CN110335238A - A kind of Automotive Paint Film Properties defect recognition system constituting method of deep learning - Google Patents
A kind of Automotive Paint Film Properties defect recognition system constituting method of deep learning Download PDFInfo
- Publication number
- CN110335238A CN110335238A CN201910378419.3A CN201910378419A CN110335238A CN 110335238 A CN110335238 A CN 110335238A CN 201910378419 A CN201910378419 A CN 201910378419A CN 110335238 A CN110335238 A CN 110335238A
- Authority
- CN
- China
- Prior art keywords
- paint film
- deep learning
- film properties
- defect
- defect area
- 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 128
- 239000003973 paint Substances 0.000 title claims abstract description 62
- 238000013135 deep learning Methods 0.000 title claims abstract description 47
- 238000000034 method Methods 0.000 title claims abstract description 22
- 238000005070 sampling Methods 0.000 claims abstract description 57
- 238000001514 detection method Methods 0.000 claims abstract description 39
- 238000012360 testing method Methods 0.000 claims description 7
- 238000013461 design Methods 0.000 claims description 4
- 238000002360 preparation method Methods 0.000 claims description 3
- 230000011218 segmentation Effects 0.000 claims description 3
- 238000006467 substitution reaction Methods 0.000 claims description 3
- 238000012546 transfer Methods 0.000 claims description 3
- 239000011248 coating agent Substances 0.000 claims description 2
- 238000000576 coating method Methods 0.000 claims description 2
- 230000005540 biological transmission Effects 0.000 claims 1
- 235000013399 edible fruits Nutrition 0.000 claims 1
- 238000012545 processing Methods 0.000 claims 1
- 230000007812 deficiency Effects 0.000 abstract description 4
- 238000010422 painting Methods 0.000 abstract description 3
- 230000002708 enhancing effect Effects 0.000 description 3
- 238000012986 modification Methods 0.000 description 3
- 230000004048 modification Effects 0.000 description 3
- 238000010276 construction Methods 0.000 description 2
- 238000009795 derivation Methods 0.000 description 2
- 238000007689 inspection Methods 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 208000003464 asthenopia Diseases 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000005520 cutting process Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 210000005036 nerve Anatomy 0.000 description 1
- 230000002035 prolonged effect Effects 0.000 description 1
- 238000013138 pruning Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
- G06T2207/30156—Vehicle coating
Abstract
The invention discloses a kind of Automotive Paint Film Properties defect recognition system constituting methods of deep learning, include the following steps: that acquiring several vehicle body paint film original images to be detected is labeled, the mark includes the defects of mark vehicle body paint film original image region and classification, an xml formatted file is generated to the picture after every mark after mark, forms mark sample set；Multi-angle several times is carried out using sampling block to each defect area marked in every picture of mark sample set and cuts out sampling, obtains defect area sampling set；Sampling block is square；Defect area sampling set is trained and is tested using deep learning algorithm, building obtains the Automotive Paint Film Properties defect recognition model based on deep learning；The Automotive Paint Film Properties defect recognition system based on deep learning of building.The present invention can effectively car assisted film defect identification, make up manually visualize detection deficiency, improve body of a motor car painting quality.
Description
Technical field
The present invention relates to depth learning technology fields, and in particular to a kind of Automotive Paint Film Properties defect recognition system of deep learning
Construction method.
Background technique
Painting dressing automobiles are a vital links in automobile production manufacturing process, and the vehicle body after carrying out coating needs to carry out
The detection and modification of surface film defect.Traditional industrial wire defect detection system is rechecked using human eye initial survey and manually, due to
It by resolution of eye, differentiates speed and worker's subjective consciousness is examined to be influenced, and prolonged intensive operational and white lamps
The reflection of light will lead to the visual fatigue of worker, and the efficiency of artificial detection is not high, missing inspection usually occurs.
Summary of the invention
In view of the deficiencies of the prior art, the present invention is intended to provide a kind of Automotive Paint Film Properties defect recognition system structure of deep learning
Construction method, can effectively car assisted film defect identification, make up manually visualize detection deficiency, improve body of a motor car paint film
Quality.
