of the present application may further include other physical modules such as a processor, a memory, and the embodiment is not limited thereto.
  • the overall intelligent automation level of the 3C industry refers to the information appliance industry that integrates the application of the three major technology products of computers, communications, and consumer electronics
  • 3C industry refers to the information appliance industry that integrates the application of the three major technology products of computers, communications, and consumer electronics
  • the display industry such as mobile phone screens
  • Most of the detection methods adopted by manufacturers on mobile phone screens can be divided into two types: manual detection methods and machine-assisted manual detection methods.
  • the manual detection method refers to relying on the naked eyes of industry experts to observe the images collected from the production environment for judgment. This method is subject to human subjective influence factors, has lower detection efficiency, and has a greater harm to the human eye.
  • the production workshop of the display screen is generally a dust-free environment, workers need to prepare for cleaning and wear dust-free clothes before entering, which may also adversely affect the health and safety of workers.
  • the machine-assisted manual detection method can also be referred to as the detection method based on the liquid crystal module detection equipment.
  • the specific principle is: first, the non-defective image is filtered by a quality inspection system with certain judgment ability, and then the industry experts will The image is detected.
  • quality inspection systems are mostly developed for expert systems and feature engineering systems, which means that experts have solidified their experience in the quality inspection system to make them have certain automation capabilities. Therefore, the machine-assisted manual detection method not only has low accuracy and poor system performance, and cannot cover all the detection standards of the manufacturer. Moreover, this method is also inefficient, and it is easy to miss and misjudge, and it is difficult to reuse the image data after detection. Dig.
  • the characteristics and determination rules are solidified into the machine based on the experience of industry experts, and it is difficult to iterate with the development of the business.
  • the detection accuracy of the quality inspection system is getting more and more Low, and may even drop to a completely unusable state.
  • the characteristics of the quality inspection system are pre-cured in the hardware by third-party suppliers.
  • upgrading not only does the production line need to be significantly modified, but it is also expensive, and it has obvious safety, standardization, and scalability aspects.
  • the shortcomings are not conducive to the optimization and upgrade of the display production line, and the business expansion capability is low.
  • both the manual detection method and the machine-assisted manual detection method have the following disadvantages: Not only are they inefficient and prone to misjudgment, but the industrial data generated by these two methods are not easy to store, manage, and reuse for secondary mining.
  • the embodiment of the present application develops an automatic and high-precision method for detecting the quality of a display screen.
  • the display screen image collected in real time on a display screen production line by using an image acquisition device is used to monitor the display screen in real time. Detect and judge the surface quality of the surface. If it is detected that there is a quality problem in the display screen collected by the current image acquisition device, then determine the position of the quality problem in the picture and its category and category instance.
  • the display screen in this embodiment may include any one of the following: a plasma screen and a liquid crystal display (liquid crystal) display (LCD) screens, light emitting diode (LED) screens, organic light-emitting diode (OLED) screens, and the like.
  • LCD liquid crystal display
  • LED light emitting diode
  • OLED organic light-emitting diode
  • the embodiments of the present application are not limited to the above-mentioned several screens, and may also include other display screens, which are not repeated here.
  • the quality problems described in the embodiments of the present application may include, but are not limited to, defect problems including different types of point defects, line defects, surface defects, and mura. Not one by one here.
  • mura refers to various trace phenomena caused by uneven brightness of the display screen.
  • “multiple” means two or more.
  • “And / or” describes the association relationship between related objects and indicates that there can be three types of relationships. For example, A and / or B can indicate that there are three cases in which A exists alone, A and B exist, and B exists alone.
  • the character "/" generally indicates that the related objects are an "or" relationship.
  • FIG. 2 is a schematic flowchart of a first embodiment of a display screen quality detection method according to an embodiment of the present application. As shown in FIG. 2, in the embodiment of the present application, the display screen quality detection method may include the following steps:
  • Step 21 Receive a quality inspection request sent by a console deployed on a display screen production line, and the quality inspection request includes a display screen image collected by an image acquisition device on the display screen production line.
  • a plurality of devices such as an image acquisition device, a console, a server group, a controller, and a database are deployed on the display screen production line.
  • the image acquisition device can be a high-precision image acquisition camera.
  • multiple displays on the display production line can be collected.
  • the screen image corresponding to the screen.
  • the console deployed on the display screen production line can send a quality inspection request to the server group where the defect detection model is deployed on the display screen production line.
  • the display screen image collected by the image acquisition device is included, so that the server in the server group that receives the quality detection request processes the received display screen image.
  • Step 22 The above display screen image is input into a defect detection model to obtain a defect detection result, wherein the defect detection model is obtained by training a historical defect display screen image using a deep convolutional neural network structure and an object detection algorithm.
  • the server receiving the quality inspection request inputs the display screen image in the quality inspection request to a defect detection model running on the server, and the defect detection model performs defect detection, and then obtains the defect detection result.
  • the defect detection model running on the server is obtained by training the historical defect display image using a deep convolutional neural network structure and an object detection algorithm. That is, the display screen image on the display screen production line is used as the input of the defect detection model, and the features in the display screen image (i.e., defects existing in the display screen image) are extracted as a defect detection model by using a deep convolutional neural network structure and an object detection algorithm Output, and then train the defect detection model.
  • the object detection means that the machine can automatically determine the object in the display screen different from other positions in the display and the area frame corresponding to the object by using the target detection method.
  • the area frame refers to the position of the object on the display screen. The extent of the area. It is worth explaining that the object detection can distinguish which positions of the display screen have defects, and the defects belong to a certain type of defect. Therefore, in the embodiment of the present application, the defect detection model is obtained by training a large number of historical defect display screen images by using a deep convolutional neural network structure and an object detection algorithm, that is, first determining different defects existing in the historical defect display screen image. Then, the defects in the image and the specific categories corresponding to the defects are separated from the perspective of the orientation or the area, and they are marked in the display screen image. Finally, they are classified and combined to obtain training.
  • the defect detection model in the embodiment of the present application is obtained by training a historical defect display screen image using a deep convolutional neural network structure and an object detection algorithm, which can be explained as follows:
  • the defect detection model is a combination training of candidate region loss, region category loss, and region boundary loss of historical defect display screen images, so that the total loss value of the candidate region loss, the region category loss, and the region boundary loss meets the preset loss. Threshold results.
  • the candidate area loss refers to the loss value between the selected defect area and the actual defect area in the historical defect display screen image
  • the area category loss refers to the loss value between the predicted defect category and the actual defect category in the selected defect area.
  • the area boundary Loss refers to the loss between the predicted defect boundary and the actual defect boundary in the selected defect area.
  • the defect detection model is based on a deep convolutional neural network (Deep CNNs) structure.
  • the structure of the deep convolutional neural network is mainly composed of a convolutional layer, a pooling layer, and a fully connected layer. And other components, the granularity and hierarchical selection of the deep convolutional neural network can be determined according to the actual situation, which is not limited in the embodiments of the present application.
  • the convolution operation of the convolution layer refers to the use of convolution kernels with different weights to scan and convolve the display screen image or the image feature map obtained after at least one convolution process, extract various types of features from it, and obtain The process of an image feature map.
  • the convolution kernel is a weight matrix, that is, the weight used in the convolution is represented by a matrix, the matrix is the same size as the corresponding image area, its rows and columns are odd numbers, and it is a weight matrix.
  • the pooling operation of the pooling layer refers to performing a dimensionality reduction operation on the feature map output by the convolution layer, retaining the main features in the feature map.
  • the fully connected layer maps the feature map generated by the convolutional layer into a fixed-length (generally the number of image categories in the input image data set) feature vector, which contains the combined information of all features of the input image. That is, the most characteristic image features in the image are retained to complete the image classification task.
  • the embodiment of the present application can use such a deep neural network model with convolution, pooling, and fully connected operations, and has characteristics such as deformation, blurring, and lighting changes of the display screen image collected by the image acquisition device on the display screen production line. High robustness and higher generalizability for classification tasks.
  • the object detection algorithm may be Faster RCNN algorithm.
  • Faster RCNN algorithm first uses the convolution operation of the convolutional neural network structure to obtain its feature map, and then calculates the selected defect of the display screen image. Whether the area contains specific defects. If there are defects, on the one hand, the loss value between the selected defect area and the actual defect area (candidate area loss) can be calculated; on the other hand, a convolutional neural network can be used for feature extraction, and then the prediction of the Defect category and defect boundary, and then calculate the loss value between the predicted defect category and the actual defect category in the selected defect area (region category loss) and the loss value between the predicted defect boundary and the actual defect boundary in the selected defect area ( Area boundary loss). If a specific defect is not included in the selected defect area of the display image, it is not classified.
  • a combination of a Faster RCNN algorithm and a Region Proposal Network can also be adopted; the candidate region network is to obtain whether the area in the original image of the display screen image contains a specific object (defect ), If there is an object in an area containing the original image, the original image is corresponding to the feature map extracted by the Faster RCNN algorithm, and the type of the object is predicted at the area on the feature map and the area is on the display screen s position.
  • the defect detection model is the result of combined training on candidate region loss, region category loss, and region boundary loss of historical defect display screen images, and a loss function including candidate region loss, region category loss, and region boundary loss can be obtained.
  • the loss function is used to evaluate the difference between the output of the deep convolutional neural network and the actual value during the training phase of the defect detection model, and then the weight value between each neuron is updated with the value of the loss function.
  • the purpose of training deep convolutional neural networks is to minimize the value of the loss function.
  • the preset loss threshold refers to a value that meets the business requirements of the display screen.
  • the depth of the deep convolutional neural network required for training the above defect detection model may be different, and it can be determined according to the actual situation, which is not limited in this embodiment.
  • Step 23 Determine the defect on the display screen corresponding to the display screen image, the defect type corresponding to the defect, and the position corresponding to the defect according to the defect detection result.
  • the defect of the display screen corresponding to the display screen image, the defect type corresponding to the defect, and the defect may be determined according to the defect detection result. Corresponding position.
  • the defect detection result that can be obtained by the defect detection model may include a defect category (there are several types of defects on the display screen), and a defect location (the specificity of each defect). position). That is, the defect detection model in the embodiment of the present application can detect several types of defect types in the display screen image and the specific number of each type of defect.
  • FIG. 3 is a schematic flow chart of the second embodiment of the display screen quality detection method provided in the embodiment of the present application.
  • step 23 determine the corresponding display screen image After the defect on the display screen, the defect type corresponding to the defect, and the position corresponding to the defect
  • it may further include step 24:
  • the quality of the display screen corresponding to the display screen image is determined according to the production stage information and the above-mentioned defect detection result; the production stage information is used to indicate the manufacturer, production environment, and type of the display screen corresponding to.
  • information on a variety of different production stages may result in different defect detection results during the quality inspection process of the display screens.
  • a liquid crystal display it generally goes through production stages such as thin film transistor processing, color filter processing, unit assembly and module assembly.
  • LED screens they generally go through the stages of patch, plug-in, wave soldering, post-soldering, testing, and module assembly.
  • Different types of display screens undergo different production stages. Therefore, when analyzing the defect detection results obtained above, it is necessary to combine the production stage information of each display screen to determine the quality of the display screen.
  • the display screen quality detection method receives a quality test request sent by a console deployed on a display screen production line, and the quality test request includes a display screen image collected by an image acquisition device on the display screen production line, and The display screen image is input to the defect detection model to obtain a defect detection result, and the defect on the display screen, the defect type corresponding to the defect, the position corresponding to the defect, and / or the quality of the display screen are determined according to the defect detection result. Since the above defect detection model is obtained by training a historical defect display screen image using a deep convolutional neural network structure and an object detection algorithm, the defect detection results obtained by using the defect detection model have high classification accuracy and strong intelligence capabilities. The system performance has been improved, and the business scalability is high.
