CN110751170A - Panel quality detection method, system, terminal device and computer readable medium - Google Patents
Panel quality detection method, system, terminal device and computer readable medium Download PDFInfo
- Publication number
- CN110751170A CN110751170A CN201910843194.4A CN201910843194A CN110751170A CN 110751170 A CN110751170 A CN 110751170A CN 201910843194 A CN201910843194 A CN 201910843194A CN 110751170 A CN110751170 A CN 110751170A
- Authority
- CN
- China
- Prior art keywords
- panel
- quality
- loss function
- value
- probability
- 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.)
- Pending
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 46
- 230000006870 function Effects 0.000 claims abstract description 73
- 238000009826 distribution Methods 0.000 claims abstract description 70
- 238000013136 deep learning model Methods 0.000 claims abstract description 37
- 238000012549 training Methods 0.000 claims abstract description 18
- 238000000034 method Methods 0.000 claims abstract description 16
- 238000004364 calculation method Methods 0.000 claims description 24
- 238000004590 computer program Methods 0.000 claims description 8
- 238000012545 processing Methods 0.000 claims description 7
- 238000007689 inspection Methods 0.000 claims description 5
- 238000003860 storage Methods 0.000 claims description 3
- 238000013135 deep learning Methods 0.000 abstract description 5
- 230000007547 defect Effects 0.000 description 26
- 238000004519 manufacturing process Methods 0.000 description 6
- 238000013527 convolutional neural network Methods 0.000 description 5
- 238000012360 testing method Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 4
- 238000012795 verification Methods 0.000 description 4
- 239000000463 material Substances 0.000 description 3
- 238000013441 quality evaluation Methods 0.000 description 3
- 238000013459 approach Methods 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000007635 classification algorithm Methods 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 230000011218 segmentation Effects 0.000 description 2
- 239000000126 substance Substances 0.000 description 2
- 230000000007 visual effect Effects 0.000 description 2
- 208000003464 asthenopia Diseases 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 239000004973 liquid crystal related substance Substances 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- 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
- 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; 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
Abstract
The invention discloses a panel quality detection method, a system, a terminal device and a computer readable medium, wherein the method comprises the following steps: calculating to obtain a predicted grade value of the panel image according to the probability value of the panel image belonging to different quality grades, and generating a first loss function according to the predicted grade value and the real label value of the panel image; fitting the probability value with a normal distribution curve to obtain prediction grade probability distribution; obtaining a second loss function according to the prediction grade probability distribution and the real probability distribution corresponding to the real label value; weighting the first loss function and the second loss function to obtain a final loss function of the deep learning model, and training the deep learning model based on the final loss function; detecting the panel image to be detected by using the trained deep learning model; according to the invention, the accuracy of the quality grade predicted value of the deep learning model is improved by learning the distribution of the prediction grade probability and utilizing the correlation among different grades in the deep learning level.
Description
Technical Field
The invention belongs to the technical field of panel automatic defect detection, and particularly relates to a panel quality detection method and system based on distributed learning, terminal equipment and a computer readable medium.
Background
With the popularization and rapid update of mobile phones and consumer electronics, the liquid crystal display and the OLED display have great output requirements, and the manufacturing process and the detection technology of the panel are rapidly developed. Although the manufacturing process of the panel is promoted year by year, the appearance of the visual defect cannot be completely avoided due to the complicated structure and the various manufacturing processes. Defects can be roughly classified into three types according to their shapes: point defects, line defects, and area defects. Among the three defects, a point defect is the most common defect, which is divided into a bright point and a dark point, and the size of the defect is one pixel; the line defects are also divided into bright lines and dark lines, the size of the defects is a line segment with the width of a whole pixel, and the point defects and the line defects are both electrical defects. The Mura defect is a common surface defect, belongs to a non-electrical defect and is one of the most difficult defects to detect in all panel display defects.
