CN110135521A - Pole-piece pole-ear defects detection model, detection method and system based on convolutional neural networks - Google Patents
Pole-piece pole-ear defects detection model, detection method and system based on convolutional neural networks Download PDFInfo
- Publication number
- CN110135521A CN110135521A CN201910450580.7A CN201910450580A CN110135521A CN 110135521 A CN110135521 A CN 110135521A CN 201910450580 A CN201910450580 A CN 201910450580A CN 110135521 A CN110135521 A CN 110135521A
- Authority
- CN
- China
- Prior art keywords
- model
- pole
- tab
- pole piece
- piece
- 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.)
- Withdrawn
Links
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/84—Systems specially adapted for particular applications
- G01N21/88—Investigating the presence of flaws or contamination
- G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
-
- G—PHYSICS
- 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
- 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
- G06F18/2163—Partitioning the feature space
-
- 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
- G06F18/25—Fusion techniques
- G06F18/254—Fusion techniques of classification results, e.g. of results related to same input data
-
- 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
- G06N3/044—Recurrent networks, e.g. Hopfield 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
- G06N3/045—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
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/84—Systems specially adapted for particular applications
- G01N21/88—Investigating the presence of flaws or contamination
- G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
- G01N2021/8854—Grading and classifying of flaws
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/84—Systems specially adapted for particular applications
- G01N21/88—Investigating the presence of flaws or contamination
- G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
- G01N2021/8883—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges involving the calculation of gauges, generating models
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)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Biophysics (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Immunology (AREA)
- Pathology (AREA)
- Biochemistry (AREA)
- Analytical Chemistry (AREA)
- Chemical & Material Sciences (AREA)
- Signal Processing (AREA)
- Image Analysis (AREA)
- Investigating Materials By The Use Of Optical Means Adapted For Particular Applications (AREA)
Abstract
The present invention relates to pole-piece pole-ear detections, and in particular to pole-piece pole-ear defects detection model, detection method and system based on convolutional neural networks, model construction: image carries out data segmentation after double gauss difference；Data mark；Neural network building: model 1 is the neural network model for extracting image fixed dimension feature；Model 2 adds classification layer on the basis of model 1, for classifying to tab defect characteristic, model 3 carries out multiple dimensioned boundary recurrence to the feature that pole piece defects detection goes out on the basis of model 1 and determines Main Boundaries framework to determine anchor frame orientation, then by non-maxima suppression method；The feature vector that model 4 returns main body output to pole piece defect boundary on the basis of model 1 is classified；Model training；Create loss function；Model evaluation；It is fused into an optimum fusion model；Model reasoning output test result.The present invention improves pole-piece pole-ear defects detection efficiency and accuracy rate.
Description
Technical field
The present invention relates to pole-piece pole-ear detections, and in particular to the pole-piece pole-ear defects detection mould based on convolutional neural networks
Type, detection method and system.
Background technique
Power battery refers to larger power capacity and output power, can configure electric bicycle, electric car, electronic
The battery of equipment and tool drives power supply is also typically included in military submarine and high-grade intelligent robot and enterprises and institutions
The standing power supply etc. of the energy storage system, communication and command system that use.With emerging electric bicycle, electric car exploitation and
It commercially produces, the development of novel submarines and UAV navigation, so that demand of the society to novel power battery is significantly
Increase.
Power lithium ion battery pole piece in process of production, can be because due to coating machine, roll squeezer etc. in coating process
Reason causes the dew foils of positive and negative anodes, blackening, speck, the defects of dropping off, and can seriously affect the performance and used life of battery in this way.
Therefore after film-making can by artificial detection or use the automatic detection of conventional machines vision, but due to manually vulnerable to it is subjective because
Element influence cause missing inspection to take place frequently, detection efficiency is low, and conventional machines vision algorithm can not cover production in various defects and
Classifying quality is poor so as to cause erroneous detection frequent occurrence, therefore the convolutional neural networks based on deep learning and computer vision are examined
Surveying will replace artificial detection and conventional machines vision-based detection to become the following the main direction of development.
Power lithium ion battery pole piece automatic testing method (201310549948.8), it is mechanical which describes emphatically detection
Structure, and there is no emphasis description to visible detection method.
The power lithium ion battery pole piece defects detection of conventional machines vision at present sweeps camera, highlighted line style by industrial line
Light source, industrial personal computer and image processing algorithm are constituted.But since power lithium-ion battery positive/negative plate needs to detect tow sides, because
This needs 2 groups of cameras to work at the same time to meet testing requirements.Due to different from the testing requirements of tab to pole piece, so right
Should every group of camera can all combine there are many algorithm.In order to classify to defect, will form after the superposition of these algorithm couples
The mapping relations of a set of complexity, detection stability will receive larger impact, and operator is needed to have very strong priori knowledge
Deposit could debug system.Pole piece and tab welding one figure are shown in that Fig. 1, tab are shown in that Fig. 2, pole-piece pole-ear defect picture are shown in
Attached drawing 6.
Summary of the invention
The purpose of the invention is to overcome, detection complicated based on the installation and debugging of conventional machines vision prescription in the prior art
It environmental requirement harshness and needs to debug the deficiency that worker has stronger priori knowledge, provides for power lithium ion battery pole piece
Pole-piece pole-ear defects detection model, detection method and the system based on convolutional neural networks of tab.
Technical solution of the present invention: the construction method of the pole-piece pole-ear defects detection model based on convolutional neural networks, packet
It includes:
(1) image preprocessing: input picture is subjected to double gauss difference, is divided into two class data of pole piece and tab；
(2) data mark: position mark are carried out to the defects of the pole piece image data that step (1) segmentation obtains, to pole
Ear image data carries out classification annotation；
(3) building of neural network: four convolutional Neural models of design, model 1 are to extract image fixed dimension feature
Neural network model；Model 2 adds classification layer, for classifying to tab defect characteristic, model 3 on the basis of model 1
Multiple dimensioned boundary recurrence is carried out to determine anchor frame orientation to the feature that pole piece defects detection goes out on the basis of model 1, then is passed through
Non-maxima suppression method determines Main Boundaries framework；It is defeated that model 4 returns main body to pole piece defect boundary on the basis of model 1
Feature vector out is classified；
It regard model 1 as image fixed dimension Feature Selection Model, regard model 2 as tab defect classification model, model
3,4 become pole piece defects detection model by Multiscale Fusion；
(4) model training model training: is carried out to tab defect classification model and pole piece defects detection model；
(5) it creates loss function: losing letter for tab defect classification model and pole piece defects detection model creation respectively
Two loss additions are obtained final loss function by number；
(6) model evaluation: tab defect classification model and pole piece defects detection model are assessed respectively；
(7) Model Fusion: obtaining the optimal models of tab defect classification and pole piece defects detection according to model evaluation result,
It is fused into an optimum fusion model using two kinds of models as submodel by concentrate function, to accelerate to detect speed；
(8) tab and pole piece image data that are partitioned by image preprocessing data reasoning: are converted into tensor data
The optimum fusion model for entering step (7) simultaneously, during data are propagated forward, the tensor data of tab enter tab and lack
Classification submodel is fallen into, the tensor data of pole piece enter pole piece defects detection submodel, and finally the adjustment Jing Guo confidence threshold value is defeated
Testing result out.
Further, in step (4), pole piece image data and tab image data are divided into 80% training data respectively
Collection and 20% validation data set, using multireel lamination, the method for small convolution kernel is collected using training set and verifying respectively to tab
Defect classification model and pole piece defects detection model are trained and verify after each iteration to training set.
Further, in step (5), for tab defect classification model, Softmax cross entropy loss function is created, is used
Difference between characterization authentic specimen and prediction probability；For pole piece defects detection model, Huber loss function is created, is used
Main body is returned in calculating bounding box；Then the two losses are added and obtain final loss function.
Further, in step (6), tab defect classification results are evaluated using accuracy rate, to pole piece defects detection side
The prediction of boundary's frame is using mean absolute error come evaluation result；If tab defect classification results Average Accuracy less than 98.5%,
The prediction of pole piece defects detection bounding box is averaged recall rate less than 98.5%, then adjusts hyper parameter according to rear re -training.
The present invention also provides a kind of pole-piece pole-ear defect inspection method based on convolutional neural networks, comprising:
S1: the image data of acquisition is passed through cable transmission to industrial personal computer by acquisition image；
S2: pre-processing image data, is divided into two class data of pole piece and tab by double gauss difference；
S3: by two groups of image data conversions of step S2 at tensor data and input previous methods building optimum fusion mould
Type；
S4: the tensor data of tab export tab classification results by tab defect categorical reasoning；The tensor data of pole piece
By pole piece defects detection, defective locations classification results, frame coordinate, confidence level are exported；
S5: product is divided by qualified and unqualified two kinds of categories by rating scale.
Further, in step S1, industrial camera parallel collection image is scanned using line in the tow sides of pole-piece pole-ear.
The present invention also provides a kind of pole-piece pole-ear defect detecting system based on convolutional neural networks, comprising:
Image data acquisition unit, it is parallel using line scanning industrial camera, it is adopted in the tow sides of the pole-piece pole-ear of welding
Collect image, and passes through cable transmission to industrial personal computer；
Pre-processing image data unit is divided into two class data of pole piece and tab by double gauss difference；
Detection unit constructs pole-piece pole-ear defects detection model using mentioned-above method, two groups of image datas is turned
Change Cheng Zhangliang data and inputs the pole-piece pole-ear defects detection model；
As a result output unit exports tab classification results, pole piece defective locations classification results, frame after testing after unit
Coordinate, confidence level, and qualified and unqualified conclusion is provided by rating scale.
Pole piece is made of the coating of collector foil and its two sides, and tab welding is in one end of pole piece.Rating scale is each
The product qualification evaluation criterion of enterprise.
The invention has the advantages that
1, the defect classification of different location can be merged by same convolutional neural networks exports.Two groups of camera acquisitions can be simultaneously
Row, which calculates, reduces detection time-consuming, improves efficiency.
2, by the method for deep learning, neural network can learn image data feature and be subject to extensive.Avoid traditional calculation
Method coupling superposition, final detection effect are also much better than traditional algorithm.
Detailed description of the invention
Fig. 1 is the pole-piece pole-ear exemplary diagram of power lithium-ion battery welding；
Fig. 2 is dynamical lithium-ion battery lug exemplary diagram；
Fig. 3 is pole piece defects detection model structure；
Fig. 4 is model training flow chart；
Fig. 5 is 1 pole-piece pole-ear defect detecting system structure chart of embodiment；
Fig. 6 is dynamical lithium-ion battery lug and tab defect exemplary diagram, and wherein a is tab breach, and b is tab gap, c
For pole piece particle, d is pole piece bubble, and e is that pole piece reveals foil.
Specific embodiment
Present invention will be further explained below with reference to specific examples.It should be understood that these embodiments are merely to illustrate the present invention
Rather than it limits the scope of the invention.In addition, it should also be understood that, after reading the content taught by the present invention, those skilled in the art
Member can various modifications may be made or change, such equivalent forms equally fall within the application the appended claims and limited to the present invention
Range.
Embodiment 1
A kind of pole-piece pole-ear defect detecting system based on convolutional neural networks, comprising:
Image data acquisition unit, it is parallel using line scanning industrial camera, it is adopted in the tow sides of the pole-piece pole-ear of welding
Collect image, and passes through cable transmission to industrial personal computer.
