CN110111332A - Collagent casing for sausages defects detection model, detection method and system based on depth convolutional neural networks - Google Patents
Collagent casing for sausages defects detection model, detection method and system based on depth convolutional neural networks Download PDFInfo
- Publication number
- CN110111332A CN110111332A CN201910421030.2A CN201910421030A CN110111332A CN 110111332 A CN110111332 A CN 110111332A CN 201910421030 A CN201910421030 A CN 201910421030A CN 110111332 A CN110111332 A CN 110111332A
- Authority
- CN
- China
- Prior art keywords
- model
- casing
- defect
- sausages
- collagent
- 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- Withdrawn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
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/136—Segmentation; Edge detection involving thresholding
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Biophysics (AREA)
- General Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Quality & Reliability (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The present invention relates to Collagent casing for sausages defects detections, and in particular to Collagent casing for sausages defects detection model, detection method and system based on depth convolutional neural networks, model construction: data mark；Convolutional neural networks building: model 1 is the basic model based on residual error network, is used to extract casing image data structure feature from raw image data；Model 2 and model 3 are down-sampling model, for obtaining the broader feature visual field of casing；Model 4 is casing defect class prediction model, for the tagsort to extraction；Model 5 is casing defect bounding box rendering model, generates anchor frame to the characteristic pattern extracted for basis and predicts classification and offset；Model training；Construct loss function；Model evaluation；Model reasoning: removing similar predicted boundary frame, export after confidence threshold value adjusts, and detection method and system are according to the model construction.The present invention improves casing defects detection efficiency and accuracy rate.
Description
Technical field
The present invention relates to Collagent casing for sausages defects detections, and in particular to the collagen based on depth convolutional neural networks
Casing defects detection model, detection method and system.
Background technique
Deep learning is with the multistage representative learning method indicated.Since initial data, in every level-one, deep learning
The expression of this grade is transformed to more advanced expression by simple function.Therefore, deep learning model also can be regarded as by being permitted
The function that more simple functions are combined.When compound function is enough, deep learning model can express extremely complex change
It changes.
Convolutional neural networks are using raw image data as input, successively alternately across the convolution of convolutional layer and pond layer
The operation such as pond, finally passes through full articulamentum.Network weight is updated using back-propagation algorithm by the loss of learning tasks
And biasing, training network model, then picture is imported in network model and extracts characteristic of division, to complete the function of identification.
Compared to traditional Collagent casing for sausages detection method, the detection method based on convolutional neural networks can be in image data extraction to higher-dimension
Feature is spent, image data can be coped in this way and deviated in a degree of dimensional variation and deformation with form.Also, by from a large amount of
The original feature of the Collagent casing for sausages that data sample kind learns, it is ensured that feature separability with higher.Therefore it is rolling up
In product neural network, satisfactory testing result can be obtained using simple loss function.
Collagent casing for sausages often has individual existing defects in production, as shown in Figure 3.Artificial defect detection scheme due to
Sausage casing contracting machine revolving speed is too fast, leads to not real-time detection, therefore can only be rechecked after casing lasso trick, and casing is long after lasso trick
Degree will be compressed to 20 centimetres or so by 18 meters to 22 meters, it is difficult to find lesser defect, therefore omission factor is very big.
The appearance quality detection of conventional machines vision Collagent casing for sausages at present, by industrial camera, tubular light source, industrial personal computer
And programmed algorithm is constituted.But due to being inflated condition before sausage casing contracting, 3 groups of cameras is needed to cooperate to meet testing requirements.
So, corresponding every group of camera can all have its specific algorithm parameter, and will form a set of complexity after the superposition of these parameters
Adjustment relationship.Since casing diameter has 36 nearly 20 seed type from Φ 18 to Φ, so system needs storage correspondence each type of
Hundreds of parameters, and need operator that there is very strong priori knowledge deposit could debug to system.
Although conventional machines visible sensation method can compensate manually can not real-time detection short slab, due to algorithm complexity
Do, needing operator to have very strong priori knowledge could adjust, therefore conventional machines vision-based detection go out defect exist with
The characteristics of machine is strong, poor reliability, and the numerical value being calculated by machine vision algorithm can not do the short slab of classification judgement.