The term "Mura" is from Japanese, meaning spots and stains, and Mura defect is a macroscopic display defect with characteristics of indefinite size, indefinite shape, uneven gray distribution, blurred edge, and the like. Generally, the Mura defect is caused by a combination of material overlapping, material unevenness, processing environment variation, and poor quality of the material itself. In the actual production process of the panel, the Mura defect may be generated due to many factors of the manufacturing process including the workshop environment and the like.
In the actual production process of the panel, panel manufacturers can sell the panel in a grading way according to the severity of the Mura defect, but at present, the panel grade is judged by training a batch of detection personnel, and the detection personnel perform visual judgment and grading according to experience. On one hand, human eye detection has subjectivity, so that complete consistency of detection standards cannot be guaranteed among different detection personnel; on the other hand, the detection mode has low efficiency and high cost, the time for detecting a 2.4-inch LCD image generally takes even forty seconds, and the detection personnel can have visual fatigue after carrying out long-term detection work and even can hurt the eyes of people. Therefore, a more efficient and more standardized panel quality evaluation method is pursued.
Deep learning, particularly convolutional neural networks in deep learning, has been successful in the fields of image classification, target detection, image semantic segmentation, instance segmentation and the like in succession since 12 months 2012. At present, the evaluation of panel picture quality by using a convolutional neural network is a new field, and a method generally adopted is to treat panel picture quality detection as a classification problem, or extract features through the convolutional neural network, simply weight class probabilities output by the last layer of network and labels, and implement a simple scoring algorithm. However, in the task of evaluating the quality of the panel pictures, the picture grades are ordered, and people expect that similar panel pictures can obtain approximate prediction grade values, which reflects the difference between the problem of evaluating the quality of the pictures and the general classification problem; the current multi-classification algorithm cannot describe the association between different classes (for example, the panel quality is 1-10 levels, and the association between 8 levels and 7 levels is obviously stronger than that between 8 levels and 2 levels); if the current multi-classification algorithm is adopted for panel quality detection, the 8-level panel may be wrongly classified into 2 levels instead of 7 levels, and thus, the accuracy of the convolutional neural network-based display panel detection method is to be improved.
Disclosure of Invention
In view of at least one of the defects or the improvement requirements of the prior art, the present invention provides a panel quality detection method, a system, a terminal device and a computer readable medium, which aims to solve the problem that the detection accuracy of the existing panel quality detection method based on deep learning needs to be improved.
To achieve the above object, according to a first aspect of the present invention, there is provided a panel quality detection method based on distribution learning, including the steps of:
inputting panel images with labels of different quality grades into a deep learning model to train the panel images; calculating to obtain a predicted grade value of the panel image according to a probability value of the panel image predicted by the deep learning model and belonging to different quality grades; generating a first loss function according to the prediction grade value and the real label value of the panel image;
fitting the probability values of the panel images belonging to different quality grades with a normal distribution curve to obtain prediction grade probability distribution; obtaining a second loss function according to the prediction grade probability distribution and the real probability distribution corresponding to the real label value;
weighting the first loss function and the second loss function to obtain a final loss function of the deep learning model, and training the deep learning model until the final loss function is minimized;
and detecting the panel image to be detected by using the trained deep learning model to obtain the quality grade of the panel image to be detected.
Preferably, in the panel quality detection method, the probability values of the panel images predicted by the deep learning model and belonging to different quality levels are as follows: dividing the characteristic data of each quality grade by the sum of the characteristic data of all the quality grades;
wherein K represents a maximum quality level; i is a natural number from 1 to K; p is a radical ofiA probability value representing an ith quality level;characteristic data representing the ith quality level.
Preferably, in the panel quality detection method, the method for calculating the predicted gradation value of the panel image includes: the sum of the products of the probability values of the quality classes and the corresponding quality class values;
wherein K represents a maximum quality level; i is a natural number from 1 to K; label represents a predicted rank value of the panel image; p is a radical ofiRepresenting the probability value of the i-th quality level.
