CN110363781A - Molten bath profile testing method based on deep neural network - Google Patents
Molten bath profile testing method based on deep neural network Download PDFInfo
- Publication number
- CN110363781A CN110363781A CN201910581281.7A CN201910581281A CN110363781A CN 110363781 A CN110363781 A CN 110363781A CN 201910581281 A CN201910581281 A CN 201910581281A CN 110363781 A CN110363781 A CN 110363781A
- Authority
- CN
- China
- Prior art keywords
- image
- molten bath
- network
- original
- bath
- 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.)
- Granted
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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/12—Edge-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/13—Edge 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
- G06T7/155—Segmentation; Edge detection involving morphological operators
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30108—Industrial image inspection
- G06T2207/30164—Workpiece; Machine component
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of molten bath profile testing method based on deep neural network, crater image when acquisition welding, cuts out de-redundancy background, obtains original bath image set；Dividing mark sample is made, constitutes molten bath picture indicia data set with original bath image set；It is trained using confrontation network DCGAN is generated, generates image similar with original bath image, constitute the molten bath picture indicia data set expanded with molten bath picture indicia data set；Color science and morphologic data augmentation are carried out, is put into semantic segmentation network and is trained, extracts molten bath profile.The present invention makes network model have better generalization ability, and preferably improve the segmentation precision to the weak fringe region in molten bath while saving manpower and time.
Description
Technical field
A kind of field of machine vision of the present invention, and in particular to molten bath profile testing method based on deep neural network.
Background technique
Automatic welding has greatly been liberated manually, is widely used in terms of intelligence production.It is regarded using computer
Feel and extract, analyze molten bath profile, is conducive to real-time detection and control automatic Arc Welding, guarantees the quality of welding product.Tradition
Method carries out edge detection using Canny operator, CV active contour etc., however, crater image is by Welder in reality
The influence of skill and material, the case where being easy to appear intensity profile unevenness and interfered by arc light, conventional method is difficult to extract at this time
Accurate and complete closure molten bath profile.
Summary of the invention
The purpose of the present invention is to provide a kind of molten bath profile testing method based on deep neural network.
The technical solution for realizing the aim of the invention is as follows: a kind of molten bath contour detecting side based on deep neural network
Method includes the following steps:
Step 1 establishes molten bath visual sensing system, and crater image when acquisition is welded is cut out de-redundancy background, obtained original
Crater image collection；
Step 2, the crater image obtained based on step 1 make its corresponding dividing mark sample, with original bath image
Collection constitutes molten bath picture indicia data set；
Step 3, the molten bath picture indicia data set obtained based on step 2 are instructed using confrontation network DCGAN is generated
Practice, generate image similar with original bath image, constitutes the molten bath picture indicia number expanded with molten bath picture indicia data set
According to collection；
Step 4 carries out color science and morphologic data augmentation to the crater image flag data collection of expansion, is put into language
It is trained in justice segmentation network；
Step 5 extracts molten bath profile using the semantic network network model that training obtains described in step 4.
Compared with prior art, the present invention its remarkable advantage are as follows: 1) using generate confrontation network generate similar image with
The method of machine number control data augmentation intensity is being saved in the case where the data set scale of construction does not reach segmentation task precision enough
While manpower and time, network model is made to have better generalization ability；2) by the feature of the different scale of residual error network
It is merged, preferably improves the segmentation precision to the weak fringe region in molten bath.
Detailed description of the invention
Fig. 1 is the schematic diagram for the molten bath visual sensing system that the present invention establishes.
Fig. 2 is the crater image after window of the present invention is cut.
Fig. 3 is crater image of the present invention and corresponding marker samples schematic diagram.
Fig. 4 is the generation image in DCGAN training process of the present invention, wherein (a) is the generation image of epoch=10, (b)
For the generation image of epoch=120, (c) the generation image of epoch=300.
Fig. 5 is the generation image after present invention segmentation screening.
Fig. 6 is that the present invention generates image and the corresponding marker samples schematic diagram of original image.
Fig. 7 is the structural schematic diagram of Res-Seg network of the present invention.
Fig. 8 is the flow chart of data augmentation of the present invention.
Fig. 9 is the method for the present invention flow chart
The schematic diagram of Figure 10 test result of the present invention, wherein (a) is the test result of Canny, it (b) is the test knot of CV
Fruit, (c) test result for being ENet, (d) test result for being ResNet50 are (e) test result of inventive algorithm.
Specific embodiment
In the following with reference to the drawings and specific embodiments, the present invention program is further illustrated.
A kind of molten bath profile testing method based on deep neural network, includes the following steps:
Step 1 establishes molten bath visual sensing system, and crater image when acquisition is welded is cut out de-redundancy background, obtained original
Crater image collection；
The molten bath visual sensing system of foundation is as shown in Figure 1, since welding gun under TIG weld technique and workpiece contact position exist
Biggish arc light arranges the optical filtering of 660nm for the adverse effect for inhibiting arc light to generate contour detecting before camera lens
Piece, to obtain clearly high quality graphic.Due to the image information for the only molten bath peripheral region that the present invention is concerned about, so background
Information is redundancy, and the fewer background information ingredient the better, is thus emphasis to collected molten bath color image using molten bath zone
(original image) carries out window and cuts.After the crater image for acquiring 1920*1200,400*400 size windows are carried out according to molten bath zone
It cuts, the image after cutting is as shown in Figure 2.
Step 2, the crater image obtained based on step 1 make its corresponding dividing mark sample, with original bath image
Collection constitutes molten bath picture indicia data set；
When making its corresponding dividing mark sample, the pixel filling for being 255 by the molten bath zone gray scale for needing to divide,
The pixel filling that background area gray scale is 0, as shown in Figure 3.
Step 3, the molten bath picture indicia data set obtained based on step 2 are instructed using confrontation network DCGAN is generated
Practice, generate image similar with original bath image, constitutes the molten bath picture indicia number expanded with molten bath picture indicia data set
According to collection；
The image of original bath image set is spliced into one according to the batch sizes (such as Batch Size=4) of setting first
Image is opened, is sequentially sent to generate in confrontation network DCGAN.Every training Jing Guo 100 Batch Size sizes, to original data set
Middle picture is tested, and test results are shown in figure 4, it can be seen that in training process, with the increasing of exercise wheel number (epoch)
Add, generation picture is gradually clear and becomes closer in original bath image.Then after training, the network of generation is utilized
Model tests the image of original bath image set, generates spelling with Batch Size size for setting quantity (such as 100)
Generation figure made of connecing.Then generation figure is cut into individual molten bath picture, quality is retained after screening and is preferably generated
Picture, as shown in Figure 5.Finally, as shown in fig. 6, searching for corresponding original image and dividing mark based on newly-generated crater image
Sample (label) collectively forms the crater image flag data collection of expansion.The mode of this data supplement is with a kind of saving manpower
The mode of (not needing to make extra label again) increases the sample size of data set with the time.
Step 4 carries out color science and morphologic data augmentation to the crater image flag data collection of expansion, is put into language
It is trained in justice segmentation network；
The semantic segmentation network Res-Seg of design as shown in fig. 7, the network be based on residual error network ResNet-50 develop and
Come, carries out progressive convolution operation to input first with ResNet-50 network, obtain the characteristic pattern of different scale, if
Segmentation result identical with dimension of picture size is inputted is obtained, the characteristic pattern for obtaining convolution is needed to be upsampled to original image size
Size calculates loss function with marker samples；
As shown in Figure 7, the feature f based on 1/32 size of original image1/32After carrying out up-sampling operation, the picture of reduction is only
Only the feature in the convolution kernel in the last one scale convolutional layer, only using this layer of feature go determine divide the result is that than
More unilateral.Therefore, the present invention is using iteration forward, first by f1/32The feature of 1/16 scale size of original image is reduced to up-sampling
f1/16, by itself and the f that is exported in convolution process1/16Carry out addition fusion.Feature after fusion is similarly operated and convolution
The feature f of 1/8 original image scale in the process1/8Addition fusion is carried out, obtained feature is to being upsampled to original image size, the above operation
It as shown in formula (1), does so, low layer and high-rise feature is sufficiently merged into output, effectively improve the essence of Target Segmentation；
Wherein D represents up-sampling operation.
In addition, the design of network losses function is as shown in formula (2):
Wherein E is softmax function, and i and j determine that pixel is to be located at foreground area f to be splitgIn or background area
bgIn, yijIndicate the binary system predicted value of pixel, a is the pixel number ratio that background accounts for image, and b is the pixel number that prospect accounts for image
Than.
Before data set to be put into Res-Seg network and is trained, crater image and its correspondence markings sample are carried out
Data augmentation, step the metamorphosis and color adjustment etc. such as rotate, cut as shown in figure 8, this part introduces molten bath picture
Operation.The ability that the requirement of model generalization ability makes these operations need to have adjustable intensity is (angle of picture rotation, bright
Spend the size etc. of adjustment), the concept of random number control is incorporated herein, before each picture is passed to and is trained in network, root
It is adjusted according to the random number of generation in the intensity of previously mentioned operation, then incoming network.The network for enabling to training to obtain in this way
Model has more robustness.Steps are as follows for augmentation:
A, maximum rotation angle, maximum zoom ratio value (scale) and maximum are set and cut long width values；
B, generate the random floating point M between 0 to 1, on the basis of M with maximum rotation angle, maximum scale value and most
Length and width value is cut greatly and carries out multiplication operations, generates the random floating point of control metamorphosis, control rotation, scaling, cutting operation；
Multiplication operations are carried out with scale on the basis of 2*M-1, generate the random floating point of control color change, control brightness, saturation
Degree, contrast, acutance, Gaussian Blur operation；
C, it is loaded into the crater image and its corresponding label of the crater image flag data collection expanded, executes rotation and contracting
Operation is put, the positive and negative of the random floating point of color change is controlled according to step b and judges whether to cut and color conversion is grasped
Make, if being negative, without cutting and color conversion operation, otherwise image and marker samples is cut and color turns
It changes.
Step 5 extracts molten bath profile using the semantic network network model that training obtains described in step 4.
Embodiment
In order to verify the validity of the present invention program, the molten bath figure in test set is tested using the network model that training obtains
Picture, it is as shown in Figure 9 with the comparing result of traditional algorithm on the original color image that is added to after segmentation result contouring.It can be seen that this
It is more smoother than the contour line that traditional contour line extraction method or traditional semantic segmentation network are extracted to invent the contour line extracted, more
Close to true molten bath boundary.
The segmentation precision of target and background, P are calculated according to formula (3)iiIndicate the pixel correctly classified, Pij(i≠j)
Indicate that, by the pixel of mistake classification, k indicates classification sum:
The segmentation precision of the method for the present invention and conventional method is as shown in table 1.By in table it is found that the method for the present invention segmentation essence
Degree increase compared with conventional method, it may be said that the bright present invention it is this based on generate confrontation network to data set carry out supplement and it is right
The mode that residual error Multi-Layer Networks feature is merged is conducive to improve segmentation task precision.
The segmentation precision contrast table of 1 target and background of table
In order to verify the robustness of network model, test, test result such as table 2 have also been made in the data other than test set
It is shown.Accuracy rate of the present invention can achieve 92%, and about 2 percentage points of standard is improved on former ResNet-50 basic network
True rate, and 7 percentage points or so are improved on former ResNet-101 basic network, one of major reason is: ResNet-
The 101 network numbers of plies are too deep, and structure is complicated, and parameter is excessive, and network is caused to occur over-fitting on the training data, thus training pattern
The effect that can not have been shown on the unknown data other than training data.
The contrast table of 2 contours extract precision of table
Claims (7)
1. a kind of molten bath profile testing method based on deep neural network, which comprises the steps of:
Step 1 establishes molten bath visual sensing system, and crater image when acquisition is welded cuts out de-redundancy background, obtains original bath
Image set；
Step 2, the crater image obtained based on step 1 make its corresponding dividing mark sample, with original bath image set structure
At molten bath picture indicia data set；
Step 3, the molten bath picture indicia data set obtained based on step 2 are trained using confrontation network DCGAN is generated, raw
At image similar with original bath image, the molten bath picture indicia data set expanded is constituted with molten bath picture indicia data set；
Step 4 carries out color science and morphologic data augmentation to the crater image flag data collection of expansion, is put into semantic point
It cuts in network and is trained；
Step 5 extracts molten bath profile using the semantic network network model that training obtains described in step 4.
2. the molten bath profile testing method according to claim 1 based on deep neural network, which is characterized in that step 1
In, it is contemplated that there are biggish arc lights for welding gun and workpiece contact position under TIG weld technique, in order to inhibit arc light to produce contour detecting
Raw adverse effect arranges the optical filter of 660nm before camera lens, to obtain clearly high quality graphic.
3. the molten bath profile testing method according to claim 1 based on deep neural network, which is characterized in that step 2
In, when making dividing mark sample, the pixel filling for being 255 by the molten bath zone gray scale for needing to divide, background area ash
The pixel filling that degree is 0.
4. the molten bath profile testing method according to claim 1 based on deep neural network, which is characterized in that step 3
In, the image of original bath image set is spliced first, in accordance with the batch sizes of setting, and be sequentially sent to generate confrontation net
In network DCGAN；Then after training, the image of original bath image set is tested using the network model of generation,
Generate the generation figure being spliced according to batch sizes；Then generation figure is cut into individual molten bath picture, and screens and meets item
The molten bath picture of part；Corresponding original image and dividing mark sample are finally searched for according to newly-generated crater image, collectively formed
The crater image flag data collection of expansion.
5. the molten bath profile testing method according to claim 1 based on deep neural network, which is characterized in that step 4
In, before data set to be put into Res-Seg network and is trained, by metamorphosis and color adjustment to crater image and its
Correspondence markings sample carries out data augmentation, the specific steps are as follows:
A, maximum rotation angle, maximum zoom ratio value and maximum are set and cut long width values；
B, generate the random floating point M between 0 to 1, on the basis of M with maximum rotation angle, maximum zoom ratio value and most
Length and width value is cut greatly and carries out multiplication operations, generates the random floating point of control metamorphosis, control rotation, scaling, cutting operation；
Multiplication operations are carried out with maximum zoom ratio value on the basis of 2*M-1, generate the random floating point of control color change, are controlled bright
Degree, saturation degree, contrast, acutance, Gaussian Blur operation；
C, the crater image and its corresponding dividing mark sample of the crater image flag data collection expanded are loaded into, execute rotation and
Zoom operations control the positive and negative of the random floating point of color change according to step b and judge whether to cut and color conversion
Operation, without cutting and color conversion operation, otherwise cuts image and marker samples and color turns if being negative
It changes.
6. the molten bath profile testing method according to claim 1 based on deep neural network, which is characterized in that step 4
In, semantic segmentation network is based on residual error network ResNet-50 and is evolved, and carries out first with ResNet-50 network to input
Progressive convolution operation obtains the characteristic pattern of different scale, and it is identical then to obtain original image size according to formula (1) operation
Characteristic pattern, with marker samples calculate loss function；
Wherein D represents up-sampling operation, f1/32Indicate the feature based on 1/32 size of original image of convolution output, f1/16Indicate convolution
The feature based on 1/16 size of original image of output.
7. the molten bath profile testing method according to claim 1 based on deep neural network, which is characterized in that step 4
In, the loss function of semantic segmentation network is designed as shown in formula (2):
Wherein E is softmax function, and i and j determine that pixel is to be located at foreground area f to be splitgIn or background area bgIn,
yijIndicate the binary system predicted value of pixel, a is the pixel number ratio that background accounts for image, and b is the pixel number ratio that prospect accounts for image.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910581281.7A CN110363781B (en) | 2019-06-29 | 2019-06-29 | Molten pool contour detection method based on deep neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910581281.7A CN110363781B (en) | 2019-06-29 | 2019-06-29 | Molten pool contour detection method based on deep neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110363781A true CN110363781A (en) | 2019-10-22 |
| CN110363781B CN110363781B (en) | 2022-08-12 |
Family
ID=68215976
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910581281.7A Active CN110363781B (en) | 2019-06-29 | 2019-06-29 | Molten pool contour detection method based on deep neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110363781B (en) |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111932539A (en) * | 2020-10-13 | 2020-11-13 | 南京知谱光电科技有限公司 | Molten pool image and depth residual error network-based height and penetration collaborative prediction method |
| WO2021068982A1 (en) * | 2020-03-19 | 2021-04-15 | 南通大学 | Quaternion multi-degree-of-freedom neuron-based multispectral welding image recognition method |
| CN112906833A (en) * | 2021-05-08 | 2021-06-04 | 武汉大学 | Plasma energy deposition image identification method based on full convolution neural network |
| CN113077423A (en) * | 2021-03-22 | 2021-07-06 | 中国人民解放军空军工程大学 | Laser selective melting pool image analysis system based on convolutional neural network |
| CN113118465A (en) * | 2019-12-31 | 2021-07-16 | 韩国科学技术院 | Method and device for estimating a weld puddle depth during a 3D printing process, and 3D printing system |
| CN113240668A (en) * | 2021-06-08 | 2021-08-10 | 南京师范大学 | Weld pool image quality evaluation method based on image digital feature distribution |
| CN113343996A (en) * | 2021-05-11 | 2021-09-03 | 武汉大学 | Method for calculating directional energy deposition area of plasma based on deep convolutional network |
| CN113554587A (en) * | 2021-05-31 | 2021-10-26 | 江苏大学 | Molten pool image geometric feature extraction method and system based on deep learning |
| CN113673529A (en) * | 2021-08-16 | 2021-11-19 | 连城凯克斯科技有限公司 | Semantic segmentation model training method, silicon fusion state detection method and electronic equipment |
| CN115424129A (en) * | 2022-10-13 | 2022-12-02 | 哈尔滨市科佳通用机电股份有限公司 | Abnormal detection method and detection system for wallboard damage |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20110216965A1 (en) * | 2010-03-05 | 2011-09-08 | Microsoft Corporation | Image Segmentation Using Reduced Foreground Training Data |
| CN105243355A (en) * | 2015-09-09 | 2016-01-13 | 大连理工大学 | Event-driven remote wireless coalbed methane well station abnormal scene safety monitoring method |
-
2019
- 2019-06-29 CN CN201910581281.7A patent/CN110363781B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20110216965A1 (en) * | 2010-03-05 | 2011-09-08 | Microsoft Corporation | Image Segmentation Using Reduced Foreground Training Data |
| CN105243355A (en) * | 2015-09-09 | 2016-01-13 | 大连理工大学 | Event-driven remote wireless coalbed methane well station abnormal scene safety monitoring method |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113118465A (en) * | 2019-12-31 | 2021-07-16 | 韩国科学技术院 | Method and device for estimating a weld puddle depth during a 3D printing process, and 3D printing system |
| CN113118465B (en) * | 2019-12-31 | 2023-08-22 | 韩国科学技术院 | Method and apparatus for estimating puddle depth during 3D printing process, and 3D printing system |
| WO2021068982A1 (en) * | 2020-03-19 | 2021-04-15 | 南通大学 | Quaternion multi-degree-of-freedom neuron-based multispectral welding image recognition method |
| CN111932539A (en) * | 2020-10-13 | 2020-11-13 | 南京知谱光电科技有限公司 | Molten pool image and depth residual error network-based height and penetration collaborative prediction method |
| CN111932539B (en) * | 2020-10-13 | 2021-02-02 | 南京知谱光电科技有限公司 | Molten pool image and depth residual error network-based height and penetration collaborative prediction method |
| CN113077423A (en) * | 2021-03-22 | 2021-07-06 | 中国人民解放军空军工程大学 | Laser selective melting pool image analysis system based on convolutional neural network |
| CN112906833A (en) * | 2021-05-08 | 2021-06-04 | 武汉大学 | Plasma energy deposition image identification method based on full convolution neural network |
| CN112906833B (en) * | 2021-05-08 | 2021-08-17 | 武汉大学 | Plasma energy deposition image identification method based on full convolution neural network |
| CN113343996A (en) * | 2021-05-11 | 2021-09-03 | 武汉大学 | Method for calculating directional energy deposition area of plasma based on deep convolutional network |
| CN113554587A (en) * | 2021-05-31 | 2021-10-26 | 江苏大学 | Molten pool image geometric feature extraction method and system based on deep learning |
| CN113554587B (en) * | 2021-05-31 | 2024-05-14 | 江苏大学 | Deep learning-based molten pool image geometric feature extraction method and system |
| CN113240668A (en) * | 2021-06-08 | 2021-08-10 | 南京师范大学 | Weld pool image quality evaluation method based on image digital feature distribution |
| CN113240668B (en) * | 2021-06-08 | 2024-04-16 | 南京师范大学 | Image digital feature distribution-based generated molten pool image quality evaluation method |
| CN113673529A (en) * | 2021-08-16 | 2021-11-19 | 连城凯克斯科技有限公司 | Semantic segmentation model training method, silicon fusion state detection method and electronic equipment |
| CN115424129A (en) * | 2022-10-13 | 2022-12-02 | 哈尔滨市科佳通用机电股份有限公司 | Abnormal detection method and detection system for wallboard damage |
| CN115424129B (en) * | 2022-10-13 | 2023-08-11 | 哈尔滨市科佳通用机电股份有限公司 | Abnormality detection method and abnormality detection system for wallboard damage |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110363781B (en) | 2022-08-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110363781A (en) | Molten bath profile testing method based on deep neural network | |
| CN110570410B (en) | Detection method for automatically identifying and detecting weld defects | |
| CN107644415B (en) | A kind of text image method for evaluating quality and equipment | |
| CN113239930B (en) | Glass paper defect identification method, system, device and storage medium | |
| CN106709936A (en) | Single target tracking method based on convolution neural network | |
| CN110853015A (en) | Aluminum profile defect detection method based on improved Faster-RCNN | |
| CN111753805B (en) | Method and device for detecting wearing of safety helmet | |
| CN112070727B (en) | Metal surface defect detection method based on machine learning | |
| CN110308151A (en) | A kind of surface of workpiece defect identification method and device based on machine vision | |
| CN109086652A (en) | Handwritten word model training method, Chinese characters recognition method, device, equipment and medium | |
| CN110148122A (en) | Apple presentation quality stage division based on deep learning | |
| CN104867144A (en) | IC element solder joint defect detection method based on Gaussian mixture model | |
| CN104867145B (en) | IC element welding point defect detection methods based on VIBE models | |
| CN110020691B (en) | Liquid crystal screen defect detection method based on convolutional neural network impedance type training | |
| CN111062961A (en) | Contact lens edge defect detection method based on deep learning | |
| CN114926407A (en) | Steel surface defect detection system based on deep learning | |
| CN110263794A (en) | Safety belt images steganalysis method based on data enhancing | |
| CN106934455A (en) | Remote sensing image optics adapter structure choosing method and system based on CNN | |
| CN112132196A (en) | Cigarette case defect identification method combining deep learning and image processing | |
| CN109685030A (en) | A kind of mug rim of a cup defects detection classification method based on convolutional neural networks | |
| CN116385758A (en) | Detection method for damage to surface of conveyor belt based on YOLOv5 network | |
| CN115526852A (en) | Molten pool and splash monitoring method in selective laser melting process based on target detection and application | |
| CN110135423A (en) | The training method and optical character recognition method of text identification network | |
| CN109815957A (en) | A kind of character recognition method based on color image under complex background | |
| CN116452505A (en) | Continuous casting billet internal defect detection and rating method based on improved YOLOv5 |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |