CN110009622B - Display panel appearance defect detection network and defect detection method thereof - Google Patents
Display panel appearance defect detection network and defect detection method thereof Download PDFInfo
- Publication number
- CN110009622B CN110009622B CN201910270972.5A CN201910270972A CN110009622B CN 110009622 B CN110009622 B CN 110009622B CN 201910270972 A CN201910270972 A CN 201910270972A CN 110009622 B CN110009622 B CN 110009622B
- Authority
- CN
- China
- Prior art keywords
- detection
- modules
- defect detection
- layer
- module
- 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.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR 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; CALCULATING OR 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; CALCULATING OR 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
Abstract
The invention discloses a display panel appearance defect detection network and a defect detection method thereof, wherein the network comprises at least two feature extraction modules, at least two detection prediction output modules, an up-sampling module and an output layer, the feature extraction module is used for extracting features of an image to be processed, and outputting feature mapping maps of corresponding positions and depths, the up-sampling module is used for up-sampling feature mapping maps of different resolutions to feature mapping maps with consistent resolutions, and the detection prediction output module is used for outputting prediction results of corresponding scales after stacking and convolution dimension reduction are carried out on the received feature mapping maps; and outputting the detection result obtained by the output layer. The method not only reduces the requirement of the deep learning model on the number of training samples, but also greatly reduces the over-detection rate and the missing-detection rate of the defect detection of the display panel, and is more robust.
Description
Technical Field
The invention relates to the technical field of defect detection, in particular to a display panel appearance defect detection network and a defect detection method thereof.
Background
In the process of manufacturing the display panel, the quality of the display panel and the output result of the finished product grade are finally caused due to the defects of scratches, indentations, fragments, dust, stains and the like on the display panel. Therefore, during the manufacturing process of the display panel, it is important to perform defect detection on each component constituting the display panel and the surface of the display panel which is finally assembled successfully. At present, the detection of each component of the display panel and the display panel thereof mainly depends on human eye observation and detection by means of traditional image processing algorithms. The defect detection based on human eyes has strong subjective factors; furthermore, the human eyes can also have visual fatigue when being detected for a long time; the panel defect detection algorithm based on the traditional image processing algorithm has poor generalization capability, redundant parameters to be adjusted, low intelligent degree and high over-detection rate and omission factor, and the factors are superposed to cause the later maintenance cost of the display panel defect detection equipment to be very high, so that the panel defect detection algorithm cannot be applied in a large scale.
The inventor of the present application finds that the method of the prior art has at least the following technical problems in the process of implementing the present invention:
[1] constructing various feature extractors for extracting defect features according to defect features in a defect sample, training a classifier based on a machine learning algorithm based on feature vectors extracted by the feature extractors, and then matching the classifier with a sliding window algorithm to finish defect detection. The main problem of the algorithm is that different feature extraction methods need to be designed according to different defects, and once the shape and the area of the defect are greatly changed, effective defect detection cannot be carried out; in addition, the method also causes missed detection and over-detection aiming at weak defects;
[2] the method has the advantages that the defect detection algorithm developed based on the traditional image processing algorithm has the biggest problems of low generalization capability, excessive parameters needing to be adjusted and manual intervention in the detection process, so that the whole defect detection process cannot be completely automated, and a large amount of over-detection and omission detection can be caused due to numerous uncontrollable factors;
[3] and the quality inspectors perform manual inspection one by one. Thus, not only can a large amount of labor cost and time cost be brought, but also the efficiency is low; along with the increase of the eye fatigue degree of people, a large amount of missed detection and over detection of weak defects can be caused.
Therefore, the technical problem that the detection accuracy is not high exists in the prior art.
Disclosure of Invention
In view of the above, the present invention provides a display panel appearance defect detection network and a defect detection method thereof, so as to solve or at least partially solve the technical problem of low detection accuracy in the prior art.
The invention provides a first aspect of a display panel appearance defect detection network, comprising: at least two feature extraction modules, at least two detection prediction output modules, an up-sampling module and an output layer,
the characteristic extraction module is used for extracting the characteristics of the image to be processed and outputting a characteristic mapping chart corresponding to the position and the depth; the characteristic extraction modules are connected in a dense connection mode, each characteristic extraction module comprises four DBL modules, and the four DBL modules are connected in the dense connection mode;
the up-sampling module is used for up-sampling the feature mapping maps with different resolutions to the feature mapping maps with consistent resolutions and outputting the feature mapping maps to the detection prediction output module;
the detection prediction output module is used for outputting a prediction result of a corresponding scale after the received feature mapping is laminated and subjected to convolution dimensionality reduction;
and the output layer is used for obtaining and outputting the detection result according to the prediction result of the corresponding scale output by the detection prediction output module.
In one embodiment, the number of the feature extraction modules and the detection prediction output modules is six.
In one embodiment, each detection prediction output module corresponds to three target detection algorithms, wherein the target detection algorithms are used for predicting target detection boxes on the feature map.
In one embodiment, the detection prediction output module comprises: the system comprises a connecting layer and three DBL modules, wherein the connecting layer is used for stacking received feature maps, and the DBL modules are used for performing convolution dimensionality reduction on the stacked feature maps to obtain detection prediction outputs on different scales.
In one embodiment, the output layer is specifically configured to:
carrying out threshold filtering on a rectangular frame confidence coefficient threshold and a category probability threshold of a target detection frame in detection prediction output;
and carrying out non-maximum suppression operation on the result after threshold filtering to obtain the final detection output result of the whole detection network.
In one embodiment, each DBL module includes a convolutional layer, a BatchNorm layer, and an activation function ReLU, which are connected in sequence.
Based on the same inventive concept, the second aspect of the present invention provides a defect detection method, comprising:
and inputting the panel image to be processed into the defect detection network in the first aspect to obtain a defect detection result.
One or more technical solutions in the embodiments of the present application have at least one or more of the following technical effects:
the invention provides a display panel appearance defect detection network, which comprises at least two feature extraction modules, at least two detection prediction output modules, an up-sampling module and an output layer, wherein the feature extraction modules can be used for extracting features of an image to be processed and outputting feature mapping maps of corresponding positions and depths, short connection is carried out between the feature extraction modules in a dense connection mode, each feature extraction module comprises four DBL modules, and the four DBL modules are in short connection in the dense connection mode; the characteristic mapping maps with different resolutions are up-sampled to characteristic mapping maps with consistent resolutions through an up-sampling module; the received feature mapping image is subjected to stacking and convolution dimensionality reduction through a detection prediction output module, and a prediction result of a corresponding scale is output; and finally, obtaining a detection result through an output layer and outputting the detection result.
In the display panel appearance defect detection network provided by the invention, the four DBL modules of each feature extraction module are connected in a dense connection mode, and different feature extraction modules are also connected in a dense connection mode. In other words, besides the conventional sequential connection, the features extracted by the convolution of the layer are densely transmitted to the next non-adjacent DBL module through the short connection, so that the features extracted from the sample by the DBL module can be better utilized, the requirement of the network on the sample amount is reduced, and the accuracy of the detection network can be improved.
The defect detection network also performs multi-scale connection prediction, and comprises at least two detection prediction output modules, wherein different detection prediction output modules are respectively input from corresponding feature extraction modules, and the feature extraction modules respectively output different positions and different network depths of the network, so that each detection prediction output module can obtain detection output on one scale, the problems of small samples, large size and large detection defect scale span in the appearance defect detection of the display panel are solved, and the problem that the later maintenance cost of the display equipment is too high to be applied in a large scale due to parameter redundancy and poor robustness in the traditional image processing algorithm is solved.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to these drawings without creative efforts.
FIG. 1 is a schematic view of a defect scale span of a defect image;
fig. 2 is a schematic structural diagram of a detection network DCDDNet constructed in a specific example;
fig. 3 is a schematic diagram illustrating a connection manner of a mainstream network in the prior art;
FIG. 4 is a schematic diagram illustrating a connection method of a densely connected neural network according to an embodiment of the present invention;
fig. 5 is a schematic diagram of the structure of each component of the detection network DCDDNet according to the embodiment of the present invention.