Preferably, in the panel quality detecting method, the first loss function is: dividing the sum of the differences between the real label values of different panel images and the predicted grade values of the real label values by the number of the panel images to obtain an average value;
wherein L is1Representing a first loss function; n represents the total amount of panel images in the input deep learning model; j is a natural number from 1 to N; y isjA true tag value representing the jth panel image; labeljIndicating the predicted level value of the jth panel image.
Preferably, in the panel quality detecting method, the second loss function is: predicting the difference value of the information entropies of the level probability distribution and the real probability distribution;
wherein L isklRepresenting a second loss function; n is a radical of1Representing a prediction level probability distribution; n is a radical of2Representing the true probability distribution corresponding to the true tag value.
According to the second aspect of the present invention, there is also provided a panel quality detection system based on distribution learning, including a first calculation unit, a fitting unit, a second calculation unit, and a third calculation unit;
the first calculating unit is used for calculating the predicted grade value of the panel image according to the probability value of the panel image predicted by the deep learning model and belonging to different quality grades; generating a first loss function according to the prediction grade value and the real label value of the panel image;
the fitting unit is used for fitting the probability values of the panel images belonging to different quality grades with a normal distribution curve to obtain prediction grade probability distribution;
the second calculating unit is used for obtaining a second loss function according to the prediction grade probability distribution and the real probability distribution corresponding to the real label value;
and the third calculation unit is used for weighting the first loss function and the second loss function to obtain a final loss function of the deep learning model.
Preferably, in the panel quality detection system, the probability values of the panel images predicted by the deep learning model and belonging to different quality levels are as follows: dividing the characteristic data of each quality grade by the sum of the characteristic data of all the quality grades;
wherein K represents a maximum quality level; i is a natural number from 1 to K; p is a radical ofiA probability value representing an ith quality level;characteristic data representing an ith quality level;
the first calculating unit obtains a predicted grade value of the panel image by calculating the sum of products of the probability values of the quality grades and the corresponding quality grade values;
wherein K represents a maximum quality level; i is a natural number from 1 to K; label represents a predicted rank value of the panel image; p is a radical ofiRepresenting the probability value of the i-th quality level.
Preferably, in the panel quality detection system, the first calculation unit obtains the first loss function by dividing the sum of differences between the true label values and the predicted gradation values of different panel images by the number of panel images;
wherein L is1Representing a first loss function; n represents the total number of panel images in the input depth learning model; j is a natural number from 1 to N; y isjA true tag value representing the jth panel image; labeljA predicted level value representing the jth panel image;
the second calculating unit obtains a second loss function by calculating a difference value of the information entropy of the prediction level probability distribution and the actual probability distribution;
wherein L isklRepresenting a second loss function; n is a radical of1Representing a prediction level probability distribution; n is a radical of2Representing the true probability distribution corresponding to the true tag value.
According to a third aspect of the present invention, there is also provided a terminal device, comprising at least one processing unit, and at least one memory unit,
wherein the storage unit stores a computer program that, when executed by the processing unit, causes the processing unit to execute the steps of any of the panel quality detection methods described above.
According to a fourth aspect of the present invention, there is also provided a computer-readable medium having stored therein a computer program executable by a terminal device, the program, when run on the terminal device, causing the terminal device to perform any of the steps of the panel quality detecting method.
In general, compared with the prior art, the above technical solution contemplated by the present invention can achieve the following beneficial effects:
according to the panel quality detection method, the system, the terminal device and the computer readable medium, the deep learning model is trained by combining a first loss function of a regression grade and a second loss function of prediction grade probability distribution, the first loss function enables the prediction grade of the image quality scoring system to be close to the grade of a real label, the second loss function enables the prediction grade probability distribution to be close to the actual probability distribution, and through learning the distribution of the prediction grade probability, the correlation among different grades is utilized in the deep learning level, so that the accuracy of probability distribution of different quality grades of a panel image output by the deep learning model is improved; the two loss functions are weighted and summed to guide the neural network to train, so that the accuracy of the quality grade predicted value of the panel to be detected is effectively improved.