Pre-processing image data unit is divided into two class data of pole piece and tab by double gauss difference.
Detection unit constructs pole-piece pole-ear defects detection model using the method that embodiment 2 is recorded, by two groups of image datas
Conversion Cheng Zhangliang data simultaneously input the pole-piece pole-ear defects detection model.
As a result output unit exports tab classification results, pole piece defective locations classification results, frame after testing after unit
Coordinate, confidence level, and qualified and unqualified conclusion is provided by rating scale.
Embodiment 2
A kind of pole-piece pole-ear defect inspection method based on convolutional neural networks, comprising:
S1: the image data of acquisition is passed through cable transmission to industrial personal computer by acquisition image.
S2: pre-processing image data, is divided into two class data of pole piece and tab by double gauss difference.
S3: at tensor data and inputting pole-piece pole-ear defects detection model for two groups of image data conversions of step S2, should
Model building method is as follows:
(1) image preprocessing: input picture is subjected to double gauss difference, is divided into two class data of pole piece and tab.
(2) data mark: position mark are carried out to the defects of the pole piece image data that step (1) segmentation obtains, to pole
Ear image data carries out classification annotation.
(3) building of neural network: four convolutional Neural models of design, model 1 are to extract image fixed dimension feature
Neural network model；Model 2 adds classification layer, for classifying to tab defect characteristic, model 3 on the basis of model 1
Multiple dimensioned boundary recurrence is carried out to determine anchor frame orientation to the feature that pole piece defects detection goes out on the basis of model 1, then is passed through
Non-maxima suppression method determines Main Boundaries framework；It is defeated that model 4 returns main body to pole piece defect boundary on the basis of model 1
Feature vector out is classified；
Model 1: the Artificial Neural Network Structures for extracting image fixed dimension spy are as follows:
Model 3,4: pole piece defects detection model structure is shown in attached drawing 3.
It regard model 1 as image fixed dimension Feature Selection Model, regard model 2 as tab defect classification model, model
3,4 become pole piece defects detection model by Multiscale Fusion.
(4) model training: respectively by pole piece image data and tab image data be divided into 80% training dataset with
20% validation data set, using multireel lamination, the method for small convolution kernel, using training set and verifying collection respectively to tab defect
Disaggregated model and pole piece defects detection model are trained and verify after each iteration to training set.Model training process
Figure is shown in attached drawing 4.
(5) it creates loss function: being directed to tab defect classification model, create Softmax cross entropy loss function, be used for table
Levy the difference between authentic specimen and prediction probability；For pole piece defects detection model, Huber loss function is created, based on
It calculates bounding box and returns main body；Then the two losses are added and obtain final loss function.
(6) model evaluation: evaluating tab defect classification results using accuracy rate, to the pre- of pole piece defects detection bounding box
It surveys using mean absolute error come evaluation result；If tab defect classification results Average Accuracy is less than 98.5%, pole piece defect
The prediction of detection bounding box is averaged recall rate less than 98.5%, then adjusts hyper parameter according to rear re -training.
(7) Model Fusion: obtaining the optimal models of tab defect classification and pole piece defects detection according to model evaluation result,
It is fused into an optimum fusion model using two kinds of models as submodel by concentrate function, to accelerate to detect speed.
(8) tab and pole piece image data that are partitioned by image preprocessing data reasoning: are converted into tensor data
The optimum fusion model for entering step (7) simultaneously, during data are propagated forward, the tensor data of tab enter tab and lack
Classification submodel is fallen into, the tensor data of pole piece enter pole piece defects detection submodel, and finally the adjustment Jing Guo confidence threshold value is defeated
Testing result out.
S4: the tensor data of tab export tab classification results by tab defect categorical reasoning；The tensor data of pole piece
By pole piece defects detection, defective locations classification results, frame coordinate, confidence level are exported.
S5: product is divided by qualified and unqualified two kinds of categories by rating scale.
The present invention just has been combined loss function during model training with model evaluation result to decide whether chasing after
Addend according to or the super ginseng retraining of adjustment, the convolutional neural networks model for finally taking classification optimal with border effect merged after when
Make final network model.Therefore defect reasoning process kind by confidence threshold value setting excluded 98.5% or more it is not true
Determine factor, therefore the method for the present invention is more than or equal to 98.5% to the accuracy rate degree of defects detection.
Embodiment 3
Test company: Tianjin Jiewei Power Industry Co., Ltd.
CompanyAddress: Tianjin Xiqing District auto industry garden open source road 11
10 unqualified are mixed with 10 qualified pole pieces with tab combination sample are chosen to put, wherein No. 1
Pole piece is to reveal thin defect, and No. 5 pole pieces have a dark trace defect, and No. 6 pole pieces have an air blister defect, and No. 9 pole pieces have a grain defect, 10
Number tab have breach defect, No. 13 tabs have bending defect, and No. 14 tabs have wrinkle defect, and No. 17 tabs have scratch
Defect, No. 19 tabs and pole piece respectively have fold and reveal thin defect, and No. 20 tabs and pole piece respectively have bending and dark trace defect.Benefit
20 products are detected with the method for embodiment 2.
Specific detection process is as follows:
S1: the image data of acquisition is passed through cable transmission to industrial personal computer by acquisition image；
S2: pre-processing image data, is divided into two class data of pole piece and tab by double gauss difference；
S3: two groups of image data conversions of step S2 at tensor data and are inputted into the method building that embodiment 2 is recorded
Optimum fusion model；
S4: the tensor data of tab export tab classification results by tab defect categorical reasoning；The tensor data of pole piece
By pole piece defects detection, defective locations classification results, frame coordinate, confidence level are exported；
S5: product is divided by qualified and unqualified two kinds of categories by rating scale.
Testing result shows that No. 1 pole piece is to reveal thin defect in 20 products, and No. 5 pole pieces have dark trace defect, No. 6
Pole piece has air blister defect, and No. 9 pole pieces have grain defect, and No. 10 tabs have breach defect, and No. 13 tabs have bending defect,
No. 14 tabs have wrinkle defect, and No. 17 tabs have scratch defects, and No. 19 tabs and pole piece respectively have fold and reveal thin defect,
No. 20 tabs and pole piece respectively have bending and dark trace defect, remaining is qualified product and identical as presetting, and shows side of the invention
Case can be used for the defects detection of pole-piece pole-ear.
Claims (7)
1. the construction method of the pole-piece pole-ear defects detection model based on convolutional neural networks characterized by comprising
(1) image preprocessing: input picture is subjected to double gauss difference, is divided into two class data of pole piece and tab；
(2) data mark: position mark are carried out to the defects of the pole piece image data that step (1) segmentation obtains, to tab figure
As data carry out classification annotation；
(3) building of neural network: four convolutional Neural models of design, model 1 are the nerve for extracting image fixed dimension feature
Network model；Model 2 adds classification layer on the basis of model 1, and for classifying to tab defect characteristic, model 3 is in mould
Multiple dimensioned boundary recurrence is carried out to the feature that pole piece defects detection goes out to determine anchor frame orientation on the basis of type 1, then passes through non-pole
Big value suppressing method determines Main Boundaries framework；Model 4 returns main body output to pole piece defect boundary on the basis of model 1
Feature vector is classified；
It regard model 1 as image fixed dimension Feature Selection Model, regard model 2 as tab defect classification model, model 3,4 is logical
Cross multiple dimensioned be fused into as pole piece defects detection model；
(4) model training model training: is carried out to tab defect classification model and pole piece defects detection model；
(5) it creates loss function: being directed to tab defect classification model and pole piece defects detection model creation loss function respectively, it will
Two loss additions obtain final loss function；
(6) model evaluation: tab defect classification model and pole piece defects detection model are assessed respectively；
(7) Model Fusion: the optimal models of tab defect classification and pole piece defects detection are obtained according to model evaluation result, are passed through
Two kinds of models are fused into an optimum fusion model by concentrate function, to accelerate to detect speed；
(8) tab and pole piece image data that are partitioned by image preprocessing data reasoning: are converted into tensor data simultaneously
The optimum fusion model for entering step (7), during data are propagated forward, the tensor data of tab enter tab defect point
Class submodel, the tensor data of pole piece enter pole piece defects detection submodel, and finally the adjustment Jing Guo confidence threshold value exports inspection
Survey result.
2. construction method according to claim 1, which is characterized in that in step (4), respectively by pole piece image data and pole
Ear image data is divided into 80% training dataset and 20% validation data set, using multireel lamination, the method for small convolution kernel,
Tab defect classification model and pole piece defects detection model are trained respectively using training set and verifying collection and changed every time
Training set is verified after generation.
3. construction method according to claim 1, which is characterized in that in step (5), for tab defect classification model,
Softmax cross entropy loss function is created, for characterizing the difference between authentic specimen and prediction probability；It is examined for pole piece defect
Model is surveyed, Huber loss function is created, returns main body for calculating bounding box；Then the two losses are added and are obtained finally
Loss function.
4. construction method according to claim 1, which is characterized in that in step (6), tab is evaluated using accuracy rate and is lacked
Classification results are fallen into, to the prediction of pole piece defects detection bounding box using mean absolute error come evaluation result；If tab defect point
For class result Average Accuracy less than 98.5%, the prediction of pole piece defects detection bounding box is averaged recall rate less than 98.5%, then adjusts
Whole hyper parameter is according to rear re -training.
5. the pole-piece pole-ear defect inspection method based on convolutional neural networks characterized by comprising
S1: the image data of acquisition is passed through cable transmission to industrial personal computer by acquisition image；
S2: pre-processing image data, is divided into two class data of pole piece and tab by double gauss difference；
S3: by two groups of image data conversions of step S2 at tensor data and the building that any one of inputs claim 1-4 it is optimal
Fusion Model；
S4: the tensor data of tab export tab classification results by tab defect categorical reasoning；The tensor data of pole piece pass through
Pole piece defects detection exports defective locations classification results, frame coordinate, confidence level；
S5: product is divided by qualified and unqualified two kinds of categories by rating scale.
6. pole-piece pole-ear defect inspection method according to claim 5, which is characterized in that in step S1, in pole-piece pole-ear
Tow sides using line scan industrial camera parallel collection image.
7. the pole-piece pole-ear defect detecting system based on convolutional neural networks comprising:
Image data acquisition unit, it is parallel using line scanning industrial camera, figure is acquired in the tow sides of the pole-piece pole-ear of welding
Picture, and pass through cable transmission to industrial personal computer；
Pre-processing image data unit is divided into two class data of pole piece and tab by double gauss difference；
Detection unit constructs pole-piece pole-ear defects detection model using the described in any item methods of claim 1-4, by two group pictures
As data convert Cheng Zhangliang data and input the pole-piece pole-ear defects detection model；
As a result output unit, exports tab classification results after testing after unit, pole piece defective locations classification results, frame are sat
Mark, confidence level, and qualified and unqualified conclusion is provided by rating scale.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910450580.7A CN110135521A (en) | 2019-05-28 | 2019-05-28 | Pole-piece pole-ear defects detection model, detection method and system based on convolutional neural networks |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910450580.7A CN110135521A (en) | 2019-05-28 | 2019-05-28 | Pole-piece pole-ear defects detection model, detection method and system based on convolutional neural networks |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110135521A true CN110135521A (en) | 2019-08-16 |
Family
ID=67582087
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910450580.7A Withdrawn CN110135521A (en) | 2019-05-28 | 2019-05-28 | Pole-piece pole-ear defects detection model, detection method and system based on convolutional neural networks |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110135521A (en) |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111274985A (en) * | 2020-02-06 | 2020-06-12 | 咪咕文化科技有限公司 | Video text recognition network model, video text recognition device and electronic equipment |
| CN111307818A (en) * | 2020-02-25 | 2020-06-19 | 华南理工大学 | Visual online detection method for laser welding spot of lithium battery tab |
| CN111696108A (en) * | 2020-08-17 | 2020-09-22 | 广东利元亨智能装备股份有限公司 | Model training method, welding spot defect detection method and device and electronic equipment |
| CN111815572A (en) * | 2020-06-17 | 2020-10-23 | 深圳市大德激光技术有限公司 | Method for detecting welding quality of lithium battery based on convolutional neural network |
| CN112017150A (en) * | 2020-04-30 | 2020-12-01 | 河南爱比特科技有限公司 | Intelligent visual detection method and equipment for surface defects of lithium ion battery pole piece |
| CN112270429A (en) * | 2020-08-31 | 2021-01-26 | 中国科学院合肥物质科学研究院 | Cloud edge cooperation-based power battery pole piece manufacturing equipment maintenance method and system |
| CN113129257A (en) * | 2019-12-30 | 2021-07-16 | 美光科技公司 | Apparatus and method for determining wafer defects |
| CN113450302A (en) * | 2020-03-24 | 2021-09-28 | 东莞新能德科技有限公司 | Tab detection method and device based on machine learning and computer readable storage medium |
| WO2021248554A1 (en) * | 2020-06-11 | 2021-12-16 | 深圳市信宇人科技股份有限公司 | High-speed and high-precision burr detection method and detection system for lithium ion battery electrode sheet |
| CN113822281A (en) * | 2020-06-19 | 2021-12-21 | 富士通株式会社 | Apparatus, method and storage medium for multi-objective optimization |
| DE102020207613A1 (en) | 2020-06-19 | 2021-12-23 | Volkswagen Aktiengesellschaft | Method for evaluating a cutting edge of a body |
| CN117710376A (en) * | 2024-02-05 | 2024-03-15 | 宁德时代新能源科技股份有限公司 | Tab defect detection method and device and electronic equipment |
-
2019
- 2019-05-28 CN CN201910450580.7A patent/CN110135521A/en not_active Withdrawn
Cited By (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113129257A (en) * | 2019-12-30 | 2021-07-16 | 美光科技公司 | Apparatus and method for determining wafer defects |
| US11922613B2 (en) | 2019-12-30 | 2024-03-05 | Micron Technology, Inc. | Apparatuses and methods for determining wafer defects |
| CN111274985B (en) * | 2020-02-06 | 2024-03-26 | 咪咕文化科技有限公司 | Video text recognition system, video text recognition device and electronic equipment |
| CN111274985A (en) * | 2020-02-06 | 2020-06-12 | 咪咕文化科技有限公司 | Video text recognition network model, video text recognition device and electronic equipment |
| CN111307818A (en) * | 2020-02-25 | 2020-06-19 | 华南理工大学 | Visual online detection method for laser welding spot of lithium battery tab |
| CN113450302A (en) * | 2020-03-24 | 2021-09-28 | 东莞新能德科技有限公司 | Tab detection method and device based on machine learning and computer readable storage medium |
| CN112017150A (en) * | 2020-04-30 | 2020-12-01 | 河南爱比特科技有限公司 | Intelligent visual detection method and equipment for surface defects of lithium ion battery pole piece |
| WO2021248554A1 (en) * | 2020-06-11 | 2021-12-16 | 深圳市信宇人科技股份有限公司 | High-speed and high-precision burr detection method and detection system for lithium ion battery electrode sheet |
| CN111815572B (en) * | 2020-06-17 | 2022-03-08 | 深圳市大德激光技术有限公司 | Method for detecting welding quality of lithium battery based on convolutional neural network |
| CN111815572A (en) * | 2020-06-17 | 2020-10-23 | 深圳市大德激光技术有限公司 | Method for detecting welding quality of lithium battery based on convolutional neural network |
| CN113822281A (en) * | 2020-06-19 | 2021-12-21 | 富士通株式会社 | Apparatus, method and storage medium for multi-objective optimization |
| DE102020207613A1 (en) | 2020-06-19 | 2021-12-23 | Volkswagen Aktiengesellschaft | Method for evaluating a cutting edge of a body |
| CN111696108B (en) * | 2020-08-17 | 2021-07-09 | 广东利元亨智能装备股份有限公司 | Model training method, welding spot defect detection method and device and electronic equipment |
| CN111696108A (en) * | 2020-08-17 | 2020-09-22 | 广东利元亨智能装备股份有限公司 | Model training method, welding spot defect detection method and device and electronic equipment |
| CN112270429A (en) * | 2020-08-31 | 2021-01-26 | 中国科学院合肥物质科学研究院 | Cloud edge cooperation-based power battery pole piece manufacturing equipment maintenance method and system |
| CN117710376A (en) * | 2024-02-05 | 2024-03-15 | 宁德时代新能源科技股份有限公司 | Tab defect detection method and device and electronic equipment |
| CN117710376B (en) * | 2024-02-05 | 2024-06-07 | 宁德时代新能源科技股份有限公司 | Tab defect detection method and device and electronic equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110135521A (en) | Pole-piece pole-ear defects detection model, detection method and system based on convolutional neural networks | |
| CN110543878B (en) | Pointer instrument reading identification method based on neural network | |
| CN109919934B (en) | Liquid crystal panel defect detection method based on multi-source domain deep transfer learning | |
| CN103324937B (en) | The method and apparatus of label target | |
| CN111368690B (en) | Deep learning-based video image ship detection method and system under influence of sea waves | |
| CN108711148B (en) | Tire defect intelligent detection method based on deep learning | |
| CN110910350B (en) | Nut loosening detection method for wind power tower cylinder | |
| CN111476307B (en) | Lithium battery surface defect detection method based on depth field adaptation | |
| CN116385958A (en) | Edge intelligent detection method for power grid inspection and monitoring | |
| CN115131747A (en) | Knowledge distillation-based power transmission channel engineering vehicle target detection method and system | |
| CN111783616B (en) | Nondestructive testing method based on data-driven self-learning | |
| CN115984543A (en) | Target detection algorithm based on infrared and visible light images | |
| Song et al. | Surface defect detection of aeroengine blades based on cross-layer semantic guidance | |
| Ying et al. | Automatic detection of photovoltaic module cells using multi-channel convolutional neural network | |
| CN117495891B (en) | Point cloud edge detection method and device and electronic equipment | |
| CN112837281B (en) | Pin defect identification method, device and equipment based on cascade convolution neural network | |
| Jiang et al. | MA-YOLO: a method for detecting surface defects of aluminum profiles with attention guidance | |
| CN106814088A (en) | Based on machine vision to the detection means and method of cell piece colour sorting | |
| CN113724233B (en) | Transformer equipment appearance image defect detection method based on fusion data generation and transfer learning technology | |
| Xu et al. | Research on improved residual network classification method for defect recognition of thermal battery | |
| CN115294061A (en) | Method and system for detecting surface defects of paper-plastic composite bag based on machine vision | |
| CN109785261A (en) | A kind of airborne LIDAR three-dimensional filtering method based on gray scale volume element model | |
| Zhang et al. | Research on image defect detection of silicon panel based on prewitt and canny operator | |
| CN115100098A (en) | Printed circuit board AOI intelligent detection equipment based on deep learning algorithm | |
| CN113723400A (en) | Electrolytic cell polar plate fault identification method, system, terminal and readable storage medium based on infrared image |
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 | ||
| WW01 | Invention patent application withdrawn after publication | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20190816 |