CN113822842A - Industrial defect detection method based on multi-task learning - Google Patents
Industrial defect detection method based on multi-task learning Download PDFInfo
- Publication number
- CN113822842A CN113822842A CN202110484644.2A CN202110484644A CN113822842A CN 113822842 A CN113822842 A CN 113822842A CN 202110484644 A CN202110484644 A CN 202110484644A CN 113822842 A CN113822842 A CN 113822842A
- Authority
- CN
- China
- Prior art keywords
- classification
- defect
- picture
- defect detection
- model
- 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
- 230000007547 defect Effects 0.000 title claims abstract description 81
- 238000001514 detection method Methods 0.000 title claims abstract description 42
- 238000000034 method Methods 0.000 claims abstract description 20
- 238000013145 classification model Methods 0.000 claims abstract description 14
- 230000006870 function Effects 0.000 claims abstract description 14
- 230000002159 abnormal effect Effects 0.000 claims abstract description 11
- 238000005070 sampling Methods 0.000 claims description 6
- 238000007781 pre-processing Methods 0.000 claims description 4
- 238000012545 processing Methods 0.000 claims description 4
- 230000004913 activation Effects 0.000 claims description 2
- 238000012986 modification Methods 0.000 claims description 2
- 230000004048 modification Effects 0.000 claims description 2
- 238000013527 convolutional neural network Methods 0.000 abstract description 8
- 238000003384 imaging method Methods 0.000 abstract description 7
- 230000000694 effects Effects 0.000 abstract description 5
- 230000008859 change Effects 0.000 abstract description 4
- 238000012549 training Methods 0.000 description 16
- 238000012360 testing method Methods 0.000 description 6
- 230000008569 process Effects 0.000 description 5
- 238000011156 evaluation Methods 0.000 description 4
- 238000002474 experimental method Methods 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 1
- 239000003086 colorant Substances 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 230000002950 deficient Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000007717 exclusion Effects 0.000 description 1
- 239000012634 fragment Substances 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Classifications
-
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/047—Probabilistic or stochastic 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- 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
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10004—Still image; Photographic image
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biophysics (AREA)
- Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Probability & Statistics with Applications (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to the technical field of industrial defect detection, and discloses an industrial defect detection method based on multi-task learning, which is characterized in that a defect classification task is subdivided into two subtasks, a normal/abnormal (ok/ng) classification problem (marked as task1) and a multilabel classification problem (marked as task2) of n defect types are respectively solved, and a classification model based on a Convolutional Neural Network (CNN) is constructed. The classification model consists of a base model and a head. The base model is responsible for extracting image features of an input image to obtain a corresponding feature image, and the base models of different tasks share network weight by adopting a hardharing connection mode. The head is an output layer, and two branches are led out from the base model and are respectively used for solving task1 and task 2; the two branches are respectively composed of a fully connected layer and a sigmod function, and the probability of the ng type and the type probability of the n defects are output. The method can solve the problem that the detection effect is unstable due to the fact that the existing industrial defect detection method is easily interfered by imaging conditions, small difference between the defect and the background, low image contrast, large change of the size and appearance of the same type of defect and the like.
Description
Technical Field
The invention relates to the technical field of industrial defect detection, in particular to an industrial defect detection method based on multi-task learning.
Background
Industrial defect detection is a technique for automatically identifying a defective portion in an image taken by an industrial camera by a machine vision algorithm. Specifically, the industrial defect detection needs to judge whether a defect exists in an image and identify the type of the defect, so as to analyze the defect degree of an industrial product. The technology can be widely applied to various industrial fields to replace manual detection, and the production efficiency, the detection precision and the stability of products are improved.
In a conventional industrial defect detection method based on machine vision, a proper imaging scheme (such as bright field imaging, dark field imaging, hybrid imaging and the like) is generally selected according to the reflection property of the surface of a detected object, so as to obtain an image with uniform illumination, and whether defects exist in the image is identified by adopting an image processing algorithm or a mode of manually designing a feature and a classifier. However, in a real industrial environment, imaging conditions are complex and changeable, which causes problems of large image noise, brightness change and the like, and in addition, the defects have small difference with the background, low image contrast, large change of the same type of defect scale and appearance and the like, and the conventional detection method is easily interfered by the factors, so that the detection effect is unstable, and the actual use requirement cannot be met.