Drawings
FIG. 1 is a flow chart of a panel quality detection method provided by an embodiment of the invention;
FIG. 2 is an overall framework diagram of a training phase provided by an embodiment of the present invention;
FIG. 3 is an overall framework diagram of the prediction phase provided by an embodiment of the present invention;
fig. 4 is a logic block diagram of a panel quality detection system according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. In addition, the technical features involved in the embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.
In order to enable the panel quality evaluation method to be more efficient and achieve the accuracy rate close to or even exceeding the manual judgment level, the invention provides a method for detecting the panel picture quality based on a convolutional neural network of distribution learning. The KL divergence (Kullback-Leibler divergence) can measure the distance between two probability distributions well, the closer the two distributions are, the smaller the KL divergence, and if the farther the KL divergence is, the larger the KL divergence.
The first Loss function L1 Loss makes the prediction level of the picture quality scoring system approach the true label level, and the second Loss function KL Loss makes the predicted level probability distribution approach the actual probability distribution. Predicted level probability distribution of panel image fitting of positive distribution curve N by predicted level probability value1(μ1,σ1 2) The actual grade probability distribution of the panel image is N2(μ2,σ2 2) Wherein the mean value mu2For true tag values, variance σ2Calculated from the dataset samples. The two loss functions are weighted and summed and then guide the neural network to train, and the high-accuracy end-to-end panel picture quality evaluation method is realized.
The present invention will be described in detail with reference to the following examples and drawings.
FIG. 1 is a flow chart of a panel quality detection method provided by an embodiment of the invention; as shown in fig. 1, the panel quality inspection method includes the steps of:
s0: preparing data, collecting detection panel images with different quality grades, marking the detection panel images, and adding quality grade labels; in this embodiment, the panel quality grade is divided into five grades, and the corresponding grade labels are 1, 2, 3, 4 and 5 respectively;
manually selecting a certain number of panel images with different quality grades, roughly dividing the data set according to the proportion of 6:2:2 of the number of samples of the training set, the verification set and the test set, and roughly balancing the number of the panel image samples with different quality grades in the training set, the verification set and the test set.
S1: inputting the panel images in the training set into a deep learning model for training; calculating to obtain a predicted grade value of the corresponding panel image according to the probability value of the panel image predicted by the deep learning model and belonging to different quality grades; generating a first loss function according to the prediction grade value and the real label value of the panel image;
FIG. 2 is an overall framework diagram of the training phase provided by the present embodiment; referring to FIG. 2, the last layer of fully connected layer data z predicted by the deep learning modeli(i-1 … K) obtaining probability value p of each category after SoftmaxiThe probability value piDividing the feature data of each quality grade by the sum of the feature data of all quality grades; the specific calculation formula is as follows:
wherein K represents the maximum quality level, and K is 5 in this embodiment; i is a natural number from 1 to K; p is a radical ofiA probability value representing an ith quality level;characteristic data representing the ith quality level.
Obtaining probability value p of panel image belonging to different quality levelsiThen, according to the probability value piCalculating a grade value label of the regression of the prediction result; in this embodiment, the predicted level value label of the panel image is the sum of the products of the probability values of the quality levels and the corresponding quality level values; the specific calculation formula is as follows:
this embodiment defines the Loss function L1 Loss of the expected regression as the first Loss function L1And calculating a first loss function L according to the real label value and the predicted grade value of the panel image1(ii) a In particular, the first loss function L1Dividing the sum of the differences between the real label values of different panel images and the predicted grade values of the real label values by the total amount of the panel images to obtain an average value; the calculation formula is as follows:
wherein, N represents the total amount of panel images in each batch of input deep learning models, for example, if the number of panel images input in a single time is 10, then N takes a value of 10; j is a natural number from 1 to N; y isjA true tag value representing the jth panel image; labeljIndicating the predicted level value of the jth panel image.