Summary of the invention
The purpose of the present invention is overcoming drawbacks described above, a kind of Collagent casing for sausages based on depth convolutional neural networks is provided
Defects detection model, detection method and system need complicated priori knowledge to come manually to solve existing conventional machines vision-based detection
There are randomnesss to extract casing defect characteristic and conventional machines vision-based detection strong, poor reliability, the adjustment of adjustment vision algorithm
Parameter is excessive, it is difficult to the problems such as covering all kinds.
Technical solution of the present invention: the building of the Collagent casing for sausages defects detection model based on depth convolutional neural networks
Method, comprising:
(1) data mark: being labeled to the defects of acquired image data, mark defect respectively with two o'clock coordinate
Coordinate maximum value and coordinate minimum value, mark its defect kind；
(2) building of convolutional neural networks: five convolutional Neural models of design, model 1 are the basis based on residual error network
Model is used to extract casing image data structure feature from raw image data；Model 2 and model 3 are down-sampling model, are used
To obtain the broader feature visual field of casing；Model 4 is casing defect class prediction model, for the tagsort to extraction；
Model 5 is casing defect bounding box rendering model, for according to generating anchor frame to the characteristic pattern that extracts and predict classification and partially
Shifting amount；
(3) model training: five convolutional Neural model combinations are become into a final casing defects detection model, are gone forward side by side
Row model training；
(4) loss function is constructed；
(5) model evaluation: evaluating casing defect classification results using accuracy rate, adopts to the prediction of casing defect bounding box
With mean absolute error evaluation result；
(6) model reasoning: obtaining casing defect anchor frame according to model 5 and the offset of prediction obtain predicted boundary frame, and
Merge similar predicted boundary frame by non-maxima suppression method, finally using the adjustment final output intestines of confidence threshold value
Clothing fault detection data result.
Further, in step (2), model 2 and model 3 are down-sampling model, by RELU activation primitive and batch normalization
The convolutional layer that composition and convolution kernel are 3*3, then a plus maximum pond layer composition.
Further, in step (2), model 4 and model 5 are respectively that categorical measure is identical and convolution by 5 input channels
Core is the convolutional layer of 3*3, and the convolutional layer that and convolution kernel identical as offset is 3*3 forms.
Further, in step (3), using multireel lamination, data set is divided into 80% training by the method for small convolution kernel
Collection is trained it with 20% verifying collection and verifies after each iteration to training set.
Further, step (4) constructs loss function: for the tagsort of model 4, creation intersects loss entropy function,
For the bounding box offset of model 5, the loss of L1 norm is created, then the two losses are added and obtain final loss function.
The present invention also provides the Collagent casing for sausages defect inspection methods based on depth convolutional neural networks, comprising:
S1: three groups of image datas of casing are obtained；
S2: operation is successively normalized in three groups of image datas of step S1 and is converted into tensor data；
S3: the tensor data of step S2 are inputted into trained Collagent casing for sausages defects detection model, collagen intestines
Clothing defects detection model reasoning output category result, defective locations frame coordinate and confidence level；The Collagent casing for sausages defect
Detection model is constructed according to mentioned-above method；
S4: the classification results that will be greater than confidence threshold value are closed with defective locations frame coordinate by non-maxima suppression method
And obtain final the reasoning results；
S5: the casing defect image data inside frame coordinate are extracted, and Threshold segmentation is carried out to it, calculate defect
Pixel value and coordinate value；
S6: according to the proportionate relationship of camera lens focal length and camera CMOS pixel, the pixel value of casing defect part is changed
It is counted as actual size, so that dialogue casing presentation quality is objectively evaluated.
Further, in step S1, casing is shot simultaneously with 120 ° of angles respectively using three groups of industrial cameras, acquires casing
360 ° of panoramic image datas, and be transmitted in industrial personal computer by data cable.
Further, in step S4, the threshold value of confidence level is 97%.
The present invention also provides the Collagent casing for sausages defect detecting systems based on depth convolutional neural networks, comprising:
Image data acquisition module shoots casing respectively using three groups of industrial cameras with 120 ° of angles simultaneously, acquires casing
360 ° of panoramic image datas, and be transmitted in industrial personal computer by data cable.
Three groups of image datas of image data acquisition module are successively normalized operation and turned by graph data conversion module
Change tensor data into；
Collagent casing for sausages defects detection module designs Collagent casing for sausages defects detection mould using mentioned-above method
Tensor data are inputted trained Collagent casing for sausages defects detection model and detected by type, output category result, defective bit
Set frame coordinate and confidence level；
Result treatment module, classification results and defective locations frame coordinate by confidence level greater than 97% pass through non-maximum
Suppressing method merges, and obtains final the reasoning results；
Casing defect image data inside frame coordinate are extracted, and carry out Threshold segmentation to it by evaluation module, calculate
The pixel value and coordinate value of defect out；And according to the proportionate relationship of camera lens focal length and camera CMOS pixel, by casing defect
Partial pixel value is converted into actual size, and dialogue casing presentation quality is objectively evaluated.
The invention has the advantages that
1, convolutional neural networks can be classified with multi input, and 3 groups of camera acquired image data can enter convolution mind simultaneously
Differentiated through network model.
2, parameter adjustment is carried out without personnel.It can learn characteristics of image when neural metwork training automatically, avoid manual debugging,
Final detection effect is also much better than conventional method, and accuracy rate is higher.
Detailed description of the invention
Fig. 1 is five convolutional neural networks model structures of depth convolutional neural networks；
Fig. 2 is the structure setting figure of model 2 and model 3；
Fig. 3 is Collagent casing for sausages defect map；
Fig. 4 is Collagent casing for sausages defects detection result figure；
Fig. 5 is collagen detection system block diagram；
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
Collagent casing for sausages defect detecting system (as shown in Figure 5) based on depth convolutional neural networks, comprising:
Image data acquisition module shoots casing respectively using three groups of industrial cameras with 120 ° of angles simultaneously, acquires casing
360 ° of panoramic image datas, and be transmitted in industrial personal computer by data cable；
Three groups of image datas of image data acquisition module are successively normalized operation and turned by graph data conversion module
Change tensor data into；
Collagent casing for sausages defects detection module is examined using method as described in example 2 design Collagent casing for sausages defect
Model is surveyed, tensor data are inputted into trained Collagent casing for sausages defects detection model and are detected, output category result lacks
Fall into position frame coordinate and confidence level；
Result treatment module, classification results and defective locations frame coordinate by confidence level greater than 97% pass through non-maximum
Suppressing method merges, and obtains final the reasoning results；
Casing defect image data inside frame coordinate are extracted, and carry out Threshold segmentation to it by evaluation module, calculate
The pixel value and coordinate value of defect out；And according to the proportionate relationship of camera lens focal length and camera CMOS pixel, by casing defect
Partial pixel value is converted into actual size, and dialogue casing presentation quality is objectively evaluated.
Embodiment 2
Collagent casing for sausages defect inspection method based on depth convolutional neural networks, comprising:
S1: obtaining three groups of image datas of casing, shoots intestines simultaneously respectively using three groups of industry phase cameras with 120 ° of angles
Clothing acquires 360 ° of casing panoramic image datas, and is transmitted in industrial personal computer by data cable.
S2: operation is successively normalized in three groups of image datas of step S1 and is converted into tensor data.
S3: the tensor data of step S2 are inputted into trained Collagent casing for sausages defects detection model, collagen intestines
Clothing defects detection model reasoning output category result, defective locations frame coordinate and confidence level；The Collagent casing for sausages defect
The construction method of detection model is as follows:
(1) data mark: being labeled to the defects of acquired image data, mark defect respectively with two o'clock coordinate
Coordinate maximum value and coordinate minimum value, mark its defect kind.
(2) building of convolutional neural networks: five convolutional Neural models of design, model 1 are the basis based on residual error network
Model is used to extract casing image data structure feature from raw image data；Model 2 and model 3 are down-sampling model, are used
To obtain the broader feature visual field of casing；Model 4 is casing defect class prediction model, for the tagsort to extraction；
Model 5 is casing defect bounding box rendering model, for according to generating anchor frame to the characteristic pattern that extracts and predict classification and partially
Shifting amount；Five convolutional neural networks relationship models are shown in Fig. 1.
Model 1: the basic model based on residual error network, model are set as 18 layers of structure and are as follows:
Model 2 and model 3 are down-sampling model, are made of RELU activation primitive with batch normalization and convolution kernel is 3*3's
Convolutional layer, then plus a maximum pond layer form, structure is shown in Fig. 2.
Model 4 and model 5 are respectively categorical measure is identical and convolution kernel is 3*3 convolutional layer by 5 input channels, with
Offset is identical and convolution kernel forms for the convolutional layer of 3*3.