In recent years, Convolutional Neural Networks (CNNs) have been widely used in industrial detection scenarios due to their strong feature expression capability (which can overcome interference caused by changes in defect backgrounds, colors, textures, shapes, and the like). The industrial defect detection problem based on the CNN classifier can be modeled as an abnormal image classification problem, and the target of the problem is as follows: and identifying whether the picture is a normal (ok) picture or an abnormal (ng) picture, and judging the category of the ng picture. Since there may be a plurality of defects in one image, the problem can be solved by a multi-label (multilabel) classification model. The classification problem of abnormal pictures in the industrial defect detection scene is different from the classification problem of general multilabel images in that: 1) the sample types in the industrial defect detection problem are quite unbalanced, and the number of normal samples is far higher than that of abnormal samples; 2) in the multilabel classification problem, a picture may contain any number of labels of any type, and the specificity of industrial defect detection lies in: a picture can have various defect types, but if the picture has defects, the type of the picture does not belong to a normal picture; 3) the evaluation and labeling of the model in the industrial defect detection scene are different from the general classification problem, and the importance of the capability of distinguishing normal samples from abnormal samples in the scene is much higher than the capability of distinguishing different types of abnormal samples by the model; in addition, the scene requires that the model can suppress false alarms as low as possible under the condition of extremely low false alarm rate, namely that the model has extremely high recall rate. The problem with directly applying the multilabel classification model to this scenario is that if the ok sample is used as one class of training classes, the multilabel classification problem is characterized by the fact that the classes are not mutually exclusive, while the specificity of the industrial defect detection problem is that the ok class is mutually exclusive from the other classes. If the ok samples are negative samples which do not belong to any training category, the ok samples are far more than the abnormal samples in the industrial defect detection problem, so that the number of the negative samples is far higher than that of the positive samples, which affects the training and convergence of the CNN model.
Disclosure of Invention
Technical problem to be solved
The embodiment of the invention provides an industrial defect detection method based on multi-task learning, which is used for solving the problems that the conventional industrial defect detection method is easily interfered by factors such as imaging conditions, small difference between defects and background, low image contrast, large change of the same type of defect scale and appearance and the like, so that the detection effect is unstable and the actual use requirement cannot be met.
Disclosure of the invention
The embodiment of the invention provides an industrial defect detection method based on multitask learning, which is characterized by comprising the following steps of:
the method comprises the following steps: collecting pictures to be detected, carrying out uniform preprocessing operation, and inputting the pictures to be detected into a classification model in batches;
step two: outputting a characteristic image corresponding to the picture by a basic network of the classification model;
step three: and (4) in the output layer, processing the characteristic image obtained in the step two by two branches (consisting of a full connection layer and a sigmoid activation layer), and outputting the probability that the picture belongs to the ng picture and the probability that the picture belongs to various defects.
Step four: and comprehensively considering the output values of the two branches to judge whether the picture to be detected contains defects and defect types. Judging whether the picture is a normal picture or not through ok/ng two classification branches, if not, selecting the class with the probability value output by the multilabel classification branch higher than a threshold value as a predicted defect class, and if the probability of all the classes is lower than the threshold value, selecting the class with the highest probability value as the defect class.
The multi-task classification network adopts an end-to-end mode to train the model. In the training process, the error between the predicted value and the expected output (true value) of the network output is calculated according to the preset loss function, and the weight of the network is updated through the back propagation error，The method specifically comprises the following steps:
step 1) setting parameters such as initialization network parameters, maximum iteration times, initial learning rate and the like;
step 2) under-sampling the normal sample, and over-sampling the abnormal sample;
step 3) randomly sampling samples of each batch, uniformly preprocessing each sample, and taking the preprocessed samples as the forward input of the network;
step 4) calculating the network output of each sample, calculating a loss value by combining the real value of the sample, reversely returning the loss, and updating the network weight;
step 5) judging whether the loss value reaches a target value or reaches the maximum iteration times, if so, saving the model and calculating a category classification threshold; if not, returning to execute the step 3).
Selecting FL (p) in the model training processt) The loss function is a loss function that is a function of the two classification tasks,
FL(pt)＝-(1-pt)γlog(pt)
wherein (1-p)t)γEquivalent to cross entropy lossA modulation factor, gamma is more than or equal to 0 and is a focusing hyperparameter, ptThe prediction probability of the model is specifically defined as follows:
where y ∈ { -1, 1} is a true category of the input picture, and p ∈ [0, 1] is a model prediction value, and represents a probability that the input picture belongs to the category (y { -1).