  • FIG. 4 is a schematic flowchart of a third embodiment of a display screen quality detection method according to an embodiment of the present application.
  • the embodiment of the present application further describes the method for detecting the quality of a display screen based on the embodiment shown in FIG. 2.
  • the method may further include the following steps:
  • Step 31 Perform image preprocessing on the display screen image, wherein the image preprocessing includes one or more of the following processes: trimming, cutting, rotating, reducing, and enlarging.
  • the image acquisition device deployed on the display screen production line is generally a high-precision camera. Therefore, the display screen image collected by the image acquisition device may have a large size, Either the pixel is high or the position is inappropriate. Therefore, after receiving the display screen image included in the quality detection request sent by the console, the display screen image needs to be pre-processed according to the actual situation.
  • the edge area of the display screen image is large, you can trim the display screen image to retain useful parts of the display screen image, or if the size of the display screen image is large, you can The screen image is cut and reduced, so that the display screen image sent to the defect detection model can be completely detected, and the detection accuracy highlighted by the display screen is improved. Or, if it is necessary to focus on detecting a certain area in the display screen image, the area in the display screen image may be enlarged and processed to make the pre-processed display screen image meet the detection standard.
  • the display screen image before the display screen image is input into the defect detection model to obtain a defect detection result, the display screen image is trimmed, or / and cut or / and, and / or Image preprocessing such as rotation, or / and reduction, and / or enlargement can make the display screen image sent to the defect detection model meet the detection standard, laying a foundation for defect detection in subsequent display screen images and improving display screen quality Detection accuracy.
  • FIG. 5 is a schematic flowchart of a third embodiment of a display screen quality detection method according to an embodiment of the present application.
  • the above step 22 inputting the above display screen image into a defect detection model to obtain a defect detection result
  • the above step 22 may be implemented by the following steps:
  • Step 41 Determine a detection model server carrying processing resources according to the load balancing policy.
  • a server group is deployed on the display screen production line.
  • the number of servers in the server group may be multiple, and each server runs a defect detection model.
  • the defect detection model running on each server is the same. Therefore, each server can receive the quality inspection request sent by the console, and then can use the defect detection model carried by itself to perform quality inspection on the display screen image. .
  • the console can also send a quality detection request to any server in the server group in real time.
  • the load balancing can be performed according to a preset load Strategy, from the server group, determine a detection model server that carries processing resources, that is, load balancing and scheduling in real time according to the deployment of the defect detection model on the display production line.
  • Step 42 Input the display screen image into a defect detection model running on the detection model server to obtain a defect detection result.
  • the above display screen image may be input into a defect detection model running on the detection model server, and the The defect detection model detects defects on the display screen image, and then obtains the defect detection results.
  • the defect detection model is obtained by the training module using a deep convolutional neural network structure and an object detection algorithm to train the historical defect display screen image.
  • a detection model server carrying processing resources is determined according to a load balancing policy, and the display screen image is input to a defect detection model running on the detection model server to obtain defect detection.
  • the method may further include the following steps:
  • defect detection results are stored as logs in the production database by the controller;
  • the display screen image and the defect detection result are input into the defect detection model in order to optimize the defect detection model.
  • the tester may preset a solution when the display screen is determined to be a bad screen according to the production scene and production stage information of the display screen, for example, send an alarm to the production manager through the controller Information, and / or, store the above-mentioned defect detection result as a log in the production database through the controller, and / or, send the production control instruction to the console through the controller to eliminate the defect, and / or, the above display screen image and
  • the defect detection result is input to the defect detection model in order to optimize the defect detection model and the like.
  • an alarm message may be issued to enable the production manager to locate the category of the defect in a timely manner. And location, and give the solution.
  • the above-mentioned defect detection result may be stored in the production database as a log by the controller, that is, the defect category of the display screen, and / or, a defect instance , And / or, the location of the defect is stored in the production database as a log, which can then be filtered into the training database, and the training module (which can be a software program such as a training engine) updates the above-mentioned defect detection model according to the defective display screen image.
  • the training module which can be a software program such as a training engine
  • a production control instruction may also be sent to the console through the controller to eliminate the defect. That is, the inspection model server bearing the defect inspection model can determine the cause of the defect through the controller, and then adjust the production process accordingly, that is, the inspection model server sends the production control instruction to the console through the controller to eliminate the display Defects appear to reduce the probability of a bad screen; or when it is determined that a defect exists in the display screen based on the above defect detection results, the controller can also send instructions to the business response device to reduce the outflow of bad screens, such as through the controller Send a grab command to the robot arm to grab the defective display.
  • the display screen image and the defect detection result may also be directly input into the defect detection model in order to optimize the defect detection model, that is, directly input the defect detection model.
  • the display screen image corresponding to the bad screen is used as the input of the defect detection model, and the defect detection result of the bad screen is used as the output of the defect detection model to optimize the defect detection model, thereby improving the detection accuracy of the defect detection model.
  • the embodiment of the present application is not limited to the above-mentioned one or more operations that can be performed by the detection model server when the display screen is determined to be a bad screen, which can be determined according to actual conditions, and will not be described again here.
  • the operation steps corresponding to the display quality detection method can also be distributed to the above-mentioned different Equipment to perform.
  • the image acquisition device collects the display screen image
  • the console sends the display screen image collected by the image acquisition device to the detection model server in the server group according to the load balancing policy, and the defect detection model running on the detection model server pairs the display screen.
  • the image is subjected to preset pre-processing for defect detection, and the defect detection result is given.
  • the detection model server can send the defect detection results to the controller.
  • the controller combines the actual business scenario and responds to the requirements of the above-mentioned defect detection results according to business requirements, such as alarms, storage logs, and control. Production control instructions, etc.
  • the controller can also store the defect detection result and the response processing behavior as a log in the production database, so that the training module updates the above obtained according to the display screen image and the defect detection result in the training database.
  • Defect detection model the training database stores data such as screen images with defects and corresponding defect detection results screened from the production database.
  • the defect detection model running on the server can be gradually replaced by a small-traffic online method, so as to achieve the purpose of dynamically expanding and generalizing the defect detection model with business scene and production stage information.
  • the display screen quality detection method in the embodiment of the present application is run for a period of time on the display screen production line, the accuracy of the above defect detection and defect location can be manually reviewed through the information in the production database, and then the above training database is updated to retrain the defects. Inspection model to improve the accuracy of defect detection.
  • FIG. 6 is a first schematic structural diagram of an embodiment of a display screen quality detection device according to an embodiment of the present application.
  • the display screen quality detection device provided in the embodiment of the present application may include a receiving module 51, a processing module 52, and a determining module 53.
  • the receiving module 51 is configured to receive a quality inspection request sent by a console deployed on a display screen production line, where the quality inspection request includes a display screen image collected by an image acquisition device on the display screen production line;
  • a processing module 52 is configured to input the display screen image into a defect detection model to obtain a defect detection result, wherein the defect detection model is to train a historical defect display screen image by using a deep convolutional neural network structure and an object detection algorithm. owned;
  • a determining module 53 is configured to determine a defect on a display screen corresponding to the display screen image, a defect type corresponding to the defect, and a position corresponding to the defect according to the defect detection result.
  • the defect detection model is obtained by training a historical defect display image using a deep convolutional neural network structure and an object detection algorithm, including:
  • the defect detection model is a combination training of candidate area loss, area category loss, and area boundary loss of the historical defect display screen image, so that the candidate area loss, the area category loss, and the area boundary loss are combined.
  • the candidate area loss refers to a loss value between a selected defect area and an actual defect area in the historical defect display screen image
  • the area category loss refers to a predicted defect category and an actual defect category in the selected defect area.
  • the value of the loss between the regions is the loss between the predicted defect boundary and the actual defect boundary in the selected defect region.
  • the processing module 52 is further configured to image the display screen image before inputting the display image into a defect detection model to obtain a defect detection result. Pre-processing.
  • the image preprocessing includes one or more of the following processes:
  • the processing module 52 is specifically configured to determine a detection model server carrying processing resources according to a load balancing policy, and input the display screen image to a server running at A defect detection result is obtained in the defect detection model on the detection model server.
  • the determining module 53 is further configured to determine, according to the production stage information and the defect detection result, the quality of the display screen corresponding to the display screen image.
  • the production stage information is used to indicate a manufacturer, a production environment, and a type of the display screen corresponding to the display screen.
  • the processing module 52 is further configured to determine, according to the defect detection result, in the determining module 53 that the quality of the display screen corresponding to the display screen image is good. After the bad, if it is determined that the display is a bad screen, perform one or more of the following operations:
  • the display screen image and the defect detection result are input into the defect detection model in order to optimize the defect detection model.
  • the display screen quality detection device of this embodiment may be used to execute the implementation solutions of the method embodiments shown in FIG. 2 to FIG. 5.
  • the specific implementation manner and technical effect are similar, and details are not described herein again.
  • FIG. 7 is a second schematic structural diagram of an embodiment of a display screen quality detection device according to an embodiment of the present application.
  • the display screen quality detection device may be an electronic device, and the electronic device may include a memory 61 and at least one process. ⁇ 62 ⁇ 62.
  • the memory 61 is configured to store a program instruction.
  • the processor 62 is configured to implement each step in the display screen quality detection method shown in the method embodiments shown in FIG. 2 to FIG. 5 when the program instructions are executed. For specific implementation principles, refer to the foregoing embodiment. This embodiment I won't repeat them here.
  • the electronic device may further include an input / output interface 63.
  • the input / output interface 63 may include an independent output interface and an input interface, or may be an integrated interface that integrates input and output.
  • the output interface is used to output data, and the input interface is used to obtain input data.
  • the output data is the collective name of the output in the method embodiment, and the input data is the collective name of the input in the method embodiment.
  • the present application further provides a storage medium, where the storage medium has instructions stored therein, which when run on a computer, cause the computer to execute the methods of the method embodiments shown in FIG. 2 to FIG. 5.
  • the present application also provides a program product, which includes a computer program stored in a storage medium.
  • At least one processor of the display screen quality detection device may read the computer program from a storage medium, and the at least one processor executes the computer program to cause the display screen quality detection device to execute the method in the method embodiments shown in FIGS.
  • a person of ordinary skill in the art may understand that all or part of the steps of implementing the foregoing method embodiments may be implemented by a program instructing related hardware.
  • the aforementioned program may be stored in a computer-readable storage medium.
  • the steps including the foregoing method embodiments are performed; and the foregoing storage medium includes various media that can store program codes, such as a ROM, a RAM, a magnetic disk, or an optical disc.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Software Systems (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Computational Linguistics (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Biophysics (AREA)
  • Mathematical Physics (AREA)
  • Biomedical Technology (AREA)
  • Artificial Intelligence (AREA)
  • Chemical & Material Sciences (AREA)
  • Nonlinear Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Quality & Reliability (AREA)
  • Optics & Photonics (AREA)
  • Immunology (AREA)
  • Pathology (AREA)
  • Crystallography & Structural Chemistry (AREA)
  • Biochemistry (AREA)
  • Analytical Chemistry (AREA)
  • Signal Processing (AREA)
  • Manufacturing & Machinery (AREA)
  • Investigating Materials By The Use Of Optical Means Adapted For Particular Applications (AREA)
  • Image Processing (AREA)
  • Image Analysis (AREA)
  • Testing Of Optical Devices Or Fibers (AREA)
  • Test And Diagnosis Of Digital Computers (AREA)

Abstract

本申请提供一种显示屏质量检测方法、装置、电子设备及存储介质，该方法包括：接收部署在显示屏生产线上的控制台发送的质量检测请求，该质量检测请求中包含显示屏生产线上的图像采集设备采集的显示屏图像，将该显示屏图像输入到缺陷检测模型中得到缺陷检测结果，该缺陷检测模型是采用深度卷积神经网络结构和物体检测算法对历史缺陷显示屏图像进行训练得到的，根据缺陷检测结果确定显示屏图像对应的显示屏上的缺陷、缺陷对应的缺陷类别、以及缺陷对应的位置。该技术方案的缺陷检测准确度高、系统性能好，业务扩展能力高。

Description

显示屏质量检测方法、装置、电子设备及存储介质
本申请要求于2018年07月02日提交中国专利局、申请号为201810710537.5、申请人为北京百度网讯科技有限公司、申请名称为“显示屏质量检测方法、装置、电子设备及存储介质”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请涉及计算机技术领域，尤其涉及一种显示屏质量检测方法、装置、电子设备及存储介质。
背景技术
信息显示技术在人们生活中的作用与日俱增，显示屏也因其体积小、重量轻、功率低、分辨率高、亮度高和无几何变形诸多特点被广泛应用。但在显示屏的生产过程中，由于各种原因可能导致生产的显示屏存在显示缺陷，例如，点缺陷、线缺陷和面缺陷等。因而，显示屏质量检测是生产过程中的重要环节。
现有技术中，显示屏质量检测主要采用人工检测或机器辅助的人工检测方法。具体的，人工检测方法是指依赖行业专家肉眼观察从生产环境中采集到的图片给出判断；机器辅助的人工检测方法是指首先利用固化有行业专家经验的质检系统对待检测显示屏图像进行检测，其次再由行业专家对疑似存在缺陷的图片进行检测判断。
然而，不管是人工检测方法，还是机器辅助的人工检测方法均受人的主观影响因素较大，检测准确度低、系统性能差，业务扩展能力低。
发明内容
本申请提供一种显示屏质量检测方法、装置、电子设备及存储介质，以克服现有显示屏缺陷检测方法中由于受人的主观影响因素较大，致使检测准确度低、系统性能差、业务扩展能力低的问题。
本申请的第一方面提供一种显示屏质量检测方法，包括：
接收部署在显示屏生产线上的控制台发送的质量检测请求，所述质量检测请求中包含所述显示屏生产线上的图像采集设备采集的显示屏图像；
将所述显示屏图像输入到缺陷检测模型中得到缺陷检测结果，其中，所述缺陷检测模型是对历史缺陷显示屏图像进行物体检测算法训练得到的；
根据所述缺陷检测结果确定所述显示屏图像对应的显示屏上的缺陷、所述缺陷对应的缺陷类别、以及所述缺陷对应的位置。
本实施例中，利用基于深度卷积神经网络结构和物体检测算法对历史缺陷显示屏图像进行训练得到的缺陷检测模型对显示屏图像进行缺陷检测，以确定显示屏图像对 应的显示屏上的缺陷、所述缺陷对应的缺陷类别、以及所述缺陷对应的位置。基于该缺陷检测模型得到的缺陷检测结果的精度高，智能化能力强，系统性能高，业务可扩展能力高。
可选的，所述缺陷检测模型是采用深度卷积神经网络结构和物体检测算法对历史缺陷显示屏图像进行训练得到的，包括：
所述缺陷检测模型是对所述历史缺陷显示屏图像的候选区域损失、区域类别损失、区域边界损失进行组合训练，以使所述候选区域损失、所述区域类别损失和所述区域边界损失的总损失值满足预设损失阈值的结果；
其中，所述候选区域损失指所述历史缺陷显示屏图像中选定缺陷区域与实际缺陷区域之间的损失值，所述区域类别损失指所述选定缺陷区域中预测缺陷类别与实际缺陷类别之间的损失值，所述区域边界损失指所述选定缺陷区域中预测缺陷边界与实际缺陷边界之间的损失。
在本实施例中，通过对显示屏图像的候选区域损失、区域类别损失和区域边界损失进行组合训练得到的缺陷检测模型分类精度高，提高了检测系统的性能。
可选的，所述将所述显示图像输入到缺陷检测模型中得到缺陷检测结果之前，还包括：
对所述显示屏图像进行图像预处理，其中，所述图像预处理包括下述处理中的一项或多项：
裁边、剪切、旋转、缩小、放大。
可选的，所述将所述显示屏图像输入到缺陷检测模型中得到缺陷检测结果，包括：
根据负载均衡策略，确定承载处理资源的检测模型服务器；
将所述显示屏图像输入到运行在所述检测模型服务器上的所述缺陷检测模型中得到缺陷检测结果。
在本实施例中，对获取的显示屏图像进行图像预处理，以使得处理后的显示屏图像符合检测标准，为后续显示屏图像中的缺陷检测奠定了基础，提高了显示屏质量的检测准确度。
可选的，所述方法还包括：
根据生产阶段信息以及所述缺陷检测结果，确定所述显示屏图像对应的显示屏的质量好坏；所述生产阶段信息用于指示所述显示屏对应的生产厂家、生产环境、以及所述显示屏的类型。
在本实施例中，对于不同种类的显示屏，其所经历的生产阶段不同，在对缺陷检测结果进行分析时，结合各显示屏的生产阶段信息进行来确定显示屏的质量好坏，可以提高获取显示屏质量的准确性。
可选的，所述根据生产阶段信息以及所述缺陷检测结果，确定所述显示屏图像对应的显示屏的质量好坏之后，还包括：
若确定所述显示屏为坏屏，则执行以下一项或多项操作：
通过控制器向生产管理者发送报警信息；
通过控制器将所述缺陷检测结果作为日志存储到生产数据库中；
通过控制器向所述控制台发送生产控制指令以便消除缺陷；
将所述显示屏图像和所述缺陷检测结果输入到所述缺陷检测模型中以便优化所述缺陷检测模型。
在实施例中，显示屏质量检测方法在显示屏生产线上运行一段时间后，可以人工通过生产数据库中的信息，复查上述缺陷检测和缺陷定位的准确率，然后更新上述训练数据库，重新训练缺陷检测模型，以提高缺陷检测准确率。
本申请的第二方面提供一种显示屏质量检测装置，其特征在于，包括：
接收模块，用于接收部署在显示屏生产线上的控制台发送的质量检测请求，所述质量检测请求中包含所述显示屏生产线上的图像采集设备采集的显示屏图像；
处理模块，用于将所述显示屏图像输入到缺陷检测模型中得到缺陷检测结果，其中，所述缺陷检测模型是采用深度卷积神经网络结构和物体检测算法对历史缺陷显示屏图像进行训练得到的；
确定模块，用于根据所述缺陷检测结果确定所述显示屏图像对应的显示屏上的缺陷、所述缺陷对应的缺陷类别、以及所述缺陷对应的位置。
本申请提供的显示屏质量检测装置的技术效果具体可与上述显示屏质量检测方法的技术效果相同。
可选的，所述缺陷检测模型是采用深度卷积神经网络结构和实例分割算法对历史缺陷显示屏图像进行训练得到的，包括：
所述缺陷检测模型是对所述历史缺陷显示屏图像的候选区域损失、区域类别损失和区域边界损失进行组合训练，以使所述候选区域损失、所述区域类别损失和所述区域边界损失的总损失值满足预设损失阈值的结果；
其中，所述候选区域损失指所述历史缺陷显示屏图像中选定缺陷区域与实际缺陷区域之间的损失值，所述区域类别损失指所述选定缺陷区域中预测缺陷类别与实际缺陷类别之间的损失值，所述区域边界损失指所述选定缺陷区域中预测缺陷边界与实际缺陷边界之间的损失。
可选的，所述处理模块，还用于在将所述显示图像输入到缺陷检测模型中得到缺陷检测结果之前，对所述显示屏图像进行图像预处理；
其中，所述图像预处理包括下述处理中的一项或多项：
裁边、剪切、旋转、缩小、放大。
可选的，所述处理模块，具体用于根据负载均衡策略，确定承载处理资源的检测模型服务器，将所述显示屏图像输入到运行在所述检测模型服务器上的所述缺陷检测模型中得到缺陷检测结果。
可选的，所述确定模块，还用于根据生产阶段信息以及所述缺陷检测结果，确定所述显示屏图像对应的显示屏的质量好坏；所述生产阶段信息用于指示所述显示屏对应的生产厂家、生产环境、以及所述显示屏的类型。
可选的，所述处理模块，还用于在所述确定模块根据所述缺陷检测结果确定所述显示屏图像对应的显示屏的质量好坏之后，若确定所述显示屏为坏屏，则执行以下一项或多项操作：
通过控制器向生产管理者发送报警信息；
通过控制器将所述缺陷检测结果作为日志存储到生产数据库中；
通过控制器向所述控制台发送生产控制指令以便消除缺陷；
将所述显示屏图像和所述缺陷检测结果输入到所述缺陷检测模型中以便优化所述缺陷检测模型。
本申请第三方面提供一种电子设备，包括处理器、存储器及存储在所述存储器上并可在处理器上运行的计算机程序，所述处理器执行所述程序时实现如上述第一方面以及第一方面各种可能实现方式中任一项所述的方法。
本申请第四方面提供一种存储介质，所述存储介质中存储有指令，当其在计算机上运行时，使得计算机执行如第一方面以及第一方面各种可能实现方式中任一项所述的方法。
本申请提供一种显示屏质量检测方法、装置、电子设备及存储介质，该方法包括：接收部署在显示屏生产线上的控制台发送的质量检测请求，该质量检测请求中包含显示屏生产线上的图像采集设备采集的显示屏图像，将该显示屏图像输入到缺陷检测模型中得到缺陷检测结果，该缺陷检测模型是采用深度卷积神经网络结构和物体检测算法对历史缺陷显示屏图像进行训练得到的，根据缺陷检测结果确定显示屏图像对应的显示屏上的缺陷、缺陷对应的缺陷类别、以及缺陷对应的位置。该技术方案的缺陷检测准确度高、系统性能好，业务扩展能力高。
附图说明
图1为本申请实施例提供的显示屏质量检测系统的结构示意图；
图2为本申请实施例提供的显示屏质量检测方法实施例一的流程示意图；
图3为本申请实施例提供的显示屏质量检测方法实施例二的流程示意图；
图4为本申请实施例提供的显示屏质量检测方法实施例三的流程示意图；
图5为本申请实施例提供的显示屏质量检测方法实施例三的流程示意图；
图6为本申请实施例提供的显示屏质量检测装置实施例的结构示意图一；
图7为本申请实施例提供的显示屏质量检测装置实施例的结构示意图二。
具体实施方式
为使本申请的目的、技术方案和优点更加清楚，下面将结合本申请的实施例，对本申请实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例是本申请一部分实施例，而不是全部的实施例。基于本申请中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都属于本申请保护的范围。
本申请实施例提供的显示屏质量检测方法应用于显示屏质量检测系统中，图1为本申请实施例提供的显示屏质量检测系统的结构示意图。如图1所示，该显示屏质量检测系统包括：部署在显示屏生产线上的图像采集设备11、控制台12、服务器组13、控制器14、数据库15和训练器16等多个不同的设备。
其中，图像采集设备11采集显示屏生产线上的显示屏图像，控制台12接收图像采集设备11采集的显示屏图像，并将该显示屏图像发送给服务器组13中的检测模型服务器130，检测模型服务器130将接收到的显示屏图像输入到本身运行的缺陷检测 模型中得到缺陷检测结果，控制器14接收检测模型服务器130的缺陷检测结果，并结合生产阶段信息给出业务响应，控制器14还可以将缺陷检测结果作为日志存储到数据库15中。此外，图像采集设备11采集到的显示屏图像还可以直接存储到数据库15中，作为缺陷检测模型训练的原始数据。训练器16提取数据库中的历史缺陷显示屏图像后，采用深度卷积神经网络结构和实例分割算法对历史缺陷显示屏图像进行训练得到缺陷检测模型。
可选的，上述数据库15可以包括生产数据库151和训练数据库152，生产数据库151可以接收并保存控制器14发送的缺陷检测结果以及图像采集设备11采集到的显示屏图像，训练数据库152可以存储从生产数据库151提取的历史缺陷显示屏图像和对应的原始显示屏图像，以使训练器16训练得到检测准确率高的缺陷检测模型。
可选的，本申请实施例中的训练器16可以是由硬件和/或软件功能实现的训练引擎，其作为缺陷检测模型的训练工具。
本申请实施例的显示屏质量检测系统中还可以包括处理器、存储器等其他实体模块，本实施例不限于此。
下面首先针对本申请实施例所适用的应用场景进行简要说明。
现阶段，3C产业(3C产业是指结合电脑、通讯、和消费性电子三大科技产品整合应用的资讯家电产业)的整体智能自动化程度较低，通过对手机屏等显示屏行业的调研分析可知，大部分生产厂家对手机屏采用的检测方式可以分为两种，即：人工检测方法和机器辅助的人工检测方法。