To achieve the goals above, the present invention adopts the following technical scheme:
A kind of Automotive Paint Film Properties defect recognition system constituting method of deep learning, includes the following steps:
S1, several vehicle body paint film original images to be detected of acquisition obtain original sample collection, are labeled to original sample collection,
The mark includes the defects of mark vehicle body paint film original image region and classification；LabelImg tool is used when mark, to vehicle
Body paint film original image carries out circle choosing and the mark of defect area, then generates an xml format text to the picture after every mark
Part forms mark sample set to carry out data preparation and unified specification operation；
S2, each defect area marked in every picture of mark sample set through step S1 is carried out using sampling block
Multi-angle cuts out sampling several times, obtains defect area sampling set；Sampling block is square；
S3, defect area sampling set obtained in step S2 is trained and is tested using deep learning algorithm, constructed
Obtain the Automotive Paint Film Properties defect recognition model based on deep learning；
S4, Automotive Paint Film Properties defect recognition system of the building based on deep learning:
The Automotive Paint Film Properties defect recognition system of the deep learning includes input module, detection categorization module, result treatment
Module；
The input module is associated with the detection categorization module, for being passed to Automotive Paint Film Properties figure to be detected for user
Piece；
The detection categorization module respectively with the input module and result treatment module relation, for what is uploaded to user
Automotive Paint Film Properties picture to be detected is carried out using the Automotive Paint Film Properties defect recognition model obtained in step S4 based on deep learning
Defect recognition and detection；
The recognition detection result that the result treatment module is used to will test categorization module is supplied to user.
Further, in step S1, the size of every picture is unified for 300 × 300, guarantees the Aspect Ratio of defect area
It is not changed.
Further, in step S2, the size for marking the picture of sample set is indicated using (Row_Sam, Col_Sam),
Middle Row_Sam, Col_Sam respectively indicate the width and height of picture；Sampling block is square, and top left co-ordinate is (leftx,
lefty), wide and height is Bm；It cuts out every time to leftxAnd lefty10 times are carried out in respective value range uniformly random to take
Value；The leftxMinimum value leftx_minFor 0 and rowmaxThe larger value in-Bm, maximum value leftx_maxValue range be
rowminWith the smaller value in Row_Sam-Bm；leftyMinimum value lefty_minValue is 0 and colmaxThe larger value in-Bm,
Maximum value lefty_maxValue is colminWith the smaller value in Col_Sam-Bm；
Wherein, rowmaxFor the abscissa in the lower right corner of defect area, rowminFor the abscissa in the upper left corner of defect area；
colmaxFor the ordinate in the defect area lower right corner, colminFor the ordinate in the defect area upper left corner.
Further, in step S2, the width and high Bm of sampling block random value in its selection section, to different defects
The method of determination in the selection section of the sampling block size of area size is as follows:
If r < 50and c < 50Select the Square (125,175)
Else if r < 100and c < 100Select the Square (225,375)
Else r < 150and c < 150Select the Square (425,575)
In formula, r indicates that the width of defect area, c indicate the height of defect area；If defect area is wide less than 50 and high
Less than 50, then the wide and high Bm of sampling block selects section for (125,175)；If defect area is wider than or is equal to 50 simultaneously
And less than 100, height is greater than or equal to 50 and less than 100, then the wide and high Bm of sampling block selects section for (225,375),
If defect area is wider than or is equal to 100 and less than 150, height is greater than or equal to 100 and less than 150, then sampling block
Width and high Bm select the section for (425,575).
Further, in step S2, to each defect area marked in every picture of mark sample set through step S1
Sampling is cut out using sampling block 10 multi-angles of progress.