The overall loss function is defined as: the weighted sum of the loss of the two tasks, the weight is 1, and the specific formula is as follows:
L＝L1+L2；
where L1 and L2 are the FL (pt) loss function values for task1 and task2, respectively.
(III) advantageous effects
The industrial defect detection method based on multi-task learning provided by the embodiment of the invention decomposes a defect classification task into two subtasks, and simultaneously solves the two tasks by training a CNN classification model. In consideration of the particularity of the industrial defect detection problem, namely the mutual exclusion of a normal class and a defect class and the capability of distinguishing ok samples and ng samples by a model, the invention adds a subtask of ok/ng two classification problems to solve the problem in addition to the general defect classification task. Specifically, the industrial defect detection task is decomposed into the following two subtasks: one is to judge whether the picture is a normal picture, and the other is to judge the defect type of the picture. The method has the following beneficial effects:
firstly, the invention designs a CNN classification model framework based on multi-task learning to solve the problem of industrial defect detection, the common information among a plurality of tasks needs to be modeled in the training process of the model, and better classification precision and generalization performance than a multilabel classification model can be obtained through joint task learning.
Secondly, when two associated tasks are learned, through loss functions and training modes which are reasonably designed, the learning processes of different tasks can play a role in mutual promotion, and therefore the performance of the model is improved.
Thirdly, in the reasoning process, the output values of the two branches are comprehensively considered to judge whether the input picture contains defects and defect types. Judging whether the picture is a normal picture or not through ok/ng two classification branches, if not, selecting the class exceeding the threshold value from the output values of the multilabel defect classification branches as a defect class, and if all the defect classes are lower than the threshold value, selecting the class corresponding to the maximum output value as the defect class.
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 described below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and those skilled in the art can also obtain other drawings according to the drawings without creative efforts.
FIG. 1 is a block flow diagram of a method for industrial defect detection based on multi-task learning according to an embodiment of the present invention;
FIG. 2 is a flow chart of model training of a method for industrial defect detection based on multi-task learning according to an embodiment of the present invention;
fig. 3 is a flowchart of a model test of an industrial defect detection method based on multi-task learning according to an embodiment of the present invention.
Fig. 4 is an example image of a photovoltaic cell industrial defect detection data set of an industrial defect detection method based on multitask learning in an embodiment of the present invention.
Detailed Description
In order to verify the effect of the method provided by the embodiment, the embodiment performs experiments on industrial data of the photovoltaic cell. The data set comprises ok pictures and 5 defect type pictures, the defect types are fragments, cross hidden cracks, single hidden cracks, cold joints and broken grids respectively, the data set is divided into a training set and a test set according to a data ratio of 4:1, and the number of the pictures of each type in the test set is shown in table 1. Fig. 4 shows exemplary pictures of various categories, and it can be seen from the drawings that the following difficulties mainly exist in processing the experimental data: the background of the picture is complex, and the background texture can interfere with the judgment of the defect; while some defects have similar features, some samples have greater intra-class differences than inter-class differences. In addition, in an actual industrial scene, the appearance and background of abnormal samples are changed greatly, the resolution of defects is very small, the number of some low-frequency defects is difficult to collect, the number of training samples is very small, and the interferences can make the model training difficult.
In this embodiment, serensnext 50 is used as a basic network to compare a multilabel classification method with the classification method based on multitask learning proposed by the present invention, where a 6-class multilabel classification Model (training classes are ok and 5 defect classes) and a 5-class multilabel classification Model (training classes are 5 defect classes, and ok samples are negative samples that do not belong to any of the training classes) are respectively recorded as Model1 and Model 2. In the experiment, the processes of training, testing and evaluation are completely consistent. The test results are shown in tables 1, 2 and 3, and for the evaluation method, the present example employed the Precision @ (Recall > ═ 0.995) values at the time of the false alarm rate, the false alarm rate and the Recall rate of 0.995 as evaluation criteria. Precision @ (Recall > -0.995) represents the model accuracy under the condition of extremely high Recall rate, and the missing report rate and the false report rate represent the capability of the model for judging whether the picture has defects, so that the classification method provided by the invention is superior to the multilabel classification method.
TABLE 1 number of pictures in each category of test set
Table 2 comparative experimental results
TABLE 3 comparison of missing and false alarm rates
In conclusion, experiments fully verify the advantages of the industrial defect detection method based on multi-task learning provided by the scheme.