Detailed Description
The invention mainly aims to provide a stable and high-generalization-capability deep learning detection network model which can effectively detect defects in an appearance defect image of a display panel; meanwhile, the problem that a mainstream deep learning detection network model cannot be well detected and detected in real time due to the fact that a small sample training data set, the resolution of an image to be detected is too large, and the defect scale span is too large in the field of appearance defect detection of the display panel is solved.
The method does not need to modify the structure of the current display panel defect detection system, does not increase any hardware cost, is simple and effective, and has the characteristics of easy realization, low cost and high practicability.
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The inventor of the application discovers through a large amount of research and practice that deep learning, particularly a convolutional neural network in deep learning, has great success in the fields of image recognition, target detection, image semantic segmentation, instance semantic segmentation and the like in succession since 12 months in 2012. In order to enable the generalization capability of the panel defect detection algorithm to defects to be stronger, the over-missing detection rate to be lower, the display panel defect detection system to be more intelligent and the later maintenance cost of related equipment to be lower, the invention creatively introduces the deep learning algorithm into the field of panel defect detection. However, the mainstream deep learning algorithm cannot be directly applied to the industrial display panel defect detection. The main reasons are as follows:
[1] small samples: the display panel defect detection algorithm based on the deep learning algorithm needs a large number of defect images as a training data set, and in the field of industrial display panel defect detection, it is extremely difficult to obtain a defect sample data set with a certain scale; therefore, there are very few training samples that can be used for classification or detection of display panel defects;
[2] large size: the resolution of the industrial panel defect image is generally a large-size image like 6480 pixels × 3840 pixels, so that the mainstream target detection network is directly used for the appearance defect detection of the display panel, and the forward propagation prediction process of a single image is a very time-consuming process and cannot meet the requirement of real-time detection in industrial detection;
[3] defect multi-scale: the scale change of the defects in the appearance defect image of the display panel is very large, in the appearance defect image with 1024 pixels by 1024 pixels resolution, the small defect scale can be as small as 4 pixels by 4 pixels, and the large defect scale can reach 1000 pixels by 1000 pixels, as shown in fig. 1, the defects to be detected are in the black rectangular frame, and have very large defect scale change, and the mainstream target detection network model cannot be well detected;
based on the consideration, the invention provides a display panel appearance defect detection network based on dense connection mode multi-scale prediction, which can solve the problems of small samples, large size and large detection defect scale span in display panel appearance defect detection, and also can solve the problem that the traditional image processing algorithm cannot be applied in a large scale due to overhigh later maintenance cost of display equipment caused by parameter redundancy and poor robustness.
Example one
The embodiment provides a display panel appearance defect detection network, which comprises: at least two feature extraction modules, at least two detection prediction output modules, an up-sampling module and an output layer,
the characteristic extraction module is used for extracting the characteristics of the image to be processed and outputting a characteristic mapping chart corresponding to the position and the depth; the characteristic extraction modules are connected in a dense connection mode, each characteristic extraction module comprises four DBL modules, and the four DBL modules are connected in the dense connection mode;
the up-sampling module is used for up-sampling the feature mapping maps with different resolutions to the feature mapping maps with consistent resolutions and outputting the feature mapping maps to the detection prediction output module;
the detection prediction output module is used for outputting a prediction result of a corresponding scale after the received feature mapping is laminated and subjected to convolution dimensionality reduction;
and the output layer is used for obtaining and outputting the detection result according to the prediction result of the corresponding scale output by the detection prediction output module.
Specifically, the number of the feature extraction module and the detection prediction output module may be set according to actual conditions, and the feature extraction module is composed of four DBL modules and used for feature extraction. The detection prediction output module takes the feature mapping image transmitted by the feature extraction module as input and respectively outputs prediction results on different scales.
In a specific implementation process, four DBL modules are connected in a dense connection mode; referring to fig. 3, the 3 rd DBL layer receives the output of the first DBL layer in addition to the output of the second DBL layer, and simultaneously the output of the 4 th DBL layer is output to the 4 th DBL layer and is also output to the OUT layer. Similarly, the feature extraction modules are also connected in a dense connection manner, please refer to fig. 2, and the feature extraction module is a DenseBlock _4 module.