Further, the detailed process of step S3 are as follows:
All defect area aspect ratios are clustered using K-means algorithm, obtain applicable aspect ratio；According to poly-
The aspect ratio that class obtains determines the cell size of bounding box and segmentation；
MobileNet is combined with SSD algorithm, uses MobileNet as basic network substitution SSD algorithm
VGG16 network, and after the Conv13 of MobileNet layer add 8 convolutional layers, extract Conv11, Conv13, Conv14_2,
6 convolutional layers of Conv15_2, Conv16_2, Conv17_2 are as detection layers；To one size of every layer of design, wherein Conv11
Layer having a size of 19 × 19 × 512, Conv13 layers having a size of 10 × 10 × 1024, Conv14_2 layers having a size of 5 × 5 × 512,
Conv15_2 layers having a size of 3 × 3 × 256, Conv16_2 layers having a size of 2 × 2 × 256, Conv17_2 layers having a size of 1 × 1 ×
128；Thus improved MobileNet-SSD algorithm is obtained；
Defect area sampling set obtained in step S2 is trained and is surveyed using improved MobileNet-SSD algorithm
Examination, obtains the Automotive Paint Film Properties defect recognition model based on deep learning.
Further, the Automotive Paint Film Properties defect recognition system of the deep learning is user by way of Web server
The service of Automotive Paint Film Properties defect recognition is provided；
User is accessed by browser；Pass through the Automotive Paint Film Properties defect recognition system of the access deep learning when access
URL opens webpage, uploading pictures；Input module calls detection categorization module after receiving picture, as the vapour based on deep learning
The input of vehicle film defect identification model, the Automotive Paint Film Properties defect recognition model based on deep learning will identify and testing result passes
Enter result treatment module, the results are shown on the display page for result treatment module；
Or, user by calling interface, sends HTTP request directly by picture transfer to input module, through detection classification mould
After block exports recognition detection result, user is directly returned result to by result treatment module.
The beneficial effects of the present invention are: it proposes a kind of off-line data enhancing strategy of vehicle body paint film defect sample collection, leads to
Crossing off-line data enhancing expands sample set, and by establishing sample database, data is enable sufficiently to be trained to and apply.It mentions
A kind of new vehicle paint image data enhances strategy out, is labeled respectively to picture, formulates Pruning strategy cutting and multi-angle sanction
It cuts, improves detection accuracy.The Automotive Paint Film Properties defect recognition system of deep learning, effective auxiliary steam are obtained using present invention building
The identification of vehicle film defect makes up the deficiency for manually visualizing detection, improves body of a motor car painting quality.
Specific embodiment
The invention will be further described below, it should be noted that the present embodiment premised on the technical program,
The detailed implementation method and specific operation process are given, but protection scope of the present invention is not limited to the present embodiment.
The present embodiment provides a kind of Automotive Paint Film Properties defect recognition system constituting methods of deep learning, include the following steps:
S1, several vehicle body paint film original images to be detected of acquisition obtain original sample collection, are labeled to original sample collection,
The mark includes the defects of mark vehicle body paint film original image region and classification；1abelImg tool is used when mark, to vehicle
Body paint film original image carries out circle choosing and the mark of defect area, then generates an xml format text to the picture after every mark
Part forms mark sample set to carry out data preparation and unified specification operation.
Further, the size of every picture is unified for 300 × 300, guarantees that the Aspect Ratio of defect area is not changed.
S2, each defect area marked in every picture of mark sample set through step S1 is carried out using sampling block
10 times multi-angle cuts out sampling, obtains defect area sampling set；
The size for marking the picture of sample set indicates that wherein Row_Sam, Col_Sam divide using (Row_Sam, Col_Sam)
Not Biao Shi picture width and height；Sampling block is square, and top left co-ordinate is (leftx, lefty), wide and height is Bm；
It cuts out every time to leftxAnd lefty10 uniformly random values are carried out in respective value range；The leftxMinimum value
leftx_minFor 0 and rowmaxThe larger value in-Bm, maximum value leftx_maxValue range be rowminIn Row_Sam-Bm
Smaller value；leftyMinimum value lefty_minValue is 0 and colmaxThe larger value in-Bm, maximum value lefty_maxValue is
colminWith the smaller value in Col_Sam-Bm；
Wherein, rowmaxFor the abscissa in the lower right corner of defect area, rowminFor the abscissa in the upper left corner of defect area；
colmaxFor the ordinate in the defect area lower right corner, colminFor the ordinate in the defect area upper left corner.
Top left co-ordinate (left is provided in the present embodimentx, lefty) value range derivation process:
(1) sampling block upper left corner x-axis coordinate leftxMinimum value leftx_min:
When defect area is at picture position to the left, the left edge of sampling block is overlapped with the left edge of picture, at this time leftxValue
It is 0；
When defect area is in the middle position or position to the right of picture, the right edge of sampling block is right along weight with defect area
It closes, at this time leftxValue is rowmax-Bm；
Therefore leftx_minValue is 0 and rowmaxBiggish value in-Bm；
(2) upper left corner x-axis coordinate left of sampling blockxMaximum value leftx_max:
When defect area is in picture middle position or position to the left, the left edge of sampling block is overlapped with the left edge of defect area,
Left at this timex_maxValue is rowmin；
When the defect area position to the right in picture, the right edge of sampling block is overlapped with the right edge of picture, at this time leftx_maxValue
For Row_Sam-Bm；
Therefore leftx_maxValue is rowminWith the smaller value in Row_Sam-Bm；
(3) upper left corner y-axis coordinate left of sampling blockyMinimum value lefty_min:
When the defect area position on the upper side in picture, the upper edge of sampling block is overlapped with the upper edge of picture, at this time lefty_minValue
It is 0；
When defect area is in the middle position or position on the lower side of picture, the lower edge of sampling block and the lower edge weight of defect area
It closes, at this time lefty_minValue is colmax-Bm；
Therefore lefty_minValue is 0 and colmaxThe larger value in-Bm；
(4) upper left corner y-axis coordinate left of sampling blockyMaximum value lefty_max:
When defect area is in the middle position or position on the upper side of picture, the upper edge of sampling block on defect area along weight
It closes, at this time lefty_maxValue is colmin；
When the defect area position on the lower side in picture, the lower edge of sampling block is overlapped with the lower edge of picture, at this time lefty_maxFor
Col_Sam-Bm；
Therefore lefty_maxValue is colminWith the smaller value in Col_Sam-Bm.
The derivation process of above-mentioned (1)-(4) can be represented by the formula:
(row in above formulamin, colmin) and (rowmax, colmax) be defect area in picture the upper left corner and the lower right corner
Coordinate.
Further, sampling block size selection condition is as follows:
The width and high Bm of sampling block select random value in section at it, and the present embodiment is according to the statistics to sample set and divides
Analysis determines strategy to the sampling block size that different defect area dimensioneds is different:
If r < 50and c < 50Select the Square (125,175)
Else if r < 100and c < 100Select the Square (225,375)
Else r < 150and c < 150Select theSquare (425,575)
In formula, r indicates that the width of defect area, c indicate the height of defect area；If defect area is wide less than 50 and high
Less than 50, then the wide and high Bm of sampling block selects section for (125,175)；If defect area is wider than or is equal to 50 simultaneously
And less than 100, height is greater than or equal to 50 and less than 100, then the wide and high Bm of sampling block selects section for (225,375),
If defect area is wider than or is equal to 100 and less than 150, height is greater than or equal to 100 and less than 150, then sampling block
Width and high Bm select the section for (425,575).
S3, Automotive Paint Film Properties defect recognition model of the building based on deep learning:
All defect area aspect ratios are clustered using K-means algorithm, obtain applicable aspect ratio；
Specifically, the sampling picture for obtain after multi-angle is cut out to the defect area of picture is square, therefore
Sample set off-line data enhancing strategy does not change the aspect ratio of defect area, and original sample collection is used only when cluster and is gathered
Class, can save computing resource and cluster result is constant.