Finally, it should be noted that: the above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.
Claims (4)
1. An industrial defect detection method based on multitask learning comprises the following steps:
the method comprises the following steps: collecting pictures to be detected, carrying out uniform preprocessing operation, and inputting the pictures to be detected into a classification model in batches;
step two: outputting a characteristic image corresponding to the picture by a basic network of the classification model;
step three: and (4) processing the characteristic images obtained in the second step by two branches on an output layer, wherein the two branches are a full connection layer and a sigmoid activation layer respectively, and outputting the probability that the picture belongs to the ng picture and the probability that the picture belongs to various defects.
Step four: and comprehensively considering the output values of the two branches to judge whether the picture to be detected contains defects and defect types. Judging whether the picture is a normal picture or not through ok/ng two classification branches, if not, selecting the class with the probability value output by the multilabel classification branch higher than a threshold value as a predicted defect class, and if the probability of all the classes is lower than the threshold value, selecting the class with the highest probability value as the defect class.
2. The industrial defect detection method based on multitask learning according to claim 1, characterized by comprising the following steps: the classification model in the step is trained in an end-to-end mode, namely, the error between the predicted value and the expected value of the network is calculated according to a preset loss function, and the network weight is updated through a back-propagation error to obtain an optimal model, and the method specifically comprises the following steps:
step 1) setting parameters such as initialization network parameters, maximum iteration times, initial learning rate and the like;
step 2) under-sampling the normal sample, and over-sampling the abnormal sample;
step 3) randomly sampling samples of each batch, uniformly preprocessing each sample, and taking the preprocessed samples as the forward input of the network;
step 4) calculating the network output of each sample, calculating a loss value by combining the real value of the sample, reversely returning the loss, and updating the network weight;
step 5) judging whether the loss value reaches a target value or reaches the maximum iteration times, if so, saving the model and calculating a category classification threshold; if not, returning to execute the step 3).
3. The industrial defect detection method based on multitask learning according to claim 2, characterized by comprising the following steps: selecting FL (pt) loss functions as the loss functions of two classification tasks, wherein the FL (pt) loss functions are obtained by modification on the basis of standard cross entropy loss and are suitable for solving the problem of sample imbalance in an industrial defect detection scene,
FL(pt)＝-(1-pt)γlog(pt)
wherein (1-p)t)γA modulation factor equivalent to cross entropy loss, gamma is more than or equal to 0 and is a focusing hyperparameter, (p)t) The prediction probability of the model is specifically defined as follows:
where y ∈ { -1, 1} is a true category of the input picture, and p ∈ [0, 1] is a model prediction value, and represents a probability that the input picture belongs to the category (y { -1).
4. The industrial defect detection method based on multitask learning according to claim 2, characterized in that the overall loss function is defined as: the weighted sum of the loss of the two tasks, the weight is 1, and the specific formula is as follows:
L＝L1+L2；
where L1 and L2 are the FL (pt) loss function values for task1 and task2, respectively.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110484644.2A CN113822842A (en) | 2021-04-30 | 2021-04-30 | Industrial defect detection method based on multi-task learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110484644.2A CN113822842A (en) | 2021-04-30 | 2021-04-30 | Industrial defect detection method based on multi-task learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113822842A true CN113822842A (en) | 2021-12-21 |
Family
ID=78912492
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110484644.2A Pending CN113822842A (en) | 2021-04-30 | 2021-04-30 | Industrial defect detection method based on multi-task learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113822842A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115249316A (en) * | 2022-09-22 | 2022-10-28 | 江苏智云天工科技有限公司 | Industrial defect detection method and device |
| CN117372368A (en) * | 2023-10-17 | 2024-01-09 | 苏州真目人工智能科技有限公司 | Appearance detection device and method based on cascade closed-loop deep learning algorithm |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9928448B1 (en) * | 2016-09-23 | 2018-03-27 | International Business Machines Corporation | Image classification utilizing semantic relationships in a classification hierarchy |
| CN109118485A (en) * | 2018-08-13 | 2019-01-01 | 复旦大学 | Digestive endoscope image classification based on multitask neural network cancer detection system early |
| CN110321603A (en) * | 2019-06-18 | 2019-10-11 | 大连理工大学 | A kind of depth calculation model for Fault Diagnosis of Aircraft Engine Gas Path |
| CN110930377A (en) * | 2019-11-18 | 2020-03-27 | 福州大学 | Automatic detection method for drainage pipeline abnormal type based on multitask learning |
| CN111179250A (en) * | 2019-12-30 | 2020-05-19 | 北京航空航天大学 | Industrial product defect detection system based on multitask learning |
| CN111738987A (en) * | 2020-06-01 | 2020-10-02 | 湖南品信生物工程有限公司 | Automatic identification method and device for multitask cervical cancer cells |
-
2021
- 2021-04-30 CN CN202110484644.2A patent/CN113822842A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9928448B1 (en) * | 2016-09-23 | 2018-03-27 | International Business Machines Corporation | Image classification utilizing semantic relationships in a classification hierarchy |
| CN109118485A (en) * | 2018-08-13 | 2019-01-01 | 复旦大学 | Digestive endoscope image classification based on multitask neural network cancer detection system early |
| CN110321603A (en) * | 2019-06-18 | 2019-10-11 | 大连理工大学 | A kind of depth calculation model for Fault Diagnosis of Aircraft Engine Gas Path |
| CN110930377A (en) * | 2019-11-18 | 2020-03-27 | 福州大学 | Automatic detection method for drainage pipeline abnormal type based on multitask learning |
| CN111179250A (en) * | 2019-12-30 | 2020-05-19 | 北京航空航天大学 | Industrial product defect detection system based on multitask learning |
| CN111738987A (en) * | 2020-06-01 | 2020-10-02 | 湖南品信生物工程有限公司 | Automatic identification method and device for multitask cervical cancer cells |
Non-Patent Citations (1)
| Title |
|---|
| 张博豪: "基于深层神经网络的工业图像检测技术研究", 中国优秀硕士学位论文全文数据库信息科技辑, no. 02, 15 February 2021 (2021-02-15), pages 138 - 1529 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115249316A (en) * | 2022-09-22 | 2022-10-28 | 江苏智云天工科技有限公司 | Industrial defect detection method and device |
| CN117372368A (en) * | 2023-10-17 | 2024-01-09 | 苏州真目人工智能科技有限公司 | Appearance detection device and method based on cascade closed-loop deep learning algorithm |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111553929B (en) | Mobile phone screen defect segmentation method, device and equipment based on converged network | |
| US20200349875A1 (en) | Display screen quality detection method, apparatus, electronic device and storage medium | |
| Zhu et al. | Modified densenet for automatic fabric defect detection with edge computing for minimizing latency | |
| WO2020007096A1 (en) | Method and device for detecting quality of display screen, electronic device, and storage medium | |
| CN109919934B (en) | Liquid crystal panel defect detection method based on multi-source domain deep transfer learning | |
| CN107330453B (en) | Pornographic image identification method based on step-by-step identification and fusion key part detection | |
| KR102450131B1 (en) | Systems and methods for detecting flaws on panels using images of the panels | |
| CN109829895B (en) | AOI defect detection method based on GAN | |
| CN109753878B (en) | Imaging identification method and system under severe weather | |
| CN108629370B (en) | Classification recognition algorithm and device based on deep belief network | |
| CN113822842A (en) | Industrial defect detection method based on multi-task learning | |
| CN116385430A (en) | Machine vision flaw detection method, device, medium and equipment | |
| CN117152119A (en) | Profile flaw visual detection method based on image processing | |
| CN114372980A (en) | Industrial defect detection method and system | |
| KR102450130B1 (en) | Systems and methods for detecting flaws on panels using images of the panels | |
| CN113570549A (en) | Defect detection method and device for reflective surface | |
| CN116778269A (en) | Method for constructing product surface defect detection model based on self-encoder reconstruction | |
| CN116777865A (en) | Underwater crack identification method, system, device and storage medium | |
| Ding et al. | Cognitive visual inspection service for LCD manufacturing industry | |
| Kefer et al. | An intelligent robot for flexible quality inspection | |
| Li et al. | Image object detection algorithm based on improved Gaussian mixture model | |
| Acevedo-Ávila et al. | A statistical background modeling algorithm for real-time pixel classification | |
| WO2023166776A1 (en) | Appearance analysis system, appearance analysis method, and program | |
| CN115307731B (en) | Outgoing laser line detection method of laser line projector | |
| Vidyabharathi et al. | Enhancing crack detection with convolutional neural networks and oriented non-maximal suppression |
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 |