That is to say, the dense connection in the DCDDNet detection network model proposed in the present invention not only means that the DBL modules are connected in a dense connection manner, but also each feature extraction module: the DenseBlock _4 modules are connected in a dense connection mode, so that the multiplexing rate of sample information can be improved, and the requirement on the number of training samples is reduced. According to the invention, different detection networks can be constructed based on the defect target detection field in a dense connection mode;
generally speaking, the main innovation point of the invention is to construct a display panel appearance defect detection network model based on dense connection mode multi-scale prediction, and to name the detection network as DCDDNet, the detection network model mainly comprises the following two improvements compared with the mainstream target detection network model:
[1] dense connection: in the field of actual display panel appearance detection, it is difficult to obtain an appearance defect image data set with a certain scale as a training data set of a depth detection learning model, however, if not enough training data sets exist, the finally trained defect detection model can not well detect the appearance defects in the defect image, in order to solve the problem, the invention introduces the concept of dense connection into the detection network model when constructing the detection network model, the dense connection network can extract the available information in the training sample to the maximum extent, and can reduce the loss of the extracted characteristic information to the maximum extent in the process of network transmission by a dense connection mode, therefore, the utilization rate of the training sample information is improved invisibly, and the dependence of the whole network model on the training sample is reduced.
At present, the connection mode of the mainstream deep learning detection network model is shown in fig. 3, and the network structure diagram of the dense connection mode is shown in fig. 4. In fig. 3 and 4 denotes a network input layer, Out denotes a network output layer, and DBL denotes a convolution module. As can be seen from the comparison between fig. 3 and fig. 4, in addition to the conventional sequential connection, the convolution modules in the densely-connected network also densely transfer the features extracted by convolution of the layer to the non-adjacent volume base layer via short connection in this way, so that the features extracted from the sample by each convolution layer can be better utilized, and the requirement of the network for the sample size is reduced.
In an implementation manner, in the defect detection network provided by the embodiment of the present invention, the number of the feature extraction module and the number of the detection prediction output modules are both six.
In one embodiment, each detection prediction output module corresponds to three target detection algorithms, wherein the target detection algorithms are used for predicting target detection boxes on the feature map.
Specifically, please refer to fig. 2, which is a diagram illustrating a network structure for DCDDNet detection. In the figure, a Dec block denotes a detection prediction output block, a DenseBlock _4 block denotes a feature extraction block, UP denotes an UP-sampling block, and OUT denotes an output layer.
Each Dec module is detection output on one scale, the DCDDNet detection network comprises 6 detection prediction output modules, and the inputs of the six Dec modules are respectively from 6 DenseBlock _4 modules of the DCDDNet detection network feature extraction part. And the 6 DenseBlock _4 modules output different positions of the network and different network depths respectively.
[2] Multi-scale connection prediction: there are 3 anchors in each Dec module. The Anchor is an algorithm in the field of target detection, and the main function of the Anchor is to predict a target detection frame Bounding Box on an image of an input layer or a final feature mapping image obtained by the whole feature extraction part. Therefore, the defect detection network provided by the embodiment needs to set 18 Anchor aspect ratios in total. Constructing a multi-scale prediction detection model containing 6 different network depths, and using an Anchor mechanism with 18 different scale proportions, wherein the main purpose is to solve the problem that the defect scale span is too large in the defect detection of the display panel, and the characteristic information of the target to be detected on the image can be captured on scales with different sizes by using multi-scale prediction; meanwhile, the depth of the detection network model is compressed to the maximum extent by the network construction mode, and information multiplexing of the feature extraction layer is increased, so that the consumption of single forward propagation time of a single image is reduced, and the requirement of the whole detection model on the number of training samples in the model training stage can be reduced.
In an implementation manner, in the defect detection network provided in an embodiment of the present invention, the detection prediction output module includes: the system comprises a connecting layer and three DBL modules, wherein the connecting layer is used for stacking received feature maps, and the DBL modules are used for performing convolution dimensionality reduction on the stacked feature maps to obtain detection prediction outputs on different scales.
In an implementation manner, in the defect detection network provided in the embodiment of the present invention, the output layer is specifically configured to:
carrying out threshold filtering on a rectangular frame confidence coefficient threshold and a category probability threshold of a target detection frame in detection prediction output;
and carrying out non-maximum suppression operation on the result after threshold filtering to obtain the final detection output result of the whole detection network.
In one embodiment, in the defect detection network provided by the embodiment of the present invention, each DBL module includes a convolutional layer, a BatchNorm layer, and an activation function ReLU, and the convolutional layer, the BatchNorm layer, and the activation function ReLU are sequentially connected.
In particular, the main role of convolutional layers is to extract local features in images; the BatchNorm layer, generally called as a regularization layer, is mainly used for solving the problems that overfitting, gradient disappearance and local optimal solution entering of a mode network occur; the ReLU is an activation function, mainly to increase the non-linear prediction of the whole network, which will greatly improve the ability of the network to solve responsible problems.
Please refer to fig. 5, wherein the DBL is a basic convolution feature extraction module, and the DBL module is formed by sequentially connecting a convolution layer, a BatchNorm layer, and a ReLU layer; the Dec module is formed by connecting a connection layer Cont and three DBL modules in sequence; the DenseBlock _4 module is formed by connecting 4 DBLs in a dense connection mode; the Up module in fig. 5 is an Up-sampling layer, and its main function is to Up-sample feature maps of different resolutions into feature maps of consistent resolution, and then send these feature maps into Dec. The input of the Up layer is from two places, namely the output of the Dec and different DenseBlock _4 modules, and the number of the detection network Up modules is 5.
The Dec module adds the feature mapping layers by using the connection layers Cont, and then performs convolution dimensionality reduction through the three convolution layers so as to obtain detection prediction outputs on different scales.
For an image to be detected, prediction results on respective scales are given by the prediction modules on the six scales, the results are input into the Out layer, threshold filtering is carried Out on rectangular frame confidence coefficient thresholds and category probability thresholds in target detection frames in the results in the Out layer, and the results after the threshold filtering are finally filtered by the NMS module, namely the final detection output result of the whole detection network.
It should be noted that, unless otherwise specified, specific meanings of english abbreviations referred to in the present invention are shown in table 2.
Table 2 term interpretation
Generally speaking, the main objective of the present invention is to provide a stable, highly generalized and deep learning detection network model capable of effectively detecting defects in the appearance defect image of the display panel; meanwhile, the problem that a mainstream deep learning detection network model cannot be well detected and detected in real time due to the fact that a small sample training data set, the resolution of an image to be detected is too large, and the defect scale span is too large in the field of appearance defect detection of the display panel is solved. The advantages of this detection network or algorithm are as follows:
1. the current AOI structure does not need to be modified, the hardware cost is not increased,
2. compared with the existing deep learning network model, the requirement of the deep learning model on the number of training samples is reduced, and fewer convolution layers are used in the detection network, so that the forward propagation speed is higher, and the requirement of the display panel appearance defect detection field on defect real-time detection is met.
3. Compared with the traditional image processing algorithm, the method not only greatly reduces the over-detection rate and the omission rate of the defect detection of the display panel, but also is more robust.
4. The display panel appearance defect detection system constructed based on the detection network can greatly promote the automation and the intellectualization of the display panel appearance defect detection system, improve the performance of AOI detection equipment and reduce the human input of manual detection.
Based on the same inventive concept, the application also provides a defect detection method based on the defect detection network in the first embodiment, which is detailed in the second embodiment.
Example two
The embodiment provides a defect detection method, which comprises the following steps:
and inputting the panel image to be processed into the defect detection network to obtain a defect detection result.
Since the defect detection method introduced in the second embodiment of the present invention is implemented based on the defect detection network in the first embodiment of the present invention, a person skilled in the art can understand the specific implementation of the method based on the defect detection network introduced in the first embodiment of the present invention, and thus details are not described herein again. All the methods implemented by the defect detection network given to the first embodiment of the present invention belong to the protection scope of the present invention.