It can be obtained by the matching strategy of SSD algorithm, each the ratio of width to heightAnd depth-width ratioTaking in aspect ratio
It only needs to cluster its ratio of width to height with the value for being greater than 1 in depth-width ratio when being worth in range, therefore clustering, the specific method is as follows:
The cell size of bounding box and segmentation are determined according to the aspect ratio that cluster obtains.Specifically,
Bounding box is sized to 1.5 times of aspect ratio size, and cell is sized to 1.5 times of aspect ratio size；
MobileNet is combined with SSD algorithm, uses MobileNet as basic network substitution SSD algorithm
VGG16 network, and after the Conv13 of MobileNet layer add 8 convolutional layers, extract Conv11, Conv13, Conv14_2,
6 convolutional layers of Conv15_2, Conv16_2, Conv17_2 are as detection layers；To one size of every layer of design, wherein Conv11
Layer having a size of 19 × 19 × 512, Conv13 layers having a size of 10 × 10 × 1024, Conv14_2 layers having a size of 5 × 5 × 512,
Conv15_2 layers having a size of 3 × 3 × 256, Conv16_2 layers having a size of 2 × 2 × 256, Conv17_2 layers having a size of 1 × 1 ×
128；Thus improved MobileNet-SSD algorithm is obtained；
Defect area sampling set obtained in step S2 is trained and is surveyed using improved MobileNet-SSD algorithm
Examination, obtains the Automotive Paint Film Properties defect recognition model based on deep learning；
S4, Automotive Paint Film Properties defect recognition system of the building based on deep learning:
The Automotive Paint Film Properties defect recognition system of the deep learning includes input module, detection categorization module, result treatment
Module；
The input module is associated with the detection categorization module, for being passed to Automotive Paint Film Properties figure to be detected for user
Piece；
The detection categorization module respectively with the input module and result treatment module relation, for what is uploaded to user
Automotive Paint Film Properties picture to be detected carries out defect recognition and inspection using the detection identification model of deep learning obtained in step S4
It surveys；
The recognition detection result that the result treatment module is used to will test categorization module is supplied to user.
Further, the Automotive Paint Film Properties defect recognition system of the deep learning can be used by way of Web server
Family provides the service of Automotive Paint Film Properties defect recognition；
User can be accessed by browser；Pass through the Automotive Paint Film Properties defect recognition system of the access deep learning when access
URL open webpage, uploading pictures；Input module calls detection categorization module, the detection as deep learning after receiving picture
The detection identification model of the input of identification model, deep learning will identify and testing result is passed to result treatment module, as a result locates
Managing module, the results are shown on the display page；
Or, user can also send HTTP request directly by picture transfer to input module, through detection point by calling interface
After generic module exports recognition detection result, user is directly returned result to by result treatment module.
The input module is the htm1 page rendered by Web server, uses built-in CSS style and component
Bootstrap open source front end frame.
Detection categorization module is the nucleus module of system, which receives picture input, examine to the defects of picture
It surveys and classifies, then export result.When system starts, detection categorization module is first initialized, and loads trained depth
The detection identification model of habit.After starting, for the picture of each input, picture size is first adjusted to 300 × 300, then
Afferent nerve network carries out forward-propagating, obtains testing result.
For those skilled in the art, it can be provided various corresponding according to above technical solution and design
Change and modification, and all these change and modification, should be construed as being included within the scope of protection of the claims of the present invention.
Claims (7)
1. a kind of Automotive Paint Film Properties defect recognition system constituting method of deep learning, which comprises the steps of:
S1, several vehicle body paint film original images to be detected of acquisition obtain original sample collection, are labeled to original sample collection, described
Mark includes the defects of mark vehicle body paint film original image region and classification；LabelImg tool is used when mark, to body paint
Film original image carry out defect area circle choosing and mark, then to every mark after picture generate an xml formatted file with
Data preparation and unified specification operation are carried out, mark sample set is formed；
S2, each defect area for marking through step S1 in every picture of mark sample set is carried out using sampling block it is several
Secondary multi-angle cuts out sampling, obtains defect area sampling set；Sampling block is square；
S3, defect area sampling set obtained in step S2 is trained and is tested using deep learning algorithm, building obtains
Automotive Paint Film Properties defect recognition model based on deep learning；
S4, Automotive Paint Film Properties defect recognition system of the building based on deep learning:
The Automotive Paint Film Properties defect recognition system of the deep learning includes input module, detection categorization module, result treatment module；
The input module is associated with the detection categorization module, for being passed to Automotive Paint Film Properties picture to be detected for user；
The detection categorization module is to be checked for uploading to user respectively with the input module and result treatment module relation
The Automotive Paint Film Properties picture of survey carries out defect using the Automotive Paint Film Properties defect recognition model obtained in step S4 based on deep learning
Identification and detection；
The recognition detection result that the result treatment module is used to will test categorization module is supplied to user.
2. the Automotive Paint Film Properties defect recognition system constituting method of deep learning according to claim 1, which is characterized in that step
In rapid S1, the size of every picture is unified for 300 × 300, guarantees that the Aspect Ratio of defect area is not changed.
3. the Automotive Paint Film Properties defect recognition system constituting method of deep learning according to claim 1, which is characterized in that step
In rapid S2, the size for marking the picture of sample set is indicated using (Row_Sam, Col_Sam), wherein Row_Sam, Col_Sam points
Not Biao Shi picture width and height；Sampling block is square, and top left co-ordinate is (leftx,lefty), wide and height is Bm；
It cuts out every time to leftxAnd lefty10 uniformly random values are carried out in respective value range；The leftxMinimum value
leftx_minFor 0 and rowmaxThe larger value in-Bm, maximum value leftx_maxValue range be rowminIn Row_Sam-Bm
Smaller value；leftyMinimum value lefty_minValue is 0 and colmaxThe larger value in-Bm, maximum value lefty_maxValue is
colminWith the smaller value in Col_Sam-Bm；
Wherein, rowmaxFor the abscissa in the lower right corner of defect area, rowminFor the abscissa in the upper left corner of defect area；
colmaxFor the ordinate in the defect area lower right corner, colminFor the ordinate in the defect area upper left corner.
4. the Automotive Paint Film Properties defect recognition system constituting method of deep learning according to claim 1 or 3, feature exist
In in step S2, the width and high Bm of sampling block select random value in section at it, the sampling to different defect area sizes
The method of determination in the selection section of block size is as follows:
50 Select the Square (125,175) of if r < 50and c <
100 Select the Square (225,375) of else if r < 100and c <
150 Select the Square (425,575) of else r < 150and c <
In formula, r indicates that the width of defect area, c indicate the height of defect area；If defect area is wide less than 50 and height is less than
50, then the wide and high Bm of sampling block selects section for (125,175)；If defect area is wider than or is equal to 50 and small
In 100, height is greater than or equal to 50 and less than 100, then the wide and high Bm of sampling block selects section for (225,375), if
Defect area is wider than or is equal to 100 and less than 150, and height is greater than or equal to 100 and less than 150, then the width of sampling block
Select the section for (425,575) with high Bm.
5. the Automotive Paint Film Properties defect recognition system constituting method of deep learning according to claim 1, which is characterized in that step
In rapid S2, each defect area marked in every picture of mark sample set through step S1 is carried out 10 times using sampling block
Multi-angle cuts out sampling.
6. the Automotive Paint Film Properties defect recognition system constituting method of deep learning according to claim 1, which is characterized in that step
The detailed process of rapid S3 are as follows:
All defect area aspect ratios are clustered using K-means algorithm, obtain applicable aspect ratio；According to clustering
To aspect ratio determine the cell size of bounding box and segmentation；
MobileNet is combined with SSD algorithm, uses MobileNet as the VGG16 net of basic network substitution SSD algorithm
Network, and 8 convolutional layers are added after the Conv13 of MobileNet layer, extract Conv11, Conv13, Conv14_2, Conv15_
2,6 convolutional layers of Conv16_2, Conv17_2 are as detection layers；To one size of every layer of design, wherein Conv11 layers having a size of
19 × 19 × 512, Conv13 layers having a size of 10 × 10 × 1024, Conv14_2 layers having a size of 5 × 5 × 512, Conv15_2 layers of ruler
It is very little be 3 × 3 × 256, Conv16_2 layers having a size of 2 × 2 × 256, Conv17_2 layers having a size of 1 × 1 × 128；Thus changed
Into MobileNet-SSD algorithm；
Defect area sampling set obtained in step S2 is trained and is tested using improved MobileNet-SSD algorithm,
Obtain the Automotive Paint Film Properties defect recognition model based on deep learning.
7. the Automotive Paint Film Properties defect recognition system constituting method of deep learning according to claim 1, which is characterized in that institute
The Automotive Paint Film Properties defect recognition system for stating deep learning provides Automotive Paint Film Properties defect recognition by way of Web server for user
Service；
User is accessed by browser；Pass through the URL of the Automotive Paint Film Properties defect recognition system of the access deep learning when access
Open webpage, uploading pictures；Input module calls detection categorization module after receiving picture, as the automobile coating based on deep learning
The input of film defect recognition model, the Automotive Paint Film Properties defect recognition model based on deep learning will identify and the incoming knot of testing result
Fruit processing module, the results are shown on the display page for result treatment module；
Or, user, by calling interface, transmission HTTP request is defeated through detection categorization module directly by picture transfer to input module
Out after recognition detection result, user is directly returned result to by result treatment module.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910378419.3A CN110335238A (en) | 2019-05-08 | 2019-05-08 | A kind of Automotive Paint Film Properties defect recognition system constituting method of deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910378419.3A CN110335238A (en) | 2019-05-08 | 2019-05-08 | A kind of Automotive Paint Film Properties defect recognition system constituting method of deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110335238A true CN110335238A (en) | 2019-10-15 |
Family
ID=68140060
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910378419.3A Pending CN110335238A (en) | 2019-05-08 | 2019-05-08 | A kind of Automotive Paint Film Properties defect recognition system constituting method of deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110335238A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111127699A (en) * | 2019-11-25 | 2020-05-08 | 爱驰汽车有限公司 | Method, system, equipment and medium for automatically recording automobile defect data |
| CN111583183A (en) * | 2020-04-13 | 2020-08-25 | 成都数之联科技有限公司 | Data enhancement method and system for PCB image defect detection |
| CN112651941A (en) * | 2020-12-25 | 2021-04-13 | 北京巅峰科技有限公司 | Vehicle defect identification method and device, electronic device and storage medium |
| CN113344847A (en) * | 2021-04-21 | 2021-09-03 | 安徽工业大学 | Long tail clamp defect detection method and system based on deep learning |
| FR3116119A1 (en) | 2020-11-12 | 2022-05-13 | Saint-Gobain Glass France | Method and system for inspecting transparent glazing fitted with a gasket |
| CN115619726A (en) * | 2022-09-29 | 2023-01-17 | 西安理工大学 | Lightweight defect detection method for detecting defects of automobile body paint surface |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106952250A (en) * | 2017-02-28 | 2017-07-14 | 北京科技大学 | A kind of metal plate and belt detection method of surface flaw and device based on Faster R CNN networks |
| CN107328793A (en) * | 2017-06-30 | 2017-11-07 | 航天新长征大道科技有限公司 | A kind of ornaments surface word print flaw detection method and device based on machine vision |
| CN107748893A (en) * | 2017-09-29 | 2018-03-02 | 阿里巴巴集团控股有限公司 | Lift the method, apparatus and server of car damage identification image recognition result |
| CN108319949A (en) * | 2018-01-26 | 2018-07-24 | 中国电子科技集团公司第十五研究所 | Mostly towards Ship Target Detection and recognition methods in a kind of high-resolution remote sensing image |
| CN109711474A (en) * | 2018-12-24 | 2019-05-03 | 中山大学 | A kind of aluminium material surface defects detection algorithm based on deep learning |
-
2019
- 2019-05-08 CN CN201910378419.3A patent/CN110335238A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106952250A (en) * | 2017-02-28 | 2017-07-14 | 北京科技大学 | A kind of metal plate and belt detection method of surface flaw and device based on Faster R CNN networks |
| CN107328793A (en) * | 2017-06-30 | 2017-11-07 | 航天新长征大道科技有限公司 | A kind of ornaments surface word print flaw detection method and device based on machine vision |
| CN107748893A (en) * | 2017-09-29 | 2018-03-02 | 阿里巴巴集团控股有限公司 | Lift the method, apparatus and server of car damage identification image recognition result |
| CN108319949A (en) * | 2018-01-26 | 2018-07-24 | 中国电子科技集团公司第十五研究所 | Mostly towards Ship Target Detection and recognition methods in a kind of high-resolution remote sensing image |
| CN109711474A (en) * | 2018-12-24 | 2019-05-03 | 中山大学 | A kind of aluminium material surface defects detection algorithm based on deep learning |
Non-Patent Citations (4)
| Title |
|---|
| LEOPOLDO ARMESTO 等: "Inspection system based on artificial vision for paint defects detection on cars bodies", 《2011 IEEE INTERNATIONAL CONFERENCE ON ROBOTICS AND AUTOMATION》 * |
| YITING LI 等: "Research on a Surface Defect Detection Algorithm Based on MobileNet-SSD", 《APPLIED SCIENCES》 * |
| 林博文: "基于深度学习的钢材表面缺陷识别方法研究", 《万方数据：硕士学位论文》 * |
| 骆亚微: "基于图像处理的车身漆膜缺陷识别与分类研究", 《中国优秀硕士学位论文全文数据库工程科技Ⅰ辑》 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111127699A (en) * | 2019-11-25 | 2020-05-08 | 爱驰汽车有限公司 | Method, system, equipment and medium for automatically recording automobile defect data |
| CN111583183A (en) * | 2020-04-13 | 2020-08-25 | 成都数之联科技有限公司 | Data enhancement method and system for PCB image defect detection |
| FR3116119A1 (en) | 2020-11-12 | 2022-05-13 | Saint-Gobain Glass France | Method and system for inspecting transparent glazing fitted with a gasket |
| WO2022101291A1 (en) | 2020-11-12 | 2022-05-19 | Saint-Gobain Glass France | Method and system for inspecting transparent glazing equipped with a spline |
| CN112651941A (en) * | 2020-12-25 | 2021-04-13 | 北京巅峰科技有限公司 | Vehicle defect identification method and device, electronic device and storage medium |
| CN113344847A (en) * | 2021-04-21 | 2021-09-03 | 安徽工业大学 | Long tail clamp defect detection method and system based on deep learning |
| CN113344847B (en) * | 2021-04-21 | 2023-10-31 | 安徽工业大学 | Deep learning-based long tail clamp defect detection method and system |
| CN115619726A (en) * | 2022-09-29 | 2023-01-17 | 西安理工大学 | Lightweight defect detection method for detecting defects of automobile body paint surface |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110335238A (en) | A kind of Automotive Paint Film Properties defect recognition system constituting method of deep learning | |
| CN108765412B (en) | Strip steel surface defect classification method | |
| CN109636772A (en) | The defect inspection method on the irregular shape intermetallic composite coating surface based on deep learning | |
| CN109977808A (en) | A kind of wafer surface defects mode detection and analysis method | |
| CN106296666B (en) | A kind of color image removes shadow method and application | |
| CN109087274A (en) | Electronic device defect inspection method and device based on multidimensional fusion and semantic segmentation | |
| CN110992317A (en) | PCB defect detection method based on semantic segmentation | |
| CN103745239B (en) | A kind of forest reserves measuring method based on satellite remote sensing technology | |
| CN101063660B (en) | Method for detecting textile defect and device thereof | |
| CN102974551A (en) | Machine vision-based method for detecting and sorting polycrystalline silicon solar energy | |
| CN113554080A (en) | Non-woven fabric defect detection and classification method and system based on machine vision | |
| CN113505865B (en) | Sheet surface defect image recognition processing method based on convolutional neural network | |
| CN104318051B (en) | The rule-based remote sensing of Water-Body Information on a large scale automatic extracting system and method | |
| CN104680185B (en) | Hyperspectral image classification method based on boundary point reclassification | |
| CN107305691A (en) | Foreground segmentation method and device based on images match | |
| CN110097136A (en) | Image classification method neural network based | |
| CN111160451A (en) | Flexible material detection method and storage medium thereof | |
| CN111257329A (en) | Smartphone camera defect detection method and detection system | |
| CN111415330A (en) | Copper foil appearance defect detection method based on deep learning | |
| CN111161237A (en) | Fruit and vegetable surface quality detection method, storage medium and sorting device thereof | |
| CN107576660B (en) | A kind of double yellow duck egg Automatic Visual Inspection method based on apart from contour | |
| CN113222926A (en) | Zipper abnormity detection method based on depth support vector data description model | |
| CN113221881A (en) | Multi-level smart phone screen defect detection method | |
| CN201081763Y (en) | Textile defect detector | |
| CN104732238B (en) | The method of gray level image texture feature extraction based on orientation selectivity |
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: 20191015 |