S2: probability value p of panel image output by Softmax belonging to different quality levelsi(i-1 … K) is fitted to a normal distribution curve to obtain a prediction rank probability distribution N1(μ1,σ1 2) (ii) a According to the prediction grade probability distribution and the real probability distribution N corresponding to the real label value2(μ2,σ2 2) Obtaining a second loss function; μ represents a mean value; μ represents variance;
the accuracy of the probability values (or prediction grade probability distribution) of the panel images belonging to different quality grades, which are obtained by the deep learning model prediction, directly determines the accuracy of the panel quality grade evaluation; therefore, if the accuracy of model prediction is to be improved, it is necessary to improve the accuracy of probability distributions of the Softmax output panel images belonging to different quality levels; for this reason, the present embodiment defines klloss of the rank probability distribution as the second Loss function LklCalculating a second loss function according to the prediction level probability distribution and the real probability distribution; in particular, the second loss function LklTo predict a rank probability distribution N1(μ1,σ1 2) And true probability distribution N2(μ2,σ2 2) The difference of the information entropy of (a);
s3: applying a first loss function L1And a second loss function LklWeighting to obtain a final loss function L of the deep learning model, and training the deep learning model until the final loss function is minimumCompleting transformation and training to obtain a panel picture quality detection model;
the final loss function L of the training phase is represented by L1And LklAnd (3) weighted composition:
wherein the content of the first and second substances,the weighting coefficients are adjusted according to the training situation.
S4: evaluating the panel picture quality detection model trained in the step S3 by using the verification set established in the step S0, wherein the whole framework of the prediction stage is shown in FIG. 3;
s5: detecting the panel images to be detected in the test set established in the step S0 by using the trained deep learning model to obtain the quality grade of the panel images to be detected;
in this embodiment, 600 panel samples collected actually are evaluated, and are divided into 5 levels (1, 2, 3, 4, 5), wherein 120 samples are taken for each level, 500 samples are taken as a training set (the number of samples for each level is 100), and both a verification set and a test set are 50 samples (the number of samples for each level is 10); a class 2 sample of the test set is predicted by the trained panel image quality detection model, and the probabilities of the model output classes 1 to 5 are (0.45, 0.2, 0.08, 0.12, 0.15), respectively, so that the predicted class is 2.32(0.45 × 1+0.2 + 2.08 × 3+0.12 × 4+0.15 × 5 is 2.32), which illustrates that the panel quality detection method provided by the embodiment has better accuracy.
The present embodiment also provides a panel quality detection system based on distribution learning, as shown in fig. 4, the panel quality detection system includes a first calculation unit, a fitting unit, a second calculation unit, and a third calculation unit;
the first calculating unit is used for calculating the predicted grade value of the panel image according to the probability value of the panel image predicted by the deep learning model and belonging to different quality grades; generating a first loss function according to the prediction grade value and the real label value of the panel image;
in this embodiment, the probability value p of the panel image predicted by the deep learning model belonging to different quality levelsiDividing the feature data of each quality grade by the sum of the feature data of all quality grades; the calculation formula is as follows:
wherein K represents the maximum quality level, and K is 5 in this embodiment; i is a natural number from 1 to K; p is a radical ofiA probability value representing an ith quality level;characteristic data representing an ith quality level;
obtaining probability value p of panel image belonging to different quality levelsiThen, the first calculation unit calculates the probability value piCalculating a grade value label of the regression of the prediction result; in this embodiment, the predicted level value label of the panel image is the sum of the products of the probability values of the quality levels and the corresponding quality level values; the specific calculation formula is as follows:
the first calculation unit calculates a first loss function L according to the real label value and the prediction grade value of the panel image1(ii) a In particular, the first loss function L1Dividing the sum of the differences between the real label values of different panel images and the predicted grade values of the real label values by the average value of the total amount of the panel images;
wherein N represents the total amount of panel images in each batch of input deep learning models; j is a natural number from 1 to N; y isjA true tag value representing the jth panel image; labeljA predicted level value representing the jth panel image;
the fitting unit is used for outputting probability values p of the panel images belonging to different quality levels by Softmaxi(i-1 … K) is fitted to a normal distribution curve to obtain a prediction rank probability distribution N1(μ1,σ1 2);
The present embodiment defines KL Loss of the rank probability distribution as the second Loss function LklThe second calculation unit is used for calculating the predicted level probability distribution N output by the fitting unit1(μ1,σ1 2) And a true probability distribution N corresponding to a true label value2(μ2,σ2 2) Obtaining a second loss function; obtaining a second loss function L by calculating the difference value of the information entropy of the prediction level probability distribution and the actual probability distributionkl;
A third calculation unit for calculating the first loss function L1And a second loss function LklWeighting to obtain a final loss function L of the deep learning model; the final loss function L is defined by L1And LklAnd (3) weighted composition:
wherein the content of the first and second substances,the weighting coefficients are adjusted according to the training situation.
And training the deep learning model until the final loss function is minimized, and finishing the training to obtain the panel picture quality detection model.
The present embodiment also provides a terminal device, which includes at least one processor and at least one memory, where the memory stores a computer program, and when the computer program is executed by the processor, the processor is enabled to execute the steps of the panel quality detection method. The type of processor and memory are not particularly limited, for example: the processor may be a microprocessor, digital information processor, on-chip programmable logic system, or the like; the memory may be volatile memory, non-volatile memory, a combination thereof, or the like.
The present embodiment also provides a computer-readable medium storing a computer program executable by a terminal device, and when the computer program runs on the terminal device, the terminal device is caused to execute the steps of the panel quality detection method. Types of computer readable media include, but are not limited to, storage media such as SD cards, usb disks, fixed hard disks, removable hard disks, and the like.
It will be understood by those skilled in the art that the foregoing is only a preferred embodiment of the present invention, and is not intended to limit the invention, and that any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the scope of the present invention.
Claims (10)
1. A panel quality detection method based on distribution learning is characterized by comprising the following steps:
calculating to obtain a predicted grade value of the panel image according to a probability value of the panel image predicted by the deep learning model and belonging to different quality grades; generating a first loss function according to the prediction grade value and the real label value of the panel image;
fitting the probability values of the panel images belonging to different quality grades with a normal distribution curve to obtain prediction grade probability distribution; obtaining a second loss function according to the prediction grade probability distribution and the real probability distribution corresponding to the real label value;
weighting the first loss function and the second loss function to obtain a final loss function of the deep learning model, and training the deep learning model based on the final loss function;
and detecting the panel image to be detected by using the trained deep learning model to obtain the quality grade of the panel image to be detected.
2. The panel quality detection method of claim 1, wherein the probability values of the panel images predicted by the deep learning model to belong to different quality levels are: the feature data for each quality class is divided by the sum of the feature data for all quality classes.
3. The panel quality detecting method according to claim 1, wherein the predicted level value of the panel image is a sum of products of a probability value of each quality level and a quality level value corresponding thereto.
4. The panel quality inspection method according to claim 1, wherein the first loss function is an average value obtained by dividing a sum of differences between the true label values and their predicted gradation values of different panel images by the number of panel images.
5. The panel quality inspection method according to claim 1, wherein the second loss function is a difference of information entropies of the prediction level probability distribution and the true probability distribution.
6. A panel quality detection system based on distribution learning is characterized by comprising a first calculation unit, a fitting unit, a second calculation unit and a third calculation unit;
the first calculating unit is used for calculating the predicted grade value of the panel image according to the probability value of the panel image predicted by the deep learning model and belonging to different quality grades; generating a first loss function according to the prediction grade value and the real label value of the panel image;
the fitting unit is used for fitting the probability values of the panel images belonging to different quality grades with a normal distribution curve to obtain prediction grade probability distribution;
the second calculating unit is used for obtaining a second loss function according to the prediction grade probability distribution and the real probability distribution corresponding to the real label value;
and the third calculation unit is used for weighting the first loss function and the second loss function to obtain a final loss function of the deep learning model.
7. The panel quality inspection system of claim 6, wherein the probability value of the panel image predicted by the deep learning model to belong to different quality levels is the feature data of each quality level divided by the sum of the feature data of all quality levels;
the first calculation unit obtains a predicted level value of the panel image by summing up the products of the probability values of the respective quality levels and the corresponding quality level values.
8. The panel quality inspection system according to claim 6 or 7, wherein the first calculation unit obtains the first loss function by dividing a sum of differences between real label values of different panel images and predicted level values thereof by the number of panel images;
the second calculation unit obtains a second loss function by calculating a difference value of the information entropy of the prediction level probability distribution and the true probability distribution.
9. A terminal device, characterized in that it comprises at least one processing unit, and at least one memory unit,
wherein the storage unit stores a computer program which, when executed by the processing unit, causes the processing unit to carry out the steps of the method of any of claims 1 to 5.
10. A computer-readable medium, in which a computer program is stored which is executable by a terminal device, and which, when run on the terminal device, causes the terminal device to carry out the steps of the method of any one of claims 1 to 5.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910843194.4A CN110751170A (en) | 2019-09-06 | 2019-09-06 | Panel quality detection method, system, terminal device and computer readable medium |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910843194.4A CN110751170A (en) | 2019-09-06 | 2019-09-06 | Panel quality detection method, system, terminal device and computer readable medium |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110751170A true CN110751170A (en) | 2020-02-04 |
Family
ID=69276134
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910843194.4A Pending CN110751170A (en) | 2019-09-06 | 2019-09-06 | Panel quality detection method, system, terminal device and computer readable medium |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110751170A (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111553557A (en) * | 2020-03-30 | 2020-08-18 | 徐州徐工挖掘机械有限公司 | Workshop product quality assessment method based on deep learning |
CN113343787A (en) * | 2021-05-20 | 2021-09-03 | 沈阳铸造研究所有限公司 | Deep learning-based grade evaluation method applicable to map comparison scene |
CN113610167A (en) * | 2021-08-10 | 2021-11-05 | 宿迁旺春机械制造有限公司 | Equipment risk detection method based on metric learning and visual perception |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2017191872A1 (en) * | 2016-05-04 | 2017-11-09 | 두산중공업 주식회사 | Plant abnormality detection method and system |
CN107463953A (en) * | 2017-07-21 | 2017-12-12 | 上海交通大学 | Image classification method and system based on quality insertion in the case of label is noisy |
CN108171707A (en) * | 2018-01-23 | 2018-06-15 | 武汉精测电子集团股份有限公司 | A kind of Mura defects level evaluation method and device based on deep learning |
CN109886329A (en) * | 2019-02-18 | 2019-06-14 | 中国铁建重工集团有限公司 | Rock crusher level detection method, detection system and heading equipment |
CN110147456A (en) * | 2019-04-12 | 2019-08-20 | 中国科学院深圳先进技术研究院 | A kind of image classification method, device, readable storage medium storing program for executing and terminal device |
-
2019
- 2019-09-06 CN CN201910843194.4A patent/CN110751170A/en active Pending
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2017191872A1 (en) * | 2016-05-04 | 2017-11-09 | 두산중공업 주식회사 | Plant abnormality detection method and system |
CN107463953A (en) * | 2017-07-21 | 2017-12-12 | 上海交通大学 | Image classification method and system based on quality insertion in the case of label is noisy |
CN108171707A (en) * | 2018-01-23 | 2018-06-15 | 武汉精测电子集团股份有限公司 | A kind of Mura defects level evaluation method and device based on deep learning |
CN109886329A (en) * | 2019-02-18 | 2019-06-14 | 中国铁建重工集团有限公司 | Rock crusher level detection method, detection system and heading equipment |
CN110147456A (en) * | 2019-04-12 | 2019-08-20 | 中国科学院深圳先进技术研究院 | A kind of image classification method, device, readable storage medium storing program for executing and terminal device |
Non-Patent Citations (4)
Title |
---|
房慧弟: "图像美学质量评价方法研究", 《中国优秀硕士学位论文全文数据库 (信息科技辑)》 * |
房慧弟: "图像美学质量评价方法研究", 《中国优秀硕士学位论文全文数据库 (信息科技辑)》, 15 December 2018 (2018-12-15), pages 18 * |
李万军: "《概率论与数理统计》", 电子科技大学出版社, pages: 111 - 117 * |
陈志浩: "基于深度学习的磁片表面缺陷检测研究", 《中国优秀硕士学位论文全文数据库 (信息科技辑)》, 15 July 2019 (2019-07-15) * |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111553557A (en) * | 2020-03-30 | 2020-08-18 | 徐州徐工挖掘机械有限公司 | Workshop product quality assessment method based on deep learning |
CN113343787A (en) * | 2021-05-20 | 2021-09-03 | 沈阳铸造研究所有限公司 | Deep learning-based grade evaluation method applicable to map comparison scene |
CN113343787B (en) * | 2021-05-20 | 2023-09-01 | 中国机械总院集团沈阳铸造研究所有限公司 | Deep learning-based medium-level assessment method suitable for map contrast scene |
CN113610167A (en) * | 2021-08-10 | 2021-11-05 | 宿迁旺春机械制造有限公司 | Equipment risk detection method based on metric learning and visual perception |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110569837B (en) | Method and device for optimizing damage detection result | |
CN113160192B (en) | Visual sense-based snow pressing vehicle appearance defect detection method and device under complex background | |
CN108885700A (en) | Data set semi-automatic labelling | |
CN110751170A (en) | Panel quality detection method, system, terminal device and computer readable medium | |
KR20190075707A (en) | Method for sorting products using deep learning | |
CN103745454B (en) | Method for evaluating image processing algorithms or systems based on rank learning | |
CN113592828B (en) | Nondestructive testing method and system based on industrial endoscope | |
CN114663346A (en) | Strip steel surface defect detection method based on improved YOLOv5 network | |
CN110567967B (en) | Display panel detection method, system, terminal device and computer readable medium | |
CN114494780A (en) | Semi-supervised industrial defect detection method and system based on feature comparison | |
CN115131283A (en) | Defect detection and model training method, device, equipment and medium for target object | |
CN114066857A (en) | Infrared image quality evaluation method and device, electronic equipment and readable storage medium | |
CN114945938A (en) | Method and device for detecting actual area of defect and method and device for detecting display panel | |
CN112070762A (en) | Mura defect detection method and device for liquid crystal panel, storage medium and terminal | |
CN107016416A (en) | The data classification Forecasting Methodology merged based on neighborhood rough set and PCA | |
CN111754491A (en) | Picture definition judging method and device | |
CN116958035A (en) | Product part defect detection method, device, medium and electronic equipment | |
CN109886923A (en) | It is a kind of for internet detection in measurement detection system and method based on machine learning | |
CN117392042A (en) | Defect detection method, defect detection apparatus, and storage medium | |
CN115222691A (en) | Image defect detection method, system and related device | |
CN113516652A (en) | Battery surface defect and adhesive detection method, device, medium and electronic equipment | |
CN113837220A (en) | Robot target identification method, system and equipment based on online continuous learning | |
CN112287898A (en) | Method and system for evaluating text detection quality of image | |
CN112581001A (en) | Device evaluation method and device, electronic device and readable storage medium | |
CN113112515B (en) | Evaluation method for pattern image segmentation algorithm |
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 | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20200204 |