As will be appreciated by one skilled in the art, embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
While preferred embodiments of the present invention have been described, additional variations and modifications in those embodiments may occur to those skilled in the art once they learn of the basic inventive concepts. Therefore, it is intended that the appended claims be interpreted as including preferred embodiments and all such alterations and modifications as fall within the scope of the invention.
It will be apparent to those skilled in the art that various modifications and variations can be made in the embodiments of the present invention without departing from the spirit or scope of the embodiments of the invention. Thus, if such modifications and variations of the embodiments of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to encompass such modifications and variations.
Claims (7)
1. A display panel appearance defect detection network, comprising: at least two feature extraction modules, at least two detection prediction output modules, an up-sampling module and an output layer,
the characteristic extraction module is used for extracting the characteristics of the image to be processed and outputting a characteristic mapping chart corresponding to the position and the depth; the characteristic extraction modules are connected in a dense connection mode, each characteristic extraction module comprises four DBL modules, the four DBL modules are connected in the dense connection mode, each DBL module is formed by sequentially connecting a convolution layer, a Batchnorm layer and a ReLU layer, and the dense connection comprises sequential connection between adjacent modules and connection between non-adjacent modules;
the up-sampling module is used for up-sampling the feature mapping maps with different resolutions to the feature mapping maps with consistent resolutions and outputting the feature mapping maps to the detection prediction output module;
the detection prediction output module is used for outputting a prediction result of a corresponding scale after the received feature mapping is laminated and subjected to convolution dimensionality reduction;
and the output layer is used for obtaining and outputting the detection result according to the prediction result of the corresponding scale output by the detection prediction output module.
2. The defect detection network of claim 1 wherein the number of feature extraction modules and detection prediction output modules is six.
3. The defect detection network of claim 1, wherein each detection prediction output module corresponds to three target detection algorithms, wherein the target detection algorithms are configured to perform prediction of target detection boxes on the feature map.
4. The defect detection network of claim 1, wherein the detection prediction output module comprises: the system comprises a connecting layer and three DBL modules, wherein the connecting layer is used for stacking received feature maps, and the DBL modules are used for performing convolution dimensionality reduction on the stacked feature maps to obtain detection prediction outputs on different scales.
5. The defect detection network of claim 1, wherein the output layer is specifically configured to:
carrying out threshold filtering on a rectangular frame confidence coefficient threshold and a category probability threshold of a target detection frame in detection prediction output;
and carrying out non-maximum suppression operation on the result after threshold filtering to obtain the final detection output result of the whole detection network.
6. The defect detection network of claim 1 or 4, wherein each DBL module comprises a convolutional layer, a BatchNorm layer, and an activation function ReLU, the convolutional layer, the BatchNorm layer, and the activation function ReLU being connected in sequence.
7. A method of defect detection, comprising:
inputting a panel image to be processed into the defect detection network according to any one of claims 1 to 6, and obtaining a defect detection result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910270972.5A CN110009622B (en) | 2019-04-04 | 2019-04-04 | Display panel appearance defect detection network and defect detection method thereof |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910270972.5A CN110009622B (en) | 2019-04-04 | 2019-04-04 | Display panel appearance defect detection network and defect detection method thereof |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110009622A CN110009622A (en) | 2019-07-12 |
| CN110009622B true CN110009622B (en) | 2022-02-01 |
Family
ID=67169990