其中，人工检测方法是指依赖于行业专家肉眼观察从生产环境中采集到的图像进行判断，该方法受人的主观影响因素较大、检测效率较低，且对人眼的伤害较大。此外，由于显示屏的生成车间一般为无尘环境，工作人员进去前需要进行清洁准备，穿戴无尘衣服，其还可能对工作人员的健康和安全会产生不利影响。
机器辅助的人工检测方法也可以称为基于液晶模组检测设备检测方法，具体原理为：首先由具有一定判断能力的质检系统过滤掉不存在缺陷的图像，再由行业专家对疑似存在缺陷的图像进行检测判断。在机器辅助的人工检测方法中，质检系统多为专家系统和特征工程系统发展而来，是指专家将经验固化在质检系统中，使其具有一定的自动化能力。因此，机器辅助的人工检测方法不仅准确率低，系统性能差，无法覆盖厂商所有的检测标准，而且这种方法还效率低，容易漏判误判，检测后的图像数据很难进行二次利用挖掘。此外，在上述质检系统中，特征和判定规则都是基于行业专家的经验固化到机器中的，难以随业务的发展迭代，导致随着生产工艺的发展，质检系统的检测精度越来越低，甚至可能降低到完全不可用的状态。进一步的，质检系统的特征都由第三方供应商预先固化在硬件中，升级时不仅需要对生产线进行重大改造，而且价格昂贵，其在安全性、规范化、可扩展性等方面都存在着明显不足，不利于显示屏生产线的优化升级，业务扩展能力低。
综上所述，人工检测方法和机器辅助的人工检测方法均存在如下缺点：不仅效率低下、容易出现误判，而且这两种方法产生的工业数据不易存储、管理和二次挖掘再利用。
本申请实施例基于人工智能技术在计算机视觉中的最新发展，研发一种自动化、 高精度的显示屏质量检测方法，利用图像采集设备在显示屏生产线上实时采集的显示屏图像，实时对显示屏的表面质量进行检测判断，如果检测到当前图像采集设备采集到的显示屏存在质量问题，则确定出该质量问题在图片中所在的位置及其所属的类别以及类别实例。
值得说明的是，本申请实施例可以应用于任何可以利用人眼、计算机视觉进行显示屏检测的场景，本实施例中的显示屏可以包括如下的任意一种：等离子屏幕、液晶显示(liquid crystal display，LCD)屏幕、发光二极管(light emitting diode，LED)屏幕、有机发光二极管(organic light-emitting diode，OLED)屏幕等。本申请实施例并不局限于上述的几种屏幕，其还可能包括其他显示屏幕，此处不再赘述。
可选的，本申请实施例中所述的质量问题可以包括，但是不局限于包括点缺陷、线缺陷、面缺陷和mura等不同类别的缺陷问题。此处不进行一一介绍。可选的，mura是指由于显示屏亮度不均匀，造成的各种痕迹现象。
下面，通过具体实施例对本申请的技术方案进行详细说明。需要说明的是，下面这几个具体的实施例可以相互结合，对于相同或相似的概念或过程可能在某些实施例中不再赘述。
本申请实施例中，“多个”是指两个或两个以上。“和/或”，描述关联对象的关联关系，表示可以存在三种关系，例如，A和/或B，可以表示：单独存在A，同时存在A和B，单独存在B这三种情况。字符“/”一般表示前后关联对象是一种“或”的关系。
图2为本申请实施例提供的显示屏质量检测方法实施例一的流程示意图。如图2所示，在本申请实施例中，该显示屏质量检测方法可以包括如下步骤：
步骤21：接收部署在显示屏生产线上的控制台发送的质量检测请求，该质量检测请求中包含显示屏生产线上的图像采集设备采集的显示屏图像。
可选的，在本申请实施例中，显示屏生产线上部署有图像采集设备、控制台、服务器组、控制器、数据库等多个设备。图像采集设备可以是高精度图像采集摄像头，在显示屏的生产过程中，通过调整图像采集设备的角度、光线、滤镜、倍镜、聚焦等，可以采集到多张处于显示屏生产线上的显示屏对应的显示屏图像。
当显示屏生产线上的图像采集设备采集到显示屏图像之后，部署在显示屏生产线上的控制台则可以向显示屏生产线上部署有缺陷检测模型的服务器组发送质量检测请求，该质量检测请求中包含上述图像采集设备采集的显示屏图像，以使服务器组中接收到该质量检测请求的服务器对接收到的显示屏图像进行处理。
步骤22：将上述显示屏图像输入到缺陷检测模型中得到缺陷检测结果，其中，该缺陷检测模型是采用深度卷积神经网络结构和物体检测算法对历史缺陷显示屏图像进行训练得到的。
可选的，接收到质量检测请求的服务器将质量检测请求中的显示屏图像输入到服务器上运行着的缺陷检测模型中，由缺陷检测模型执行缺陷检测，进而得到缺陷检测结果。
值得说明的是，服务器上运行的缺陷检测模型是采用深度卷积神经网络结构和物体检测算法对历史缺陷显示屏图像进行训练得到的。即，显示屏生产线上的显示屏图 像作为缺陷检测模型的输入，利用深度卷积神经网络结构和物体检测算法提取显示屏图像中的特征(即，显示屏图像中存在的缺陷)作为缺陷检测模型的输出，进而对缺陷检测模型进行训练。
具体的，物体检测是指机器可以自动从图像中用目标检测方法确定出显示屏中不同于其他位置处的物体，以及该物体对应的区域框；区域框是指物体在显示屏上的位置所在的区域的范围。值得说明的是，物体检测可以区分出显示屏的哪些位置上具有缺陷，且该缺陷具体属于某一类缺陷。因而，在本申请实施例中，缺陷检测模型是基于采用深度卷积神经网络结构和物体检测算法对大量历史缺陷显示屏图像进行训练得到，即首先确定出历史缺陷显示屏图像中存在的不同缺陷，再从方位或区域的角度出发分隔出图像中的缺陷，以及缺陷对应的具体类别，并在显示屏图像中标注出来，最后对其进行分类统计、组合训练得到。
作为一种示例，本申请实施例中的缺陷检测模型是采用深度卷积神经网络结构和物体检测算法对历史缺陷显示屏图像进行训练得到的，可以解释如下：
缺陷检测模型是对历史缺陷显示屏图像的候选区域损失、区域类别损失和区域边界损失进行组合训练，以使该候选区域损失、该区域类别损失和该区域边界损失的总损失值满足预设损失阈值的结果。
其中，候选区域损失指历史缺陷显示屏图像中选定缺陷区域与实际缺陷区域之间的损失值，区域类别损失指选定缺陷区域中预测缺陷类别与实际缺陷类别之间的损失值，区域边界损失指选定缺陷区域中预测缺陷边界与实际缺陷边界之间的损失。
在本申请实施例中，缺陷检测模型以深度卷积神经网络(Deep convolutional neural networks，Deep CNNs)结构为基础，深度卷积神经网络的结构主要是由卷积层、池化层、全连接层等组成，对于深度卷积神经网络的粒度和层次选择可以根据实际情况确定，本申请实施例并不对其进行限定。卷积层的卷积操作是指利用权值不同的卷积核对显示屏图像或者经过至少一次卷积处理后得到的图像特征图进行扫描卷积，从中提取出各种类别的特征，并重新得到一个图像特征图的过程。其中，卷积核就是权矩阵，即卷积时使用到的权用一个矩阵表示，该矩阵与对应的图像区域大小相同，其行、列都是奇数，是一个权矩阵。池化层的池化操作则是指对卷积层输出的特征图进行降维操作，保留特征图中的主要特征。在CNN网络中，全连接层将卷积层产生的特征图映射成一个固定长度(一般为输入图像数据集中的图像类别数)的特征向量，这个特征向量包含了输入图像所有特征的组合信息，即将图像中含有最具有特点的图像特征保留了下来以此完成图像分类任务。
本申请实施例可以利用这种具有卷积、池化、全连接操作的深度神经网络模型，对显示屏生产线上由图像采集设备采集到的显示屏图像的变形、模糊、光照变化等特征具有较高的鲁棒性，对于分类任务具有更高的可泛化性。
可选的，在本实施例中，物体检测算法可以是采用Faster RCNN算法，Faster RCNN算法首先利用卷积神经网络结构的卷积操作，得到其特征图，然后再计算显示屏图像的选定缺陷区域内是否包含特定缺陷。如果包含缺陷，一方面，可以计算选定缺陷区域与实际缺陷区域之间的损失值(候选区域损失)，另一方面，可以利用卷积神经网络进行特征提取，然后预测选定缺陷区域中的缺陷类别和缺陷边界，进而计算出选定 缺陷区域中预测缺陷类别与实际缺陷类别之间的损失值(区域类别损失)以及选定缺陷区域中预测缺陷边界与实际缺陷边界之间的损失值(区域边界损失)。如果显示屏图像的选定缺陷区域内不包含特定缺陷，则不进行分类。
具体的，本实施例中还可以采用Faster RCNN算法和候选区域网络(Region Proposal Network)结合的方式；该候选区域网络是获取显示屏图像的原图中的区域中是否包含有特定的物体(缺陷)，如果包含原图的某个区域中包含有物体，则将原图与Faster RCNN算法提取的特征图相对应，在特征图上的该区域处预测该物体的类型以及该区域在显示屏上的位置。
综上可知，缺陷检测模型是对历史缺陷显示屏图像的候选区域损失、区域类别损失和区域边界损失进行组合训练的结果，可以得到包含候选区域损失、区域类别损失和区域边界损失的损失函数，该损失函数在缺陷检测模型的训练阶段用于评估深度卷积神经网络输出结果与实际值的差异，然后用损失函数的值更新每个神经元之间的权重值。深度卷积神经网络的训练目的就是最小化损失函数值。
当缺陷检测模型的输出结果与显示屏图像中标记的缺陷结果之间的误差值小于预设损失阈值时，停止训练。该预设损失阈值是指符合显示屏业务要求的值。
值得说明的是，在本申请实施例中，对于不同的生产场景和显示屏图像的特点，训练上述缺陷检测模型所需要的深度卷积神经网络的深度、每层的神经元数量以及卷积层、池化层的组织方式均可能不同，其可以根据实际情况进行确定，本实施例并不对其进行限定。
步骤23：根据缺陷检测结果确定显示屏图像对应的显示屏上的缺陷、缺陷对应的缺陷类别、以及缺陷对应的位置。
可选的，在本申请的实施例中，当根据缺陷检测模型得出缺陷检测结果之后，可以依据该缺陷检测结果确定上述显示屏图像对应的显示屏的缺陷、缺陷对应的缺陷类别、以及缺陷对应的位置。
在本实施例中，当显示屏图像中存在缺陷时，该缺陷检测模型可以得出的缺陷检测结果中可以包括缺陷类别(显示屏上共存在几类缺陷)、缺陷位置(每个缺陷的具体位置)。也就是说，本申请实施例的缺陷检测模型可以检测出显示屏图像中存在几类缺陷类型，以及每类缺陷的具体个数。
相应的，图3为本申请实施例提供的显示屏质量检测方法实施例二的流程示意图，如图3所示，本实施例中，在该步骤23(根据缺陷检测结果确定显示屏图像对应的显示屏上的缺陷、缺陷对应的缺陷类别、以及缺陷对应的位置)之后，还可以包括步骤24：
根据生产阶段信息以及上述缺陷检测结果，确定显示屏图像对应的显示屏的质量好坏；生产阶段信息用于指示显示屏对应的生产厂家、生产环境、以及显示屏的类型。
具体的，显示屏的生产厂家不同、显示屏的生产环境、以及显示屏的类型不同等多种不同的生产阶段信息均可能在显示屏质量检测过程中得到不同的缺陷检测结果。例如，对于液晶显示屏，其一般要经过薄膜晶体管加工、彩色滤光器加工、单元装配和模块装配等生产阶段。对于LED屏，其一般要经过贴片、插件、波峰焊、后焊、测试、模组组装等阶段。对于不同种类的显示屏，其所经历的生产阶段不同，因而，对 上述得到的缺陷检测结果进行分析时，需要结合各显示屏的生产阶段信息进行来确定显示屏的质量好坏。
本申请实施例提供的显示屏质量检测方法，通过接收部署在显示屏生产线上的控制台发送的质量检测请求，该质量检测请求中包含显示屏生产线上的图像采集设备采集的显示屏图像，将显示屏图像输入到缺陷检测模型中得到缺陷检测结果，并根据该缺陷检测结果确定显示屏上的缺陷、缺陷对应的缺陷类别、以及缺陷对应的位置和/或显示屏的质量好坏。由于上述缺陷检测模型是采用深度卷积神经网络结构和物体检测算法对历史缺陷显示屏图像进行训练得到的，因而，利用该缺陷检测模型得到的缺陷检测结果的分类精度高，智能化能力强，系统性能有所提高，业务可扩展能力高。
可选的，图4为本申请实施例提供的显示屏质量检测方法实施例三的流程示意图。本申请实施例是在图2所示实施例的基础上对显示屏质量检测方法的进一步说明。如图4所示，本实施例提供的显示屏质量检测方法，在上述步骤22(将上述显示屏图像输入到缺陷检测模型中得到缺陷检测结果)之前，该方法还可以包括如下步骤：
步骤31：对上述显示屏图像进行图像预处理，其中，该图像预处理包括下述处理中的一项或多项：裁边、剪切、旋转、缩小、放大。
可选的，在本申请的实施例中，通常情况下，部署在显示屏生产线上的图像采集设备一般是高精度摄像头，因而，利用该图像采集设备采集到的显示屏图像可能尺寸较大、或者像素较高、或者位置不合适等。因而，当接收到控制台发送的包含在质量检测请求中的显示屏图像之后，需要根据实际情况对显示屏图像进行预处理。
例如，若显示屏图像的边缘区域较大，此时，可以对显示屏图像进行裁边处理，保留显示屏图像的有用部分，或者，若显示屏图像的尺寸较大，此时，可以对显示屏图像进行剪切、缩小处理，以使送入缺陷检测模型的显示屏图像可以完全被检测，提高显示屏凸显的检测精度。再或者，若需要着重检测显示屏图像中的某一区域，可以对显示屏图像中的该区域进行放大等处理，以使预处理后的显示屏图像符合检测标准。
本申请实施例的显示屏质量检测方法，通过在将显示屏图像输入到缺陷检测模型中得到缺陷检测结果之前，对上述显示屏图像进行裁边、或/和剪切或/和、或/和旋转、或/和缩小、或/和放大等图像预处理，可以使送入到缺陷检测模型的显示屏图像符合检测标准，为后续显示屏图像中的缺陷检测奠定了基础，提高了显示屏质量的检测准确度。
可选的，在上述实施例的基础上，图5为本申请实施例提供的显示屏质量检测方法实施例三的流程示意图。如图5所示，在本实施例中，上述步骤22(将上述显示屏图像输入到缺陷检测模型中得到缺陷检测结果)可以通过如下步骤实现：
步骤41：根据负载均衡策略，确定承载处理资源的检测模型服务器。
可选的，在本申请的实施例中，显示屏生产线上部署有一个服务器组，该服务器组中的服务器数量可以为多个，每个服务器上均运行着缺陷检测模型。可选的，每个服务器上运行的缺陷检测模型均是相同的，因此，每个服务器均可接收控制台发送的质量检测请求，进而可以利用自身承载的缺陷检测模型对显示屏图像进行质量检测。
作为一种示例，由于部署在显示屏生产线上的图像采集设备实时采集显示屏图像，因而，控制台也可以实时向服务器组中的任一服务器发送质量检测请求。
可选的，由于服务器组中每个服务器上运行的缺陷检测模型是相同的，因而，为了提高服务器上的缺陷检测模型的检测效率，保证缺陷检测模型的负载均衡，可以根据预先设置的负载均衡策略，从服务器组中确定一个承载处理资源的检测模型服务器，即根据显示屏生产线上缺陷检测模型的部署情况实时进行负载均衡和调度。
步骤42：将上述显示屏图像输入到运行在上述检测模型服务器上的缺陷检测模型中得到缺陷检测结果。
可选的，在本申请实施例中，当从服务器组中确定出承载处理资源的检测模型服务器之后，便可以将上述显示屏图像输入到该检测模型服务器上运行的缺陷检测模型中，利用该缺陷检测模型对显示屏图像的缺陷进行检测，进而得到缺陷检测结果。可选的，该缺陷检测模型是由训练模块采用深度卷积神经网络结构和物体检测算法对历史缺陷显示屏图像进行训练得到的。
本申请实施例提供的显示屏质量检测方法，通过根据负载均衡策略，确定承载处理资源的检测模型服务器，并且将上述显示屏图像输入到运行在上述检测模型服务器上的缺陷检测模型中得到缺陷检测结果，能够实现服务器上的负载均衡，提高显示屏图像的检测效率，提升显示屏质量检测系统的性能。
可选的，在本申请实施例提供的显示屏质量检测方法中，在上述步骤23(根据上述缺陷检测结果确定显示屏图像对应的显示屏的质量好坏)之后，还可以包括如下步骤：
若确定上述显示屏为坏屏，则执行以下一项或多项操作：
通过控制器向生产管理者发送报警信息；
通过控制器将上述缺陷检测结果作为日志存储到生产数据库中；
通过控制器向控制台发送生产控制指令以便消除缺陷；
将上述显示屏图像和上述缺陷检测结果输入到上述缺陷检测模型中以便优化上述缺陷检测模型。
可选的，在本申请实施例中，测试人员可以根据显示屏的生产场景和生产阶段信息，预先设置当确定显示屏为坏屏时的解决方案，比如，通过控制器向生产管理者发送报警信息，和/或，通过控制器将上述缺陷检测结果作为日志存储到生产数据库中，和/或，通过控制器向控制台发送生产控制指令以便消除缺陷，和/或，将上述显示屏图像和上述缺陷检测结果输入到上述缺陷检测模型中以便优化上述缺陷检测模型等。
具体的，作为一种示例，当根据上述缺陷检测结果确定出显示屏图像对应的显示屏是坏屏，即显示屏中存在缺陷时，可以发出报警信息，以使生产管理者及时定位缺陷的类别和位置，并且给出解决方案。
作为另一种示例，当根据上述缺陷检测结果确定显示屏中存在缺陷时，可以通过控制器将上述缺陷检测结果作为日志存储到生产数据库中，即将显示屏的缺陷类别，和/或，缺陷实例，和/或，缺陷位置作为日志存储到生产数据库中，进而可以将其筛选到训练数据库中，由训练模块(可以是训练引擎等软件程序)根据存在缺陷的显示屏图像更新上述缺陷检测模型。
作为再一种示例，当根据上述缺陷检测结果确定显示屏中存在缺陷时，还可以通过控制器向控制台发送生产控制指令以便消除缺陷。即，承载缺陷检测模型的检测模 型服务器可以通过控制器确定出缺陷出现的原因，进而根据相应的调整生产流程，也即，检测模型服务器通过控制器向控制台发送生产控制指令以消除显示屏上出现的缺陷，以减少坏屏出现的概率；或者，当根据上述缺陷检测结果确定显示屏中存在缺陷时，还可以通过控制器向业务响应装置发送指令，减少坏屏的流出，如通过控制器向机械臂发送抓取指令，将包含有缺陷的显示屏抓取。
作为又一种示例，当根据上述缺陷检测结果确定显示屏中存在缺陷时，也可以直接将上述显示屏图像和上述缺陷检测结果输入到上述缺陷检测模型中以便优化上述缺陷检测模型，即直接将坏屏对应的显示屏图像作为缺陷检测模型的输入，坏屏的缺陷检测结果作为缺陷检测模型的输出，以优化该缺陷检测模型，进而提高缺陷检测模型的检测准确度。
值得说明的是，本申请实施例并不限定在确定显示屏为坏屏时检测模型服务器可执行的上述一项或多项操作，其可根据实际情况进行确定，此处不再赘述。
可选的，对于显示屏生产线上部署的图像采集设备、控制台、服务器组、控制器、数据库等多个不同的设备，也可以将显示屏质量检测方法对应的操作步骤分散到上述多个不同的设备来执行。例如，图像采集设备采集显示屏图像，控制台根据负载均衡策略，将图像采集设备采集到的显示屏图像发送给服务器组中的检测模型服务器，由检测模型服务器上运行的缺陷检测模型对显示屏图像进行预设的预处理之后进行缺陷检测，并给出缺陷检测结果。检测模型服务器可以将缺陷检测结果发送给控制器，一方面由控制器结合实际业务场景，并根据业务需求，根据上述缺陷检测结果做出符合实际业务场景要求的响应，如报警、存储日志、控制生产控制指令等，另一方面，控制器还可以将缺陷检测结果及上述响应的处理行为作为日志存储到生产数据库中，以使训练模块根据训练数据库中的显示屏图像和缺陷检测结果更新上述得到的缺陷检测模型，该训练数据库中存储的是从生产数据库中筛选的具有缺陷的显示屏图像和对应的缺陷检测结果等数据。
值得说明的是，对于每一次优化的缺陷检测模型可通过小流量上线的方式逐步取代正在服务器上运行的缺陷检测模型，以达到缺陷检测模型随业务场景和生产阶段信息动态扩展泛化的目的。当本申请实施例中显示屏质量检测方法在显示屏生产线上运行一段时间后，可以人工通过生产数据库中的信息，复查上述缺陷检测和缺陷定位的准确率，然后更新上述训练数据库，重新训练缺陷检测模型，以提高缺陷检测准确率。
下述为本申请装置实施例，可以用于执行本申请方法实施例。对于本申请装置实施例中未披露的细节，请参照本申请方法实施例。
图6为本申请实施例提供的显示屏质量检测装置实施例的结构示意图一。如图6所示，本申请实施例提供的显示屏质量检测装置可以包括：接收模块51、处理模块52和确定模块53。
其中，该接收模块51，用于接收部署在显示屏生产线上的控制台发送的质量检测请求，所述质量检测请求中包含所述显示屏生产线上的图像采集设备采集的显示屏图像；
处理模块52，用于将所述显示屏图像输入到缺陷检测模型中得到缺陷检测结果，其中，所述缺陷检测模型是采用深度卷积神经网络结构和物体检测算法对历史缺陷显 示屏图像进行训练得到的；
确定模块53，用于根据所述缺陷检测结果确定所述显示屏图像对应的显示屏上的缺陷、所述缺陷对应的缺陷类别、以及所述缺陷对应的位置。
可选的，所述缺陷检测模型是采用深度卷积神经网络结构和物体检测算法对历史缺陷显示屏图像进行训练得到的，包括：
所述缺陷检测模型是对所述历史缺陷显示屏图像的候选区域损失、区域类别损失和区域边界损失进行组合训练，以使所述候选区域损失、所述区域类别损失和所述区域边界损失的总损失值满足预设损失阈值的结果；
其中，所述候选区域损失指所述历史缺陷显示屏图像中选定缺陷区域与实际缺陷区域之间的损失值，所述区域类别损失指所述选定缺陷区域中预测缺陷类别与实际缺陷类别之间的损失值，所述区域边界损失指所述选定缺陷区域中预测缺陷边界与实际缺陷边界之间的损失。
可选的，在本申请的一种可能实现方式中，所述处理模块52，还用于在将所述显示图像输入到缺陷检测模型中得到缺陷检测结果之前，对所述显示屏图像进行图像预处理。
其中，所述图像预处理包括下述处理中的一项或多项：
裁边、剪切、旋转、缩小、放大。
可选的，在本申请的另一种可能实现方式中，所述处理模块52，具体用于根据负载均衡策略，确定承载处理资源的检测模型服务器，将所述显示屏图像输入到运行在所述检测模型服务器上的所述缺陷检测模型中得到缺陷检测结果。
可选的，在本申请的再一种可能实现方式中，所述确定模块53，还用于根据生产阶段信息以及所述缺陷检测结果，确定所述显示屏图像对应的显示屏的质量好坏；所述生产阶段信息用于指示所述显示屏对应的生产厂家、生产环境、以及所述显示屏的类型。
可选的，在本申请的又一种可能实现方式中，所述处理模块52，还用于在所述确定模块53根据所述缺陷检测结果确定所述显示屏图像对应的显示屏的质量好坏之后，若确定所述显示屏为坏屏，则执行以下一项或多项操作：
通过控制器向生产管理者发送报警信息；
通过控制器将所述缺陷检测结果作为日志存储到生产数据库中；
通过控制器向所述控制台发送生产控制指令以便消除缺陷；
将所述显示屏图像和所述缺陷检测结果输入到所述缺陷检测模型中以便优化所述缺陷检测模型。
本实施例的显示屏质量检测装置可用于执行图2至图5所示方法实施例的实现方案，具体实现方式和技术效果类似，这里不再赘述。
图7为本申请实施例提供的显示屏质量检测装置实施例的结构示意图二，如图7所示，该显示屏质量检测装置可以为电子设备，该电子设备可以包括：存储器61和至少一个处理器62。
存储器61，用于存储程序指令。
处理器62，用于在程序指令被执行时实现如上述图2至图5所示方法实施例所示 的显示屏质量检测方法中的各个步骤，具体实现原理可参见上述实施例，本实施例此处不再赘述。
该电子设备还可以包括及输入/输出接口63。
输入/输出接口63可以包括独立的输出接口和输入接口，也可以为集成输入和输出的集成接口。其中，输出接口用于输出数据，输入接口用于获取输入的数据，上述输出的数据为上述方法实施例中输出的统称，输入的数据为上述方法实施例中输入的统称。
本申请还提供一种存储介质，所述存储介质中存储有指令，当其在计算机上运行时，使得计算机执行如图2至图5所示方法实施例的方法。
本申请还提供一种程序产品，该程序产品包括计算机程序，该计算机程序存储在存储介质中。显示屏质量检测装置的至少一个处理器可以从存储介质读取该计算机程序，至少一个处理器执行该计算机程序使得显示屏质量检测装置执行图2至图5所示方法实施例的方法。
本领域普通技术人员可以理解：实现上述各方法实施例的全部或部分步骤可以通过程序指令相关的硬件来完成。前述的程序可以存储于一计算机可读取存储介质中。该程序在执行时，执行包括上述各方法实施例的步骤；而前述的存储介质包括：ROM、RAM、磁碟或者光盘等各种可以存储程序代码的介质。
以上所述，仅为本申请的具体实施方式，但本申请的保护范围并不局限于此，任何熟悉本技术领域的技术人员在本申请揭露的技术范围内，可轻易想到变化或替换，都应涵盖在本申请的保护范围之内。因此，本申请的保护范围应以所述权利要求的保护范围为准。

Claims (14)

  1. 一种显示屏质量检测方法，其特征在于，包括：
    接收部署在显示屏生产线上的控制台发送的质量检测请求，所述质量检测请求中包含所述显示屏生产线上的图像采集设备采集的显示屏图像；
    将所述显示屏图像输入到缺陷检测模型中得到缺陷检测结果，其中，所述缺陷检测模型是对历史缺陷显示屏图像进行物体检测算法训练得到的；
    根据所述缺陷检测结果确定所述显示屏图像对应的显示屏上的缺陷、所述缺陷对应的缺陷类别、以及所述缺陷对应的位置。
  2. 根据权利要求1所述的方法，其特征在于，所述缺陷检测模型是采用深度卷积神经网络结构和物体检测算法对历史缺陷显示屏图像进行训练得到的，包括：
    所述缺陷检测模型是对所述历史缺陷显示屏图像的候选区域损失、区域类别损失、区域边界损失进行组合训练，以使所述候选区域损失、所述区域类别损失和所述区域边界损失的总损失值满足预设损失阈值的结果；
    其中，所述候选区域损失指所述历史缺陷显示屏图像中选定缺陷区域与实际缺陷区域之间的损失值，所述区域类别损失指所述选定缺陷区域中预测缺陷类别与实际缺陷类别之间的损失值，所述区域边界损失指所述选定缺陷区域中预测缺陷边界与实际缺陷边界之间的损失。
  3. 根据权利要求2所述的方法，其特征在于，所述将所述显示图像输入到缺陷检测模型中得到缺陷检测结果之前，还包括：
    对所述显示屏图像进行图像预处理，其中，所述图像预处理包括下述处理中的一项或多项：
    裁边、剪切、旋转、缩小、放大。
  4. 根据权利要求1-3任一项所述的方法，其特征在于，所述将所述显示屏图像输入到缺陷检测模型中得到缺陷检测结果，包括：
    根据负载均衡策略，确定承载处理资源的检测模型服务器；
    将所述显示屏图像输入到运行在所述检测模型服务器上的所述缺陷检测模型中得到缺陷检测结果。
  5. 根据权利要求1-3任一项所述的方法，其特征在于，所述方法还包括：
    根据生产阶段信息以及所述缺陷检测结果，确定所述显示屏图像对应的显示屏的质量好坏；所述生产阶段信息用于指示所述显示屏对应的生产厂家、生产环境、以及所述显示屏的类型。
  6. 根据权利要求4所述的方法，其特征在于，所述根据生产阶段信息以及所述缺陷检测结果，确定所述显示屏图像对应的显示屏的质量好坏之后，还包括：
    若确定所述显示屏为坏屏，则执行以下一项或多项操作：
    通过控制器向生产管理者发送报警信息；
    通过所述控制器将所述缺陷检测结果作为日志存储到生产数据库中；
    通过所述控制器向所述控制台发送生产控制指令以便消除缺陷；
    将所述显示屏图像和所述缺陷检测结果输入到所述缺陷检测模型中以便优化所述 缺陷检测模型。
  7. 一种显示屏质量检测装置，其特征在于，包括：
    接收模块，用于接收部署在显示屏生产线上的控制台发送的质量检测请求，所述质量检测请求中包含所述显示屏生产线上的图像采集设备采集的显示屏图像；
    处理模块，用于将所述显示屏图像输入到缺陷检测模型中得到缺陷检测结果，其中，所述缺陷检测模型是采用深度卷积神经网络结构和物体检测算法对历史缺陷显示屏图像进行训练得到的；
    确定模块，用于根据所述缺陷检测结果确定所述显示屏图像对应的显示屏上的缺陷、所述缺陷对应的缺陷类别、以及所述缺陷对应的位置。
  8. 根据权利要求7所述的装置，其特征在于，
    所述缺陷检测模型是采用深度卷积神经网络结构和实例分割算法对历史缺陷显示屏图像进行训练得到的，包括：
    所述缺陷检测模型是对所述历史缺陷显示屏图像的候选区域损失、区域类别损失和区域边界损失进行组合训练，以使所述候选区域损失、所述区域类别损失和所述区域边界损失的总损失值满足预设损失阈值的结果；
    其中，所述候选区域损失指所述历史缺陷显示屏图像中选定缺陷区域与实际缺陷区域之间的损失值，所述区域类别损失指所述选定缺陷区域中预测缺陷类别与实际缺陷类别之间的损失值，所述区域边界损失指所述选定缺陷区域中预测缺陷边界与实际缺陷边界之间的损失。
  9. 根据权利要求7所述的装置，其特征在于，
    所述处理模块，还用于在将所述显示图像输入到缺陷检测模型中得到缺陷检测结果之前，对所述显示屏图像进行图像预处理；
    其中，所述图像预处理包括下述处理中的一项或多项：
    裁边、剪切、旋转、缩小、放大。
  10. 根据权利要求7-9任一项所述的装置，其特征在于，
    所述处理模块，具体用于根据负载均衡策略，确定承载处理资源的检测模型服务器，将所述显示屏图像输入到运行在所述检测模型服务器上的所述缺陷检测模型中得到缺陷检测结果。
  11. 根据权利要求7-9任一项所述的装置，其特征在于，所述确定模块，还用于根据生产阶段信息以及所述缺陷检测结果，确定所述显示屏图像对应的显示屏的质量好坏；所述生产阶段信息用于指示所述显示屏对应的生产厂家、生产环境、以及所述显示屏的类型。
  12. 根据权利要求11所述的装置，其特征在于，
    所述处理模块，还用于在所述确定模块根据所述缺陷检测结果确定所述显示屏图像对应的显示屏的质量好坏之后，若确定所述显示屏为坏屏，则执行以下一项或多项操作：
    通过控制器向生产管理者发送报警信息；
    通过控制器将所述缺陷检测结果作为日志存储到生产数据库中；
    通过控制器向所述控制台发送生产控制指令以便消除缺陷；
    将所述显示屏图像和所述缺陷检测结果输入到所述缺陷检测模型中以便优化所述缺陷检测模型。
  13. 一种电子设备，其特征在于，包括：至少一个处理器和存储器；
    所述存储器存储计算机执行指令；
    所述至少一个处理器执行所述存储器存储的计算机执行指令，执行权利要求1-6任一项所述的方法。
  14. 一种存储介质，其特征在于，所述存储介质上存储有计算机执行指令，当所述计算机执行指令被处理器执行时，实现权利要求1-6任一项所述的方法。
PCT/CN2019/083164 2018-07-02 2019-04-18 显示屏质量检测方法、装置、电子设备及存储介质 Ceased WO2020007096A1 (zh)

Priority Applications (3)

Application Number Priority Date Filing Date Title
KR1020197034323A KR102229594B1 (ko) 2018-07-02 2019-04-18 디스플레이 스크린 품질 검출 방법, 장치, 전자기기 및 저장매체
JP2019563779A JP6921241B2 (ja) 2018-07-02 2019-04-18 ディスプレイスクリーン品質検査方法、装置、電子機器及び記憶媒体
US16/939,277 US11488294B2 (en) 2018-07-02 2020-07-27 Method for detecting display screen quality, apparatus, electronic device and storage medium

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201810710537.5 2018-07-02
CN201810710537.5A CN108846841A (zh) 2018-07-02 2018-07-02 显示屏质量检测方法、装置、电子设备及存储介质

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US16/939,277 Continuation US11488294B2 (en) 2018-07-02 2020-07-27 Method for detecting display screen quality, apparatus, electronic device and storage medium

Publications (1)

Publication Number Publication Date
WO2020007096A1 true WO2020007096A1 (zh) 2020-01-09

Family

ID=64200206

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/083164 Ceased WO2020007096A1 (zh) 2018-07-02 2019-04-18 显示屏质量检测方法、装置、电子设备及存储介质

Country Status (5)

Country Link
US (1) US11488294B2 (zh)
JP (1) JP6921241B2 (zh)
KR (1) KR102229594B1 (zh)
CN (1) CN108846841A (zh)
WO (1) WO2020007096A1 (zh)

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111709931A (zh) * 2020-06-16 2020-09-25 四川赛康智能科技股份有限公司 一种耐张线夹缺陷检测识别报告自动获取方法及系统
CN111798447A (zh) * 2020-07-18 2020-10-20 太原理工大学 一种基于Faster RCNN的深度学习塑化材料缺陷检测方法
CN112465758A (zh) * 2020-11-19 2021-03-09 河南省科学院应用物理研究所有限公司 一种铝基覆铜板质量的诊断方法
CN113592832A (zh) * 2021-08-05 2021-11-02 深圳职业技术学院 一种工业品缺陷检测方法和装置
CN113838015A (zh) * 2021-09-15 2021-12-24 上海电器科学研究所（集团）有限公司 一种基于网络协同的电器产品外观缺陷检测方法
CN114299012A (zh) * 2021-12-28 2022-04-08 以萨技术股份有限公司 一种基于卷积神经网络的物体表面缺陷检测方法及系统
CN116128740A (zh) * 2022-10-10 2023-05-16 深圳康荣电子有限公司 一种用于液晶显示屏的智能图像处理方法及系统
CN116912205A (zh) * 2023-07-13 2023-10-20 上海频准激光科技有限公司 一种基于神经网络模型的光纤熔接质量预测方法
CN116958046A (zh) * 2023-06-08 2023-10-27 西安电子科技大学 一种基于深度学习的屏幕Mura检测方法及设备

Families Citing this family (64)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10740913B2 (en) 2017-12-12 2020-08-11 Samsung Electronics Co., Ltd. Ultrafast, robust and efficient depth estimation for structured-light based 3D camera system
CN108846841A (zh) * 2018-07-02 2018-11-20 北京百度网讯科技有限公司 显示屏质量检测方法、装置、电子设备及存储介质
WO2020061736A1 (zh) * 2018-09-25 2020-04-02 西安诺瓦电子科技有限公司 显示设备故障检测方法、设备、系统及计算机可读介质
CN111077166A (zh) * 2018-10-19 2020-04-28 北京金山云网络技术有限公司 液晶屏的瑕疵检测方法、装置及终端设备
CN109544548A (zh) 2018-11-30 2019-03-29 北京百度网讯科技有限公司 餐盒的缺陷检测方法、装置、服务器、设备和存储介质
CN111275660B (zh) * 2018-12-05 2024-04-30 合肥欣奕华智能机器股份有限公司 一种平板显示器缺陷检测方法及装置
CN109492940A (zh) * 2018-12-06 2019-03-19 华中科技大学 一种用于造型浇注生产线的铸件炉包箱质量跟踪方法
KR102206753B1 (ko) * 2019-01-24 2021-01-22 주식회사 수아랩 결함 검사 장치
CN109887450B (zh) * 2019-03-01 2023-08-22 深圳市大像视讯科技有限公司 一种led显示屏人工智能检测方法
CN111666960B (zh) * 2019-03-06 2024-01-19 南京地平线机器人技术有限公司 图像识别方法、装置、电子设备及可读存储介质
CN109934814B (zh) * 2019-03-15 2022-02-01 英业达科技有限公司 表面缺陷侦测系统及其方法
CN109993734A (zh) * 2019-03-29 2019-07-09 北京百度网讯科技有限公司 用于输出信息的方法和装置
CN110533027B (zh) * 2019-07-22 2022-09-02 杭州未名信科科技有限公司 一种基于移动设备的文本检测和识别方法与系统
US11587233B2 (en) 2019-10-17 2023-02-21 International Business Machines Corporation Display panel defect enhancement
CN110675399A (zh) * 2019-10-28 2020-01-10 上海悦易网络信息技术有限公司 屏幕外观瑕疵检测方法及设备
CN110827244A (zh) * 2019-10-28 2020-02-21 上海悦易网络信息技术有限公司 一种用于电子设备外观瑕疵检测的方法与设备
CN110992278A (zh) * 2019-11-22 2020-04-10 成都数之联科技有限公司 一种用于解决样本不均衡问题的数据增强方法及系统
KR102390058B1 (ko) * 2020-04-24 2022-04-22 경기대학교 산학협력단 Led 패널 결함 검출을 위한 데이터 생성 장치 및 방법
US11900589B2 (en) 2020-05-29 2024-02-13 Boe Technology Group Co., Ltd. Detection device of display panel and detection method thereof, electronic device and readable medium
CN111693534B (zh) 2020-06-12 2023-09-15 北京百度网讯科技有限公司 表面缺陷的检测方法、模型训练方法、装置、设备和介质
US11790533B2 (en) * 2020-07-02 2023-10-17 Sony Group Corporation Machine learning based image segmentation training with contour accuracy evaluation
US20220130079A1 (en) * 2020-10-23 2022-04-28 Siemens Medical Solutions Usa, Inc. Systems and methods for simultaneous attenuation correction, scatter correction, and de-noising of low-dose pet images with a neural network
CN113313638A (zh) * 2020-12-23 2021-08-27 深圳市杰恩世智能科技有限公司 一种外观缺陷检测方法
CN112712504B (zh) * 2020-12-30 2023-08-15 广东粤云工业互联网创新科技有限公司 基于云端的工件检测方法及系统、计算机可读存储介质
CN112836724A (zh) * 2021-01-08 2021-05-25 重庆创通联智物联网有限公司 物体缺陷识别模型训练方法、装置、电子设备和存储介质
CN112734742B (zh) * 2021-01-19 2021-11-16 上海微亿智造科技有限公司 用于提高工业质检准确率的方法和系统
CN113176275B (zh) * 2021-03-19 2022-02-01 哈工大机器人(中山)无人装备与人工智能研究院 一种用于显示面板复检的方法、装置及系统
CN115147326A (zh) * 2021-03-29 2022-10-04 北京小米移动软件有限公司 色偏检测方法及装置、存储介质
US12614378B2 (en) * 2021-07-26 2026-04-28 PAIGE.AI, Inc. Systems and methods to process electronic images to determine histopathology quality
CN113569737B (zh) * 2021-07-28 2024-11-05 合肥综合性国家科学中心人工智能研究院(安徽省人工智能实验室) 基于自主学习网络模型笔记本屏幕缺陷检测方法及介质
CN113706477B (zh) * 2021-08-10 2024-02-13 南京旭锐软件科技有限公司 缺陷类别识别方法、装置、设备及介质
CN113824827A (zh) * 2021-09-20 2021-12-21 苏州凯仕弘科技有限公司 一种手机屏幕视觉检测系统
CN113780900B (zh) * 2021-11-09 2022-04-12 深圳市裕展精密科技有限公司 基于边缘计算的焊接检测系统及方法
CN114429445B (zh) * 2021-11-09 2025-05-23 南京师范大学 一种基于MAIRNet的PCB缺陷检测与识别方法
CN114331957B (zh) * 2021-11-17 2025-05-30 北京明略软件系统有限公司 纸张缺陷检测方法、装置及电子设备
KR20240093896A (ko) * 2021-11-25 2024-06-24 가부시키가이샤 재팬 디스프레이 표시 장치의 검사 방법 및 정보 처리 장치
CN114383530A (zh) * 2021-12-08 2022-04-22 深圳市德润赛尔光电有限公司 曲面屏弧度检测系统及方法
KR102933443B1 (ko) 2022-03-24 2026-03-05 (주)인포플라 인공지능 기반으로 텍스트 및 비-텍스트 이미지를 포함하는 화면 정보를 인지하여 화면 상의 오브젝트에 이벤트를 발생시키는 방법 및 시스템
KR102494867B1 (ko) * 2022-05-05 2023-02-06 주식회사 리드텍 머신러닝에 기반하여 블랙 스크린을 감지하는 비디오 월 시스템 및 그 동작방법
CN114677377A (zh) * 2022-05-30 2022-06-28 成都数之联科技股份有限公司 显示屏缺陷检测方法、训练方法、装置、设备及介质
CN114862832B (zh) * 2022-05-31 2025-03-21 北京中祥英科技有限公司 缺陷检测模型的优化方法、装置、设备及存储介质
CN115456975B (zh) * 2022-09-01 2026-04-24 昆明理工大学 基于改进Mask R-CNN的多背景下显示屏缺陷的实时分割方法
CN115457323A (zh) * 2022-09-15 2022-12-09 广东工业大学 基于视觉检测系统的手机玻璃屏非均衡表面缺陷分类方法
CN115601309A (zh) * 2022-09-23 2023-01-13 联想（北京）有限公司（Cn） 缺陷检测方法和装置
CN115761337A (zh) * 2022-11-18 2023-03-07 武汉海微科技有限公司 显示屏排线检测方法、装置、设备及存储介质
CN115760823B (zh) * 2022-11-28 2026-03-24 杭州海康机器人股份有限公司 一种缺陷检测方法、装置及电子设备
CN115937184B (zh) * 2022-12-28 2024-07-05 中轻长泰(长沙)智能科技股份有限公司 一种物品装配检测方法、装置及终端设备
KR20240116101A (ko) * 2023-01-20 2024-07-29 삼성전자주식회사 영상 프레임의 오류를 검출하기 위한 전자 장치 및 그 동작 방법
CN117764908B (zh) * 2023-08-17 2024-06-07 上海感图网络科技有限公司 Ng图像的缺陷信息显示方法、装置、设备及存储介质
CN117197413B (zh) * 2023-10-14 2026-03-03 武汉云片松科技有限公司 仪表黑屏的检测方法、系统、设备及存储介质
KR102823116B1 (ko) 2023-10-20 2025-06-25 국립한밭대학교 산학협력단 영상장치를 이용한 차세대 스마트 led 전광판의 불량픽셀 검출을 위한 딥러닝 시스템
CN118192110B (zh) * 2024-03-21 2024-11-05 深圳市裕融科技有限公司 车载液晶显示屏的缺陷检测方法及装置
CN118192112B (zh) * 2024-03-27 2024-11-08 深圳市爱富佳科技有限公司 基于ai的液晶屏故障检测方法、装置、设备及介质
CN121039711A (zh) * 2024-03-28 2025-11-28 京东方科技集团股份有限公司 流程控制方法及装置、计算设备、存储介质和程序产品
CN118505674B (zh) * 2024-06-19 2025-02-28 深圳市爱信新材料有限公司 一种基于视觉感知的显示屏缺陷检测方法及系统
CN118747988B (zh) * 2024-07-16 2025-06-10 江西肆海电子科技有限公司 液晶显示屏功能检测方法、装置、设备及存储介质
CN118505704B (zh) * 2024-07-18 2024-10-22 成都数之联科技股份有限公司 一种面板产线缺陷检测通用模型建模检测方法
CN118604005B (zh) * 2024-08-07 2024-11-15 远东幕墙(珠海)有限公司 基于3d扫描实现双曲幕墙的单元自动化检测方法及系统
CN118864434B (zh) * 2024-08-08 2025-11-18 唐山昆铁科技有限公司 钢轨表面缺陷检测方法、系统、电子设备及可读存储介质
CN119295367B (zh) * 2024-08-19 2025-06-27 北京奔驰汽车有限公司 一种生产线冲压件质量检测方法、装置及系统
CN118761677B (zh) * 2024-09-03 2025-02-07 广州飞达音响有限公司 一种音响生产缺陷智能测试方法及系统
CN120912513A (zh) * 2025-07-07 2025-11-07 衢州元森光电技术股份有限公司 基于ai模型的显示屏生产数据管理方法及系统
CN120912535B (zh) * 2025-07-22 2026-04-03 苏州市职业大学 基于图像处理的显示屏异常点检测方法及系统
CN120563501B (zh) * 2025-07-29 2025-11-07 深圳市信润富联数字科技有限公司 缺陷检测方法、装置、存储介质以及电子设备

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20090219521A1 (en) * 2008-02-29 2009-09-03 Lg Display Co., Ltd. Inspection apparatus for liquid crystal display device and inspection method using the same
US20150168293A1 (en) * 2013-12-18 2015-06-18 Tektronix, Inc. Apparatus and Method to Measure Display Quality
CN105044127A (zh) * 2015-07-31 2015-11-11 深圳市星火辉煌系统工程有限公司 一种oled微型显示器缺陷检测装置及检测方法
CN107944551A (zh) * 2017-12-15 2018-04-20 福州大学 一种用于电润湿显示屏缺陷识别方法
CN108846841A (zh) * 2018-07-02 2018-11-20 北京百度网讯科技有限公司 显示屏质量检测方法、装置、电子设备及存储介质

Family Cites Families (31)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH08247962A (ja) * 1995-03-10 1996-09-27 Sumitomo Chem Co Ltd カラーフィルタの欠陥検出方法及び欠陥検出装置並びにパネルディスプレイの欠陥検出方法及び欠陥検出装置
US6122397A (en) * 1997-07-03 2000-09-19 Tri Path Imaging, Inc. Method and apparatus for maskless semiconductor and liquid crystal display inspection
JP4155496B2 (ja) * 2002-04-25 2008-09-24 大日本スクリーン製造株式会社 分類支援装置、分類装置およびプログラム
JP2005088303A (ja) * 2003-09-16 2005-04-07 Fuji Photo Film Co Ltd 画像記録装置及び方法
JP2010159979A (ja) * 2009-01-06 2010-07-22 Hitachi Ltd 外観検査方法及び外観検査システム
US8995747B2 (en) * 2010-07-29 2015-03-31 Sharp Laboratories Of America, Inc. Methods, systems and apparatus for defect detection and classification
CN102854194B (zh) * 2012-09-11 2014-11-05 中南大学 基于线阵ccd的物件表面缺陷检测方法及装置
CN103913468B (zh) * 2014-03-31 2016-05-04 湖南大学 生产线上大尺寸lcd玻璃基板的多视觉缺陷检测设备及方法
EP3161635B1 (en) * 2014-06-30 2023-11-01 Amazon Technologies, Inc. Machine learning service
KR101688458B1 (ko) 2016-04-27 2016-12-23 디아이티 주식회사 깊은 신경망 학습 방법을 이용한 제조품용 영상 검사 장치 및 이를 이용한 제조품용 영상 검사 방법
US11210777B2 (en) * 2016-04-28 2021-12-28 Blancco Technology Group IP Oy System and method for detection of mobile device fault conditions
EP4410155A1 (en) * 2016-05-09 2024-08-07 Grabango Co. System and method for computer vision driven applications within an environment
KR101813223B1 (ko) * 2016-05-11 2018-01-30 한국과학기술원 영상 표면의 결함을 검출 및 분류하는 방법 및 장치
CN106228532B (zh) * 2016-07-05 2019-02-01 中科院成都信息技术股份有限公司 一种lcd屏幕亚像素级缺陷检测方法
US11580398B2 (en) * 2016-10-14 2023-02-14 KLA-Tenor Corp. Diagnostic systems and methods for deep learning models configured for semiconductor applications
JP6948128B2 (ja) * 2017-01-13 2021-10-13 キヤノン株式会社 映像監視装置及びその制御方法及びシステム
JP6809250B2 (ja) * 2017-01-23 2021-01-06 株式会社リコー 情報処理装置、情報処理方法およびプログラム
CN106952250B (zh) * 2017-02-28 2021-05-07 北京科技大学 一种基于Faster R-CNN网络的金属板带表面缺陷检测方法及装置
US11379688B2 (en) * 2017-03-16 2022-07-05 Packsize Llc Systems and methods for keypoint detection with convolutional neural networks
US20180322623A1 (en) * 2017-05-08 2018-11-08 Aquifi, Inc. Systems and methods for inspection and defect detection using 3-d scanning
CN107292870B (zh) * 2017-06-07 2020-11-20 复旦大学 基于图像对齐与检测网络模型的轨道塞钉故障检测方法与系统
CN107561738B (zh) * 2017-08-30 2020-06-12 湖南理工学院 基于fcn的tft-lcd表面缺陷快速检测方法
KR101838664B1 (ko) * 2017-11-28 2018-03-14 서울전선 주식회사 케이블 표면 검사 장치
JP6992475B2 (ja) * 2017-12-14 2022-01-13 オムロン株式会社 情報処理装置、識別システム、設定方法及びプログラム
CN108154504A (zh) * 2017-12-25 2018-06-12 浙江工业大学 一种基于卷积神经网络的钢板表面缺陷的检测方法
CN108108768B (zh) * 2017-12-29 2020-09-25 清华大学 基于卷积神经网络的光伏玻璃缺陷分类方法及装置
CN108230318B (zh) * 2018-01-09 2019-04-16 北京百度网讯科技有限公司 钢包缺陷检测分类方法、装置、设备及计算机可读介质
CN108230317A (zh) * 2018-01-09 2018-06-29 北京百度网讯科技有限公司 钢板缺陷检测分类方法、装置、设备及计算机可读介质
US11763188B2 (en) * 2018-05-03 2023-09-19 International Business Machines Corporation Layered stochastic anonymization of data
CN108961238A (zh) * 2018-07-02 2018-12-07 北京百度网讯科技有限公司 显示屏质量检测方法、装置、电子设备及存储介质
CN108921840A (zh) * 2018-07-02 2018-11-30 北京百度网讯科技有限公司 显示屏外围电路检测方法、装置、电子设备及存储介质

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20090219521A1 (en) * 2008-02-29 2009-09-03 Lg Display Co., Ltd. Inspection apparatus for liquid crystal display device and inspection method using the same
US20150168293A1 (en) * 2013-12-18 2015-06-18 Tektronix, Inc. Apparatus and Method to Measure Display Quality
CN105044127A (zh) * 2015-07-31 2015-11-11 深圳市星火辉煌系统工程有限公司 一种oled微型显示器缺陷检测装置及检测方法
CN107944551A (zh) * 2017-12-15 2018-04-20 福州大学 一种用于电润湿显示屏缺陷识别方法
CN108846841A (zh) * 2018-07-02 2018-11-20 北京百度网讯科技有限公司 显示屏质量检测方法、装置、电子设备及存储介质

Cited By (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111709931A (zh) * 2020-06-16 2020-09-25 四川赛康智能科技股份有限公司 一种耐张线夹缺陷检测识别报告自动获取方法及系统
CN111709931B (zh) * 2020-06-16 2023-05-23 四川赛康智能科技股份有限公司 一种耐张线夹缺陷检测识别报告自动获取方法及系统
CN111798447A (zh) * 2020-07-18 2020-10-20 太原理工大学 一种基于Faster RCNN的深度学习塑化材料缺陷检测方法
CN112465758B (zh) * 2020-11-19 2023-04-25 河南省科学院应用物理研究所有限公司 一种铝基覆铜板质量的诊断方法
CN112465758A (zh) * 2020-11-19 2021-03-09 河南省科学院应用物理研究所有限公司 一种铝基覆铜板质量的诊断方法
CN113592832A (zh) * 2021-08-05 2021-11-02 深圳职业技术学院 一种工业品缺陷检测方法和装置
CN113838015A (zh) * 2021-09-15 2021-12-24 上海电器科学研究所（集团）有限公司 一种基于网络协同的电器产品外观缺陷检测方法
CN113838015B (zh) * 2021-09-15 2023-09-22 上海电器科学研究所（集团）有限公司 一种基于网络协同的电器产品外观缺陷检测方法
CN114299012A (zh) * 2021-12-28 2022-04-08 以萨技术股份有限公司 一种基于卷积神经网络的物体表面缺陷检测方法及系统
CN116128740A (zh) * 2022-10-10 2023-05-16 深圳康荣电子有限公司 一种用于液晶显示屏的智能图像处理方法及系统
CN116128740B (zh) * 2022-10-10 2024-05-14 深圳康荣电子有限公司 一种用于液晶显示屏的智能图像处理方法及系统
CN116958046A (zh) * 2023-06-08 2023-10-27 西安电子科技大学 一种基于深度学习的屏幕Mura检测方法及设备
CN116912205A (zh) * 2023-07-13 2023-10-20 上海频准激光科技有限公司 一种基于神经网络模型的光纤熔接质量预测方法
CN116912205B (zh) * 2023-07-13 2024-03-08 上海频准激光科技有限公司 一种基于神经网络模型的光纤熔接质量预测方法

Also Published As

Publication number Publication date
US11488294B2 (en) 2022-11-01
JP6921241B2 (ja) 2021-08-18
CN108846841A (zh) 2018-11-20
KR102229594B1 (ko) 2021-03-17
US20200357109A1 (en) 2020-11-12
JP2020528996A (ja) 2020-10-01
KR20200004824A (ko) 2020-01-14

Similar Documents

Publication Publication Date Title
WO2020007096A1 (zh) 显示屏质量检测方法、装置、电子设备及存储介质
US11380232B2 (en) Display screen quality detection method, apparatus, electronic device and storage medium
US20200380899A1 (en) Method and apparatus for detecting peripheral circuit of display screen, electronic device, and storage medium
CN115131283B (zh) 目标对象的缺陷检测、模型训练方法、装置、设备及介质
US20200355627A1 (en) Method for detecting display screen peripheral circuit, apparatus, electronic device and storage medium
CN109084955A (zh) 显示屏质量检测方法、装置、电子设备及存储介质
CN109064446A (zh) 显示屏质量检测方法、装置、电子设备及存储介质
CN108921839A (zh) 连铸坯质量检测方法、装置、电子设备及存储介质
WO2021120186A1 (zh) 分布式产品缺陷分析系统、方法及计算机可读存储介质
CN112686833B (zh) 一种基于卷积神经网络的工业产品表面缺陷检测和分类装置
CN109949286A (zh) 用于输出信息的方法和装置
JP7446060B2 (ja) 情報処理装置、プログラム及び情報処理方法
CN109087281B (zh) 显示屏外围电路检测方法、装置、电子设备及存储介质
CN108961239A (zh) 连铸坯质量检测方法、装置、电子设备及存储介质
CN116385430A (zh) 一种机器视觉瑕疵检测方法、装置、介质及设备
CN113822842A (zh) 一种基于多任务学习的工业缺陷检测方法
CN117611549A (zh) 一种芯片焊接质量实时监测方法、系统、设备及存储介质
KR102937354B1 (ko) 비전 검사 시스템의 결함 심각도 판정 시스템
CN109087282A (zh) 显示屏外围电路检测方法、装置、电子设备及存储介质
CN121904801A (zh) 一种用于在养殖场孵化过程中进行监测的图像监测方法
WO2025199865A1 (zh) 流程控制方法及装置、计算设备、存储介质和程序产品
CN121686372A (zh) 一种机器人摄像头污染状态管理方法、装置及设备
CN118015454A (zh) 一种复合绝缘子检测方法及装置

Legal Events

Date Code Title Description
ENP Entry into the national phase

Ref document number: 2019563779

Country of ref document: JP

Kind code of ref document: A

ENP Entry into the national phase

Ref document number: 20197034323

Country of ref document: KR

Kind code of ref document: A

121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19830845

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19830845

Country of ref document: EP

Kind code of ref document: A1