(3) five convolutional Neural model combinations model training: are become into a final Collagent casing for sausages defects detection
Model, and multireel lamination is used, the training set and 20% verifying collection that data set is divided into 80% by the method for small convolution kernel are to it
It is trained and training set is verified after each iteration.
(4) it constructs loss function: for the tagsort of model 4, cross entropy loss function is created, for the side of model 5
Boundary's frame offset, creation L1 norm loss, then does add operation for what the two loss functions obtained numerical value.
(5) model evaluation: evaluating casing defect classification results using accuracy rate, adopts to the prediction of casing defect bounding box
With mean absolute error evaluation result.
(6) model reasoning: obtaining casing defect anchor frame according to model 5 and the offset of prediction obtain predicted boundary frame, and
Merge similar predicted boundary frame by non-maxima suppression method, finally using the adjustment final output intestines of confidence threshold value
Clothing fault detection data result.
S4: the classification results by confidence level greater than 97% are closed with defective locations frame coordinate by non-maxima suppression method
And obtain final the reasoning results.
S5: the casing defect image data inside frame coordinate are extracted, and Threshold segmentation is carried out to it, calculate defect
Pixel value and coordinate value.
S6: according to the proportionate relationship of camera lens focal length and camera CMOS pixel, the pixel value of casing defect part is changed
It is counted as actual size, so that dialogue casing presentation quality is objectively evaluated.
The present invention just has been combined loss function during casing defects detection model training and model evaluation result is come
Decide whether finally to take classification and the optimal depth convolutional neural networks of border effect in supplemental data mark and model retraining
Model is as final network model.Therefore the uncertain factor that 97% or more has been excluded in casing defect reasoning process, because
This method for the present invention is more than or equal to 97% to the accuracy of casing defects detection.
Embodiment 3
Test company: Wuzhou Shenguan Protein Casing Co., Ltd.
CompanyAddress: Guangxi Zhuang Autonomous Region Wuzhou Wanxiu District trial zone Yue Gui industry park
Choosing 50 rice glue diameters is 26 millimeters of Collagent casing for sausages, wherein successively 10 meters, 22 meters, 37 meters of position point
Others is manufacture hair, 3 kinds of greasy dirt, stain defects, is carried out using the method for embodiment 2 to the 50 rice glue collagen albumen sausage casing continuous
Detection.
For the Collagent casing for sausages shown in a in Fig. 3, specific detection process is as follows:
S1: obtaining three groups of image datas, using three groups of industrial cameras respectively to take the photograph casing when 120 ° of angle beating-ins, acquires intestines
360 ° of panoramic image datas of clothing, and be transmitted in industrial personal computer by data cable.
S2: operation is successively normalized in three groups of image datas of step S1 and is converted into tensor data.
S3: the tensor data of step S2 are inputted into trained casing defects detection model, casing defects detection model pushes away
Manage output category result, defective locations frame coordinate and confidence level；
S4: the classification results by confidence level greater than 97% are merged with defective locations frame by non-maxima suppression method,
Obtain final the reasoning results.
S5: the casing defect image data inside frame coordinate are extracted, and Threshold segmentation is carried out to it, calculate defect
Pixel value and coordinate value.
S6: according to the proportionate relationship of camera lens focal length and camera CMOS pixel, the pixel value of casing defect part is changed
It is counted as actual size, so that dialogue casing presentation quality is objectively evaluated
Testing result is shown, sees Fig. 4, and the defect that yw, hd, mf frame select in image respectively corresponds greasy dirt, stain, hair；?
10 meters, 17 meters, 22 meters, 37 meters respectively have 3 kinds of hair, greasy dirt, stain defects, identical as presetting, and defect type as the result is shown,
Defective locations are correct.
Claims (9)
1. the construction method of the Collagent casing for sausages defects detection model based on depth convolutional neural networks, which is characterized in that packet
It includes:
(1) image labeling: being labeled the defects of acquired image data, marks the seat of defect respectively with two o'clock coordinate
Maximum value and coordinate minimum value are marked, its defect kind is marked；
(2) building of convolutional neural networks: five convolutional Neural models of design, model 1 are the basic model based on residual error network,
For extracting casing image data structure feature from raw image data；Model 2 and model 3 are down-sampling model, for obtaining
Take the feature visual field that casing is broader；Model 4 is casing defect class prediction model, for the tagsort to extraction；Model 5
For casing defect bounding box rendering model, anchor frame is generated to the characteristic pattern extracted for basis and predicts classification and offset；
(3) model training: five convolutional neural networks model splicings, which are merged, becomes a final casing defects detection model,
And carry out model training；
(4) loss function is constructed；
(5) model evaluation: evaluating casing defect classification results using accuracy rate, uses to the prediction of casing defect bounding box flat
Equal absolute error evaluation result；
(6) model reasoning: casing defect anchor frame is obtained according to model 5 and the offset of prediction obtains predicted boundary frame, and is passed through
Non-maxima suppression method merges similar predicted boundary frame, finally lacks using the adjustment final output casing of confidence threshold value
Fall into detection data result.
2. construction method according to claim 1, which is characterized in that in step (2), model 2 and model 3 are down-sampling mould
Type, is made of RELU activation primitive with batch normalization and convolution kernel is the convolutional layer of 3*3, then a plus maximum pond layer composition.
3. construction method according to claim 1, which is characterized in that in step (2), model 4 and model 5, respectively by 5
Input channel is the convolutional layer that categorical measure is identical and convolution kernel is 3*3, and convolution kernel identical as offset is the convolutional layer of 3*3
Composition.
4. construction method according to claim 1, which is characterized in that in step (3), using multireel lamination, small convolution kernel
Method by data set be divided into 80% training set and 20% verifying collection it is trained and after each iteration to training set
It is verified.
5. construction method according to claim 1, which is characterized in that step (4) constructs loss function: for model 4
Tagsort, creation intersects loss entropy function, for the bounding box offset of model 5, creates the loss of L1 norm, then by this two
A loss addition obtains final loss function.
6. the Collagent casing for sausages defect inspection method based on depth convolutional neural networks characterized by comprising
S1: three groups of image datas of casing are obtained；
S2: operation is successively normalized in three groups of image datas of step S1 and is converted into tensor data；
S3: the tensor data of step S2 are inputted into trained Collagent casing for sausages defects detection model, Collagent casing for sausages lacks
Fall into detection model reasoning output category result, defective locations frame coordinate and confidence level；The Collagent casing for sausages defects detection
Model method building according to claim 1-5；
S4: the classification results that will be greater than confidence threshold value are merged with defective locations frame coordinate by non-maxima suppression method,
Obtain final the reasoning results；
S5: the casing defect image data inside frame coordinate are extracted, and Threshold segmentation is carried out to it, calculate the picture of defect
Element value and coordinate value；
S6: according to the proportionate relationship of camera lens focal length and camera CMOS pixel, the pixel value of casing defect part is converted into
Actual size, so that dialogue casing presentation quality is objectively evaluated.
7. Collagent casing for sausages defect inspection method according to claim 6, which is characterized in that in step S1, utilize three
Group industrial camera shoots casing respectively with 120 ° of angles simultaneously, acquires 360 ° of panoramic image datas of casing, and pass through data cable
It is transmitted in industrial personal computer.
8. Collagent casing for sausages defect inspection method according to claim 6, which is characterized in that in step S4, confidence level
Threshold value be 97%.
9. the Collagent casing for sausages defect detecting system based on depth convolutional neural networks characterized by comprising
Image data acquisition module shoots casing respectively using three groups of industrial cameras with 120 ° of angles simultaneously, acquires 360 ° of casing
Panoramic image data, and be transmitted in industrial personal computer by data cable；
Three groups of image datas of image data acquisition module are successively normalized operation and are converted by graph data conversion module
Tensor data；
Collagent casing for sausages defects detection module designs Collagent casing for sausages using the described in any item methods of claim 1-5
Tensor data are inputted trained Collagent casing for sausages defects detection model and detected by defects detection model, output category
As a result, defective locations frame coordinate and confidence level；
Result treatment module, classification results and defective locations frame coordinate by confidence level greater than 97% pass through non-maxima suppression
Method merges, and obtains final the reasoning results；
Casing defect image data inside frame coordinate are extracted, and carry out Threshold segmentation to it by evaluation module, and calculating is fallen vacant
Sunken pixel value and coordinate value；And according to the proportionate relationship of camera lens focal length and camera CMOS pixel, by casing defect part
Pixel value be converted into actual size, dialogue casing presentation quality is objectively evaluated.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910421030.2A CN110111332A (en) | 2019-05-20 | 2019-05-20 | Collagent casing for sausages defects detection model, detection method and system based on depth convolutional neural networks |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910421030.2A CN110111332A (en) | 2019-05-20 | 2019-05-20 | Collagent casing for sausages defects detection model, detection method and system based on depth convolutional neural networks |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110111332A true CN110111332A (en) | 2019-08-09 |
Family
ID=67491140
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910421030.2A Withdrawn CN110111332A (en) | 2019-05-20 | 2019-05-20 | Collagent casing for sausages defects detection model, detection method and system based on depth convolutional neural networks |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110111332A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111814850A (en) * | 2020-06-22 | 2020-10-23 | 浙江大华技术股份有限公司 | Defect detection model training method, defect detection method and related device |
| CN113836850A (en) * | 2021-11-26 | 2021-12-24 | 成都数之联科技有限公司 | Model obtaining method, system and device, medium and product defect detection method |
| CN114862812A (en) * | 2022-05-16 | 2022-08-05 | 华中科技大学 | Two-stage rail transit vehicle defect detection method and system based on priori knowledge |
-
2019
- 2019-05-20 CN CN201910421030.2A patent/CN110111332A/en not_active Withdrawn
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111814850A (en) * | 2020-06-22 | 2020-10-23 | 浙江大华技术股份有限公司 | Defect detection model training method, defect detection method and related device |
| CN113836850A (en) * | 2021-11-26 | 2021-12-24 | 成都数之联科技有限公司 | Model obtaining method, system and device, medium and product defect detection method |
| CN114862812A (en) * | 2022-05-16 | 2022-08-05 | 华中科技大学 | Two-stage rail transit vehicle defect detection method and system based on priori knowledge |
| CN114862812B (en) * | 2022-05-16 | 2024-09-10 | 华中科技大学 | Priori knowledge-based two-stage rail transit vehicle defect detection method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111325713B (en) | Neural network-based wood defect detection method, system and storage medium | |
| CN107016413B (en) | A kind of online stage division of tobacco leaf based on deep learning algorithm | |
| CN112733950A (en) | Power equipment fault diagnosis method based on combination of image fusion and target detection | |
| CN107123111B (en) | Deep residual error network construction method for mobile phone screen defect detection | |
| CN108021947B (en) | A kind of layering extreme learning machine target identification method of view-based access control model | |
| CN106875373A (en) | Mobile phone screen MURA defect inspection methods based on convolutional neural networks pruning algorithms | |
| CN107944504B (en) | Board recognition and machine learning method and device for board recognition and electronic equipment | |
| Tilocca et al. | Detecting fabric defects with a neural network using two kinds of optical patterns | |
| CN110111332A (en) | Collagent casing for sausages defects detection model, detection method and system based on depth convolutional neural networks | |
| CN107220603A (en) | Vehicle checking method and device based on deep learning | |
| CN113160123A (en) | Leather defect detection method, system and device based on YOLOv5 | |
| CN110136101A (en) | A kind of tire X-ray defect detection method compared based on twin distance | |
| CN106355579A (en) | Defect detecting method of cigarette carton surface wrinkles | |
| CN110263768A (en) | A kind of face identification method based on depth residual error network | |
| JP2021515885A (en) | Methods, devices, systems and programs for setting lighting conditions and storage media | |
| CN114581456B (en) | Multi-image segmentation model construction method, image detection method and device | |
| CN114529484B (en) | Deep learning sample enhancement method for direct current component change in imaging | |
| CN114882216A (en) | Garment button quality detection method, system and medium based on deep learning | |
| CN112070712A (en) | Printing defect detection method based on self-encoder network | |
| CN111259923A (en) | Multi-target detection method based on improved three-dimensional R-CNN algorithm | |
| CN111882545B (en) | Fabric defect detection method based on bidirectional information transmission and feature fusion | |
| CN118135448A (en) | Target detection method and system used under dim light condition | |
| CN117910073A (en) | Artwork package design optimization system and method based on 3D printing technology | |
| CN116994206A (en) | Indoor personnel counting method and system based on multi-metadata fusion | |
| JP2023508641A (en) | Data augmentation-based matter analysis model learning device and method |
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 |
Application publication date: 20190809 |
|
| WW01 | Invention patent application withdrawn after publication |