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910270972.5A Active CN110009622B (en) | 2019-04-04 | 2019-04-04 | Display panel appearance defect detection network and defect detection method thereof |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110009622B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110599455A (en) * | 2019-08-13 | 2019-12-20 | 武汉精立电子技术有限公司 | Display screen defect detection network model, method and device, electronic equipment and storage medium |
| CN110490875B (en) * | 2019-10-16 | 2020-01-21 | 武汉精立电子技术有限公司 | Screen defect filtering method and device and storage medium |
| CN111539456B (en) * | 2020-04-02 | 2024-03-01 | 浙江华睿科技股份有限公司 | Target identification method and device |
| CN111968093A (en) * | 2020-08-19 | 2020-11-20 | 创新奇智(上海)科技有限公司 | Magnetic shoe surface defect detection method and device, electronic equipment and storage medium |
| CN113409290B (en) * | 2021-06-29 | 2023-12-15 | 北京兆维电子(集团)有限责任公司 | Method and device for detecting appearance defects of liquid crystal display, and storage medium |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109165550A (en) * | 2018-07-13 | 2019-01-08 | 首都师范大学 | A kind of multi-modal operation track fast partition method based on unsupervised deep learning |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9380273B1 (en) * | 2009-10-02 | 2016-06-28 | Rockwell Collins, Inc. | Multiple aperture video image enhancement system |
| CN109214505B (en) * | 2018-08-29 | 2022-07-01 | 中山大学 | Full convolution target detection method of densely connected convolution neural network |
| CN109167928B (en) * | 2018-09-06 | 2020-11-03 | 武汉精测电子集团股份有限公司 | Rapid automatic exposure method and system based on display panel defect detection |
| CN109559301A (en) * | 2018-11-20 | 2019-04-02 | 江苏第二师范学院（江苏省教育科学研究院） | A kind of EMU end to end enters the twin network method of institute's defects detection |
-
2019
- 2019-04-04 CN CN201910270972.5A patent/CN110009622B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109165550A (en) * | 2018-07-13 | 2019-01-08 | 首都师范大学 | A kind of multi-modal operation track fast partition method based on unsupervised deep learning |
Non-Patent Citations (2)
| Title |
|---|
| yolov3;永远学习中;《https://www.jianshu.com/p/ee032282440f》;20181224;第1-8页 * |
| 基于深度学习的连铸板坯表面缺陷检测系统;毛欣翔 等;《工业控制计算机》;20190331;第32卷(第3期);第66-68页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110009622A (en) | 2019-07-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110009622B (en) | Display panel appearance defect detection network and defect detection method thereof | |
| CN110503112B (en) | Small target detection and identification method for enhancing feature learning | |
| CN112884064B (en) | Target detection and identification method based on neural network | |
| CN108460764B (en) | Ultrasonic image intelligent segmentation method based on automatic context and data enhancement | |
| CN110852316B (en) | Image tampering detection and positioning method adopting convolution network with dense structure | |
| CN112967243A (en) | Deep learning chip packaging crack defect detection method based on YOLO | |
| CN113591968A (en) | Infrared weak and small target detection method based on asymmetric attention feature fusion | |
| CN111626993A (en) | Image automatic detection counting method and system based on embedded FEFnet network | |
| CN105931246A (en) | Fabric flaw detection method based on wavelet transformation and genetic algorithm | |
| CN111382785A (en) | GAN network model and method for realizing automatic cleaning and auxiliary marking of sample | |
| CN114120093B (en) | Coal gangue target detection method based on improved YOLOv algorithm | |
| CN110020658B (en) | Salient object detection method based on multitask deep learning | |
| CN110599458A (en) | Underground pipe network detection and evaluation cloud system based on convolutional neural network | |
| CN115830004A (en) | Surface defect detection method, device, computer equipment and storage medium | |
| Zhu et al. | Towards automatic wild animal detection in low quality camera-trap images using two-channeled perceiving residual pyramid networks | |
| CN116485709A (en) | Bridge concrete crack detection method based on YOLOv5 improved algorithm | |
| CN115147418A (en) | Compression training method and device for defect detection model | |
| CN116071315A (en) | Product visual defect detection method and system based on machine vision | |
| CN115240259A (en) | Face detection method and face detection system based on YOLO deep network in classroom environment | |
| CN110033443B (en) | Display panel defect detection method | |
| CN111767919B (en) | Multilayer bidirectional feature extraction and fusion target detection method | |
| CN116740572A (en) | Marine vessel target detection method and system based on improved YOLOX | |
| CN115775226A (en) | Transformer-based medical image classification method | |
| CN111539434B (en) | Infrared weak and small target detection method based on similarity | |
| CN112446292B (en) | 2D image salient object detection method and system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |