CN110473173A - A kind of defect inspection method based on deep learning semantic segmentation - Google Patents
A kind of defect inspection method based on deep learning semantic segmentation Download PDFInfo
- Publication number
- CN110473173A CN110473173A CN201910673139.5A CN201910673139A CN110473173A CN 110473173 A CN110473173 A CN 110473173A CN 201910673139 A CN201910673139 A CN 201910673139A CN 110473173 A CN110473173 A CN 110473173A
- Authority
- CN
- China
- Prior art keywords
- defect
- image
- semantic segmentation
- defects detection
- training
- 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 203
- 230000011218 segmentation Effects 0.000 title claims abstract description 60
- 238000013135 deep learning Methods 0.000 title claims abstract description 36
- 238000000034 method Methods 0.000 title claims abstract description 34
- 238000007689 inspection Methods 0.000 title claims abstract description 18
- 238000001514 detection method Methods 0.000 claims abstract description 86
- 238000012549 training Methods 0.000 claims abstract description 33
- 230000000694 effects Effects 0.000 claims abstract description 12
- 238000004519 manufacturing process Methods 0.000 claims abstract description 9
- 230000009466 transformation Effects 0.000 claims description 12
- 238000005070 sampling Methods 0.000 claims description 8
- 230000002950 deficient Effects 0.000 claims description 6
- 238000012545 processing Methods 0.000 claims description 4
- 238000012360 testing method Methods 0.000 claims description 3
- 238000013519 translation Methods 0.000 claims description 3
- 208000037656 Respiratory Sounds Diseases 0.000 claims description 2
- 238000003709 image segmentation Methods 0.000 claims description 2
- 238000007711 solidification Methods 0.000 claims description 2
- 230000008023 solidification Effects 0.000 claims description 2
- 238000012423 maintenance Methods 0.000 claims 1
- 238000011002 quantification Methods 0.000 abstract description 4
- 239000000126 substance Substances 0.000 abstract description 3
- 238000005516 engineering process Methods 0.000 description 4
- 230000006835 compression Effects 0.000 description 3
- 238000007906 compression Methods 0.000 description 3
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 239000004744 fabric Substances 0.000 description 1
- 238000003475 lamination Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 230000000630 rising effect Effects 0.000 description 1
- 238000009966 trimming Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
-
- 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]
Abstract
The present invention relates to a kind of defect inspection methods based on semantic segmentation, comprising: production object table planar defect training image collection;Complete the manual mark of surface defect image;Build the deep learning semantic segmentation defects detection model applied to body surface defects detection;Carry out the training of deep learning semantic segmentation defects detection model;Based on the deep learning semantic segmentation defects detection model trained, body surface defects detection is carried out;Obtain corresponding output has defect kind label image, obtains defects detection result.It can automate, is efficient to body surface progress defects detection, to realize the automatic identification of body surface defect and the quantification treatment of defects detection result.After detectable substance surface imperfection or scene complexity rise, it can be good at taking into account detection effect and detection efficiency.
Description
Technical field
The present invention relates to a kind of body surface defect automatic testing method and systems, and in particular to one kind is based on deep learning
The body surface defect inspection method of semantic segmentation.
Background technique
With the development of modern industry, defect detecting technique is widely used in Fabric Defect detection, workpiece surface quality
Detection, aerospace field etc..Relative to the body surface defects detection under simple scenario, modern industry is to defect detecting technique
Propose more strict requirements.With the appearance and development of the science and technology such as computer technology, artificial intelligence, it is based on machine
The body surface defect detecting technique of vision technique is applied and is given birth to, largely improve body surface defects detection effect with
And body surface defects detection rate is improved, while avoiding because the factors such as scene condition, subjective judgement influence defects detection
As a result accuracy.In technical field of machine vision, body surface defect detecting technique simulates people by computer vision
Class visual performance, the acquisition process for carrying out image from specific tested material object calculates, final to realize that industrial vision is actually detected, controls
System and corresponding scene application.
But the body surface defect detecting technique of current most of machine vision techniques both for rule of surface or
Person detects the simple situation of scene, after detectable substance surface imperfection or scene complexity rising, detection effect and detection effect
Rate just cannot be considered in terms of, and seriously affect the applicability of defects detection.
Summary of the invention
It is a kind of based on deep technical problem to be solved by the present invention lies in providing in view of the deficiency of the prior art
The defect inspection method of degree study semantic segmentation can automate, is efficient to body surface progress defects detection, thus real
The automatic identification of existing body surface defect and the quantification treatment of defects detection result.When detectable substance surface imperfection or scene
After complexity rises, it can be good at taking into account detection effect and detection efficiency.
The technical scheme adopted by the invention is as follows:
A kind of defect inspection method based on semantic segmentation, it is characterised in that specifically includes the following steps:
S1, production object table planar defect training image collection, to the body surface under different spaces pose under Same Scene
Carry out surface defect image sample collection;
S2, production object table planar defect image set, the manual mark of surface defect image is completed using image enhancement technique;
S3, surface defect training image collection TrainSet is utilizedDefectFor input, surface defect marks training image collection
TrainLabelSetDefectFor output, the deep learning semantic segmentation defects detection applied to body surface defects detection is built
Model;
S4, the training for carrying out deep learning semantic segmentation defects detection model;
S5, based on the deep learning semantic segmentation defects detection model trained, carry out body surface defects detection;
Obtain corresponding output has defect kind label image, obtains defects detection result.
Further, in step S1, capturing sample image quantity is N width, N >=300.
Further, in step S2, in the production of body surface defect map image set, it is desirable that surface defect area interception image
1/3, be not otherwise considered as defect image;In a width defect image, defective data will be far fewer than background data;By object
After surface defect image enhances Processing Algorithm, the ratio of defective data and background data is close to 1:1 in image.
Further, image enhancement technique includes Image Reversal transformation, random trimming, color jitter, translation in step S2
At least one of transformation, change of scale, contrast variation, noise disturbance and rotation transformation/reflection transformation.
Further, when step S2 marks surface defect image manually, recording surface defect training image collection is
TrainSetDefect, picture format is [m, n, 3], and recording surface defect mark training image collection corresponds to
TrainLabelSetDefect, picture format is [m, n], and m is that picture size corresponds to length, and n is that picture size corresponds to width.
Further, deep learning semantic segmentation defects detection model described in step S3 is set as to picture each in image
Element carries out classification understanding, carries out image segmentation for body surface defect area respective pixel, carries out according to different defect kinds
Classification, gives identical defect kind label for same kind of different surfaces defect and exports.
Further, deep learning semantic segmentation defects detection model includes four chief components in step S3: volume
Lamination, pond layer, down-sampled layer and up-sampling layer;The dimensionality reduction sampling of feature vector is completed in down-sampled layer;Input data format
For the defect image data of [m, n, 3], it includes not in label image that the label image of output, which is the label image of [m, n] format,
Generic defect kind and background.
The image input format of deep learning semantic segmentation defects detection model setting is [m, n, 3], by model convolution
Compression forms compressive features vector [w, h, d] after layer, pond layer, and wherein w is the span of feature vector, and h is the height of feature vector
Degree, d are the depth of feature vector;Relative to original objects surface defect image input data, the compressive features vector is corresponding empty
Between dimension it is lower but more compact.Compressive features vector [w, h, d] is input to model up-sampling layer by down-sampled layer, rebuilds
Output identical with input layer (including convolutional layer, pond layer) dimension of deep learning semantic segmentation defects detection prototype network
Vector [m, n] exports the label image [m, n] with defect classification corresponding with input picture.
The present invention is based on traditional image classification depth convolutional neural networks, to each of defect image pixel
Classification is made, i.e., image is understood in pixel scale.Deep learning semantic segmentation defects detection model will not be to target
Example distinguishes, i.e., if in image including same category of two targets, corresponds to class label having the same, i.e.,
Identical defect kind label can be given by belonging to same kind of different surfaces defect.
Further, in step S4 carry out deep learning semantic segmentation defects detection model training in the following manner into
Row:
For defect map image set ImageSetDefectImage collection LabelImageSet is marked with defectDefect, by image
Middle background information is set as 0 class detection, and the defect information including crackle class is set as 1 class detection, and respective weights ratio is pressed
Respective pixel statistics ratio is configured, i.e., 0 class and 1 class detection corresponding pixel points quantity ratio are 1:k, k in correspondence image
For constant, then corresponding training weight ratio is k:1;Then the maximum training the number of iterations of setting is N=30000, and it is most that N corresponds to numerical value
May be big, if loss value has restrained during hands-on, can deconditioning, and then detect test effect.
Further, defect map image set ImageSet is directed in step S4Defect, using manual marking software
Colabeler pixel segmentation software v2.0.4 marks off the defects of every width defect image region in defect image training set, and
Corresponding defect kind is marked, defect is formed and marks image collection LabelImageSetDefect。
Further, defect map image set ImageSet is completed in step S4DefectMark work after, by corresponding input figure
Picture and label image are converted to TFRecord format, while the depth that model training parameter completes corresponding defect map image set is arranged
Semantic segmentation defects detection model training is practised, defect is formed and marks image collection LabelImageSetDefect。
Defect map image set ImageSetDefectImage input format be that [m, n, 3] and defect mark image collection
LabelImageSetDefectImage input format be [m, n].
Further, step S5 carries out body surface based on the deep learning semantic segmentation defects detection model trained
Defects detection;The deep learning semantic segmentation defects detection model trained in step S4 is solidified, i.e. generation pb model
File;For deep learning semantic segmentation defects detection model after solidification, defect image data are inputted, obtain the tool of corresponding output
Defective type label image obtains defects detection result.
Compared with the existing technology, obtained by the present invention to have the beneficial effect that
The detection effect and detection rates of body surface defect are not ensured that using traditional image processing algorithm, this
Invention builds deep learning semantic segmentation defects detection mould by the above-mentioned defect inspection method based on deep learning semantic segmentation
Type carries out classification understanding to pixel each in image, to complete the segmentation work of body surface defect area respective pixel, base
In cured body surface defects detection model, inputting defect image can be obtained defect kind result.Higher speed can be achieved
The body surface defect kind of rate divides and quantification treatment.
Detailed description of the invention
Fig. 1 is the principle steps figure of the method for the present invention.
Fig. 2 is process step figure of the invention (being configured with implementation result picture).
Specific embodiment
The invention will be further described by 1-2 with reference to the accompanying drawing.
As shown in Figure 1, the implementation case provides a kind of body surface defects detection based on deep learning semantic segmentation
Method, implementation steps include:
S1: production object table planar defect training image collection first, to the object under different spaces pose under Same Scene
Surface carries out defect image acquisition, and capturing sample image quantity is N width (N >=300).
S2: completing the manual mark of surface defect image using image enhancement technique, and record defect map image set is
ImageSetDefect(picture format is [m, n, 3]), record defect mark image collection correspond to LabelImageSetDefect
(picture format is [m, n]).
In the production of body surface defect map image set, it is desirable that otherwise the 1/3 of surface defect area interception image is not considered as
Defect image.In a width defect image, defective data will be far fewer than background data, to improve body surface defects detection effect
Fruit, introducing image data in use the technical solution of the present invention enhances technology comprising Image Reversal transformation is trimmed, color at random
Color shake, translation transformation, change of scale, contrast variation, noise disturbance and rotation transformation/reflection transformation.By object table
After planar defect image enhancement processing algorithm, the ratio of defect and background data is close to 1:1 in image.
S3, deep learning semantic segmentation network model is built;The present invention is with traditional image classification depth convolutional Neural net
Based on network, classification is made to each of defect image pixel, i.e., image is understood in pixel scale.Depth
Object instance will not be distinguished by practising semantic segmentation defects detection model, i.e., if in image including same category of two mesh
Mark, then it corresponds to class label having the same, therefore the present invention builds deep learning semantic segmentation defects detection model, and will
It is applied to body surface defects detection, i.e., same kind of body surface defect can class label having the same.
Projected depth study semantic segmentation defects detection prototype network of the present invention includes three chief components: convolution
Layer, down-sampled layer and up-sampling layer.Body surface defect image input format is [m, n, 3], by model convolutional layer, pond layer
Compression of images is expressed as the feature vector of [w, h, d] afterwards, and wherein w is the span of feature vector, and h is the height of feature vector, d
For the depth of feature vector, relative to original image input data, it is lower but more tight which corresponds to Spatial Dimension
It causes.This compression vector is input to model up-sampling layer by the present invention, rebuilds output vector identical with mode input layer dimension, i.e.,
The corresponding defect class label image of input picture.
S4: setting model training parameter (including: defect classification respective weights ratio, maximum training the number of iterations etc.), it is complete
At the training of deep learning semantic segmentation defects detection model.
For object defect map image set ImageSetDefect, marked off using manual marking software Colabeler v2.0.4
The defects of each image region, and corresponding defect kind is marked, 0 class and 1 class defects detection item complete defect image training
Corresponding input picture and label image are converted into TFRecord format in the mark work of collection, while model training parameter is set
Complete the deep learning semantic segmentation defects detection model training of corresponding diagram image set.
S5, based on the body surface defects detection for having trained deep learning semantic segmentation defects detection model: by step S4
In the deep learning semantic segmentation defects detection model trained solidified, be converted to Tensorflow pb file, corresponding mould
Type input is the defect image of [m, n, 3] format, and model output is the defect class label image of [m, n] format, label image
In include different classes of defect kind and background.
S51, deep learning semantic segmentation defects detection model inspection result is visualized, as shown in Fig. 2, defect is examined
Surveying black region part in effect image is 0 class defects detection item, and red area is 1 class detection, is based on to realize
The body surface defects detection effect of deep learning semantic segmentation.
S52, by the quantification treatment of the achievable body surface defect of image region segmentation algorithm, i.e., pass through tradition first
ZhangShi camera calibration algorithm solves pixel coordinate to the conversion map matrix of world coordinates, and then counts red in solution S51
The corresponding pixel faces product value in color region can finally obtain the actual value of corresponding 1 class defects detection item, realize and really scheme in Fig. 2
Body surface defects detection effect in piece.
Claims (10)
1. a kind of defect inspection method based on semantic segmentation, it is characterised in that specifically includes the following steps:
S1, production object table planar defect training image collection, carry out the body surface under different spaces pose under Same Scene
Surface defect image sample collection;
S2, production object table planar defect image set, the manual mark of surface defect image is completed using image enhancement technique;
S3, surface defect training image collection TrainSet is utilizedDefectFor input, surface defect marks training image collection
TrainLabelSetDefectFor output, the deep learning semantic segmentation defects detection applied to body surface defects detection is built
Model;
S4, the training for carrying out deep learning semantic segmentation defects detection model;
S5, based on the deep learning semantic segmentation defects detection model trained, carry out body surface defects detection;It is corresponded to
Output has defect kind label image, obtains defects detection result.
2. the defect inspection method according to claim 1 based on semantic segmentation, it is characterised in that in step S1, acquire sample
This amount of images is N width, N >=300.
3. the defect inspection method according to claim 1 based on semantic segmentation, it is characterised in that in step S2, in object
In the production of surface defect image collection, it is desirable that otherwise the 1/3 of surface defect area interception image is not considered as defect image;One
In width defect image, defective data will be far fewer than background data;After body surface defect image enhances Processing Algorithm, image
The ratio of middle defective data and background data is close to 1:1;Image enhancement technique includes Image Reversal transformation, with machine maintenance in step S2
Cut, in color jitter, translation transformation, change of scale, contrast variation, noise disturbance and rotation transformation/reflection transformation extremely
Few one kind.
4. the defect inspection method according to claim 1 based on semantic segmentation, it is characterised in that step S2 is marked manually
When surface defect image, recording surface defect training image collection is TrainSetDefect, picture format is [m, n, 3], record sheet
Planar defect mark training image collection corresponds to TrainLabelSetDefect, picture format is [m, n], and m is the corresponding length of picture size
Degree, n are that picture size corresponds to width.
5. the defect inspection method according to claim 1 based on semantic segmentation, it is characterised in that deep described in step S3
Degree study semantic segmentation defects detection model is set as carrying out classification understanding to pixel each in image, for body surface defect
Region respective pixel carries out image segmentation, classifies according to different defect kinds, for same kind of different surfaces defect
It gives identical defect kind label and is exported;If in image including same category of two targets, tool is corresponded to
There is identical class label, that is, identical defect kind label can be given by belonging to same kind of different surfaces defect.
6. the defect inspection method according to claim 1 based on semantic segmentation, it is characterised in that depth in step S3
Practising semantic segmentation defects detection model includes four chief components: convolutional layer, pond layer, down-sampled layer and up-sampling layer;
The dimensionality reduction sampling of feature vector is completed in down-sampled layer;Input data format is the defect image data of [m, n, 3], the mark of output
The label image that image is [m, n] format is signed, includes different classes of defect kind and background in label image.
7. the defect inspection method according to claim 1 based on semantic segmentation, it is characterised in that depth in step S3
The image input format for practising the setting of semantic segmentation defects detection model is [m, n, 3], is compressed after model convolutional layer, pond layer
It is formed compressive features vector [w, h, d], wherein w is the span of feature vector, and h is the height of feature vector, and d is feature vector
Depth;Contracting feature vector [w, h, d] is input to model up-sampling layer by down-sampled layer, rebuilds and lacks with deep learning semantic segmentation
The identical output vector of input layer dimension [m, n] of detection model network is fallen into, that is, is exported corresponding with input picture with defect
The label image [m, n] of classification.
8. the defect inspection method according to claim 1 based on semantic segmentation, it is characterised in that carried out in step S4 deep
The training of degree study semantic segmentation defects detection model carries out in the following manner:
For defect map image set ImageSetDefectImage collection LabelImageSet is marked with defectDefect, will be carried on the back in image
Scape information is set as 0 class detection, and the defect information including crackle class is set as 1 class detection, and respective weights ratio is by correspondence
Pixels statistics ratio is configured, i.e., 0 class and 1 class detection corresponding pixel points quantity ratio are 1:k in correspondence image, and k is normal
Number, then corresponding training weight ratio is k:1;Then the maximum training the number of iterations of setting is N=30000, and N corresponds to numerical value as far as possible
Greatly, if loss value has restrained during hands-on, can deconditioning, and then detect test effect.
9. the defect inspection method according to claim 1 based on semantic segmentation, it is characterised in that for scarce in step S4
Fall into image set ImageSetDefect, defect map is marked off using manual marking software Colabeler pixel segmentation software v2.0.4
As the defects of every width defect image region in training set, and corresponding defect kind is marked, forms defect and mark image collection
LabelImageSetDefect;Complete defect map image set ImageSetDefectMark work after, will corresponding input picture and label
Image is converted to TFRecord format, while the deep learning semanteme point that model training parameter completes corresponding defect map image set is arranged
Defects detection model training is cut, defect is formed and marks image collection LabelImageSetDefect。
10. the defect inspection method according to claim 1 based on semantic segmentation, it is characterised in that step S5 is based on having instructed
Experienced deep learning semantic segmentation defects detection model carries out body surface defects detection;The depth that will have been trained in step S4
Study semantic segmentation defects detection model is solidified, i.e. generation pb model file;For deep learning semantic segmentation after solidification
Defects detection model inputs defect image data, and obtain corresponding output has defect kind label image, obtains defects detection
As a result.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910673139.5A CN110473173A (en) | 2019-07-24 | 2019-07-24 | A kind of defect inspection method based on deep learning semantic segmentation |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910673139.5A CN110473173A (en) | 2019-07-24 | 2019-07-24 | A kind of defect inspection method based on deep learning semantic segmentation |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110473173A true CN110473173A (en) | 2019-11-19 |
Family
ID=68508889
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910673139.5A Pending CN110473173A (en) | 2019-07-24 | 2019-07-24 | A kind of defect inspection method based on deep learning semantic segmentation |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110473173A (en) |
Cited By (32)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111080636A (en) * | 2019-12-23 | 2020-04-28 | 杨静日 | CNN semantic segmentation self-learning detection method for surface defects of color steel tiles |
CN111179278A (en) * | 2019-12-16 | 2020-05-19 | 精英数智科技股份有限公司 | Image detection method, device, equipment and storage medium |
CN111311544A (en) * | 2020-01-19 | 2020-06-19 | 无锡赛默斐视科技有限公司 | Floor defect detection method based on deep learning |
CN111681232A (en) * | 2020-06-10 | 2020-09-18 | 厦门理工学院 | Industrial welding image defect detection method based on semantic segmentation |
CN111709918A (en) * | 2020-06-01 | 2020-09-25 | 深圳市深视创新科技有限公司 | Product defect classification method combining multiple channels based on deep learning |
CN111862092A (en) * | 2020-08-05 | 2020-10-30 | 复旦大学 | Express delivery outer package defect detection method and device based on deep learning |
CN111932501A (en) * | 2020-07-13 | 2020-11-13 | 太仓中科信息技术研究院 | Seal ring surface defect detection method based on semantic segmentation |
CN112085722A (en) * | 2020-09-07 | 2020-12-15 | 凌云光技术股份有限公司 | Training sample image acquisition method and device |
CN112329896A (en) * | 2021-01-05 | 2021-02-05 | 武汉精测电子集团股份有限公司 | Model training method and device |
CN112365443A (en) * | 2020-10-16 | 2021-02-12 | 珠海市奥德维科技有限公司 | Hexahedron defect detection method and medium based on deep learning |
CN112488119A (en) * | 2020-12-18 | 2021-03-12 | 山西省信息产业技术研究院有限公司 | Tunnel block falling or water seepage detection and measurement method based on double-depth learning model |
CN113344888A (en) * | 2021-06-17 | 2021-09-03 | 四川启睿克科技有限公司 | Surface defect detection method and device based on combined model |
US11403069B2 (en) | 2017-07-24 | 2022-08-02 | Tesla, Inc. | Accelerated mathematical engine |
US11409692B2 (en) | 2017-07-24 | 2022-08-09 | Tesla, Inc. | Vector computational unit |
US11487288B2 (en) | 2017-03-23 | 2022-11-01 | Tesla, Inc. | Data synthesis for autonomous control systems |
US11537811B2 (en) | 2018-12-04 | 2022-12-27 | Tesla, Inc. | Enhanced object detection for autonomous vehicles based on field view |
US11561791B2 (en) | 2018-02-01 | 2023-01-24 | Tesla, Inc. | Vector computational unit receiving data elements in parallel from a last row of a computational array |
US11562231B2 (en) | 2018-09-03 | 2023-01-24 | Tesla, Inc. | Neural networks for embedded devices |
US11567514B2 (en) | 2019-02-11 | 2023-01-31 | Tesla, Inc. | Autonomous and user controlled vehicle summon to a target |
US11610117B2 (en) | 2018-12-27 | 2023-03-21 | Tesla, Inc. | System and method for adapting a neural network model on a hardware platform |
US11636333B2 (en) | 2018-07-26 | 2023-04-25 | Tesla, Inc. | Optimizing neural network structures for embedded systems |
US11665108B2 (en) | 2018-10-25 | 2023-05-30 | Tesla, Inc. | QoS manager for system on a chip communications |
US11681649B2 (en) | 2017-07-24 | 2023-06-20 | Tesla, Inc. | Computational array microprocessor system using non-consecutive data formatting |
US11734562B2 (en) | 2018-06-20 | 2023-08-22 | Tesla, Inc. | Data pipeline and deep learning system for autonomous driving |
US11748620B2 (en) | 2019-02-01 | 2023-09-05 | Tesla, Inc. | Generating ground truth for machine learning from time series elements |
US11790664B2 (en) | 2019-02-19 | 2023-10-17 | Tesla, Inc. | Estimating object properties using visual image data |
US11816585B2 (en) | 2018-12-03 | 2023-11-14 | Tesla, Inc. | Machine learning models operating at different frequencies for autonomous vehicles |
US11841434B2 (en) | 2018-07-20 | 2023-12-12 | Tesla, Inc. | Annotation cross-labeling for autonomous control systems |
CN117269168A (en) * | 2023-09-15 | 2023-12-22 | 昆山精诚得精密五金模具有限公司 | New energy automobile precision part surface defect detection device and detection method |
US11893393B2 (en) | 2017-07-24 | 2024-02-06 | Tesla, Inc. | Computational array microprocessor system with hardware arbiter managing memory requests |
US11893774B2 (en) | 2018-10-11 | 2024-02-06 | Tesla, Inc. | Systems and methods for training machine models with augmented data |
CN111311544B (en) * | 2020-01-19 | 2024-04-26 | 无锡赛默斐视科技有限公司 | Floor defect detection method based on deep learning |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20180268257A1 (en) * | 2017-03-20 | 2018-09-20 | Rolls-Royce Plc | Surface defect detection |
CN108717569A (en) * | 2018-05-16 | 2018-10-30 | 中国人民解放军陆军工程大学 | It is a kind of to expand full convolutional neural networks and its construction method |
CN109829900A (en) * | 2019-01-18 | 2019-05-31 | 创新奇智(北京)科技有限公司 | A kind of steel coil end-face defect inspection method based on deep learning |
-
2019
- 2019-07-24 CN CN201910673139.5A patent/CN110473173A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20180268257A1 (en) * | 2017-03-20 | 2018-09-20 | Rolls-Royce Plc | Surface defect detection |
CN108717569A (en) * | 2018-05-16 | 2018-10-30 | 中国人民解放军陆军工程大学 | It is a kind of to expand full convolutional neural networks and its construction method |
CN109829900A (en) * | 2019-01-18 | 2019-05-31 | 创新奇智(北京)科技有限公司 | A kind of steel coil end-face defect inspection method based on deep learning |
Cited By (39)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US11487288B2 (en) | 2017-03-23 | 2022-11-01 | Tesla, Inc. | Data synthesis for autonomous control systems |
US11681649B2 (en) | 2017-07-24 | 2023-06-20 | Tesla, Inc. | Computational array microprocessor system using non-consecutive data formatting |
US11893393B2 (en) | 2017-07-24 | 2024-02-06 | Tesla, Inc. | Computational array microprocessor system with hardware arbiter managing memory requests |
US11409692B2 (en) | 2017-07-24 | 2022-08-09 | Tesla, Inc. | Vector computational unit |
US11403069B2 (en) | 2017-07-24 | 2022-08-02 | Tesla, Inc. | Accelerated mathematical engine |
US11561791B2 (en) | 2018-02-01 | 2023-01-24 | Tesla, Inc. | Vector computational unit receiving data elements in parallel from a last row of a computational array |
US11797304B2 (en) | 2018-02-01 | 2023-10-24 | Tesla, Inc. | Instruction set architecture for a vector computational unit |
US11734562B2 (en) | 2018-06-20 | 2023-08-22 | Tesla, Inc. | Data pipeline and deep learning system for autonomous driving |
US11841434B2 (en) | 2018-07-20 | 2023-12-12 | Tesla, Inc. | Annotation cross-labeling for autonomous control systems |
US11636333B2 (en) | 2018-07-26 | 2023-04-25 | Tesla, Inc. | Optimizing neural network structures for embedded systems |
US11562231B2 (en) | 2018-09-03 | 2023-01-24 | Tesla, Inc. | Neural networks for embedded devices |
US11893774B2 (en) | 2018-10-11 | 2024-02-06 | Tesla, Inc. | Systems and methods for training machine models with augmented data |
US11665108B2 (en) | 2018-10-25 | 2023-05-30 | Tesla, Inc. | QoS manager for system on a chip communications |
US11816585B2 (en) | 2018-12-03 | 2023-11-14 | Tesla, Inc. | Machine learning models operating at different frequencies for autonomous vehicles |
US11537811B2 (en) | 2018-12-04 | 2022-12-27 | Tesla, Inc. | Enhanced object detection for autonomous vehicles based on field view |
US11908171B2 (en) | 2018-12-04 | 2024-02-20 | Tesla, Inc. | Enhanced object detection for autonomous vehicles based on field view |
US11610117B2 (en) | 2018-12-27 | 2023-03-21 | Tesla, Inc. | System and method for adapting a neural network model on a hardware platform |
US11748620B2 (en) | 2019-02-01 | 2023-09-05 | Tesla, Inc. | Generating ground truth for machine learning from time series elements |
US11567514B2 (en) | 2019-02-11 | 2023-01-31 | Tesla, Inc. | Autonomous and user controlled vehicle summon to a target |
US11790664B2 (en) | 2019-02-19 | 2023-10-17 | Tesla, Inc. | Estimating object properties using visual image data |
CN111179278A (en) * | 2019-12-16 | 2020-05-19 | 精英数智科技股份有限公司 | Image detection method, device, equipment and storage medium |
CN111080636A (en) * | 2019-12-23 | 2020-04-28 | 杨静日 | CNN semantic segmentation self-learning detection method for surface defects of color steel tiles |
CN111080636B (en) * | 2019-12-23 | 2023-08-25 | 山东安亚金属科技有限公司 | CNN semantic segmentation self-learning detection method for color steel tile surface defects |
CN111311544B (en) * | 2020-01-19 | 2024-04-26 | 无锡赛默斐视科技有限公司 | Floor defect detection method based on deep learning |
CN111311544A (en) * | 2020-01-19 | 2020-06-19 | 无锡赛默斐视科技有限公司 | Floor defect detection method based on deep learning |
CN111709918B (en) * | 2020-06-01 | 2023-04-18 | 深圳市深视创新科技有限公司 | Product defect classification method combining multiple channels based on deep learning |
CN111709918A (en) * | 2020-06-01 | 2020-09-25 | 深圳市深视创新科技有限公司 | Product defect classification method combining multiple channels based on deep learning |
CN111681232B (en) * | 2020-06-10 | 2023-03-10 | 厦门理工学院 | Industrial welding image defect detection method based on semantic segmentation |
CN111681232A (en) * | 2020-06-10 | 2020-09-18 | 厦门理工学院 | Industrial welding image defect detection method based on semantic segmentation |
CN111932501A (en) * | 2020-07-13 | 2020-11-13 | 太仓中科信息技术研究院 | Seal ring surface defect detection method based on semantic segmentation |
CN111862092A (en) * | 2020-08-05 | 2020-10-30 | 复旦大学 | Express delivery outer package defect detection method and device based on deep learning |
CN112085722A (en) * | 2020-09-07 | 2020-12-15 | 凌云光技术股份有限公司 | Training sample image acquisition method and device |
CN112085722B (en) * | 2020-09-07 | 2024-04-09 | 凌云光技术股份有限公司 | Training sample image acquisition method and device |
CN112365443A (en) * | 2020-10-16 | 2021-02-12 | 珠海市奥德维科技有限公司 | Hexahedron defect detection method and medium based on deep learning |
CN112488119A (en) * | 2020-12-18 | 2021-03-12 | 山西省信息产业技术研究院有限公司 | Tunnel block falling or water seepage detection and measurement method based on double-depth learning model |
CN112329896A (en) * | 2021-01-05 | 2021-02-05 | 武汉精测电子集团股份有限公司 | Model training method and device |
CN113344888A (en) * | 2021-06-17 | 2021-09-03 | 四川启睿克科技有限公司 | Surface defect detection method and device based on combined model |
CN117269168A (en) * | 2023-09-15 | 2023-12-22 | 昆山精诚得精密五金模具有限公司 | New energy automobile precision part surface defect detection device and detection method |
CN117269168B (en) * | 2023-09-15 | 2024-04-09 | 昆山精诚得精密五金模具有限公司 | New energy automobile precision part surface defect detection device and detection method |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110473173A (en) | A kind of defect inspection method based on deep learning semantic segmentation | |
CN111179229B (en) | Industrial CT defect detection method based on deep learning | |
CN107123111B (en) | Deep residual error network construction method for mobile phone screen defect detection | |
CN111402226A (en) | Surface defect detection method based on cascade convolution neural network | |
CN107665492B (en) | Colorectal panoramic digital pathological image tissue segmentation method based on depth network | |
CN105957086B (en) | A kind of method for detecting change of remote sensing image based on optimization neural network model | |
CN111080622B (en) | Neural network training method, workpiece surface defect classification and detection method and device | |
CN110070008B (en) | Bridge disease identification method adopting unmanned aerial vehicle image | |
CN109977808A (en) | A kind of wafer surface defects mode detection and analysis method | |
CN107607554A (en) | A kind of Defect Detection and sorting technique of the zinc-plated stamping parts based on full convolutional neural networks | |
CN111507976B (en) | Defect detection method and system based on multi-angle imaging | |
CN111062915A (en) | Real-time steel pipe defect detection method based on improved YOLOv3 model | |
CN109284779A (en) | Object detecting method based on the full convolutional network of depth | |
CN110992349A (en) | Underground pipeline abnormity automatic positioning and identification method based on deep learning | |
CN113298757A (en) | Metal surface defect detection method based on U-NET convolutional neural network | |
CN111127417B (en) | Printing defect detection method based on SIFT feature matching and SSD algorithm improvement | |
CN109584206B (en) | Method for synthesizing training sample of neural network in part surface flaw detection | |
CN115601332A (en) | Embedded fingerprint module appearance detection method based on semantic segmentation | |
CN111161237A (en) | Fruit and vegetable surface quality detection method, storage medium and sorting device thereof | |
CN115760837A (en) | Crystal quality evaluation method and system based on deep neural network | |
CN113393438A (en) | Resin lens defect detection method based on convolutional neural network | |
CN114972230A (en) | Method and system for detecting defects of circuit board on production line | |
CN114428110A (en) | Method and system for detecting defects of fluorescent magnetic powder inspection image of bearing ring | |
CN113112482A (en) | PCB defect detection method based on attention mechanism network | |
CN117237736A (en) | Daqu quality detection method based on machine vision and deep learning |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20191119 |