CN107123109A - A kind of window sliding algorithm detected for Bridge Crack - Google Patents
A kind of window sliding algorithm detected for Bridge Crack Download PDFInfo
- Publication number
- CN107123109A CN107123109A CN201710207983.XA CN201710207983A CN107123109A CN 107123109 A CN107123109 A CN 107123109A CN 201710207983 A CN201710207983 A CN 201710207983A CN 107123109 A CN107123109 A CN 107123109A
- Authority
- CN
- China
- Prior art keywords
- bridge
- mrow
- bin
- bridge crack
- msub
- 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
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
- G06T7/0004—Industrial image inspection
- G06T7/0008—Industrial image inspection checking presence/absence
-
- 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
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
Abstract
The invention discloses a kind of window sliding algorithm detected for Bridge Crack, comprise the following steps:1) builds data set, and data set is divided into after training set and checking collection, is trained in DBCC disaggregated models；2) combinations DBCC disaggregated models, detect Bridge Crack.Detect the calculating process of Bridge Crack as shown in following formula:Wherein, PbAnd P (x)c(x) it is the probability that bridge background bin and Bridge Crack bin are recognized in this identification process；F (x) is 0, and it is bridge background bin to represent current bridge bin graphic picture, and f (x) is 1, and it is Bridge Crack bin to represent current bridge bin graphic picture, and t is that a probability distinguishes threshold value.The present invention compared with directly using the Bridge Crack that traditional window sliding algorithm combination DBCC models are carried out to detect, the present invention for Bridge Crack detection more accurately, it is stronger to the rejection ability of bridge facet noise.
Description
Technical field
The invention belongs to image procossing and computer vision field, specifically related to a kind of window detected for Bridge Crack
Slide algorithm.
Background technology
Bridge needs regularly to make its health status and commented as the hinge of the traffic systems such as road, highway, railway
Estimate, and Bridge Crack drastically influence the safe operation of bridge, even more serious meeting hair as one of topmost bridge defect
Raw bridge ruins the accident that people dies.Therefore, effective detection positioning is carried out to Bridge Crack most important.
Current Bridge Crack detection method is mainly based upon image processing algorithm progress, the core of image processing algorithm
It is the combination of window sliding algorithm and disaggregated model.Traditional window sliding algorithm is that the fixed window of picture to be detected → use exists
The enterprising line slip of picture is detected, bridge picture bin → examining bridge picture bin using disaggregated model → is obtained and judges the picture
It is Bridge Crack bin or bridge background bin.Conventional disaggregated model is CIFAR10 models or DBCC models.
If directly being possible to detect using traditional window sliding algorithm and DBCC models couplings detection Bridge Crack
Excessive bridge noise bin, so as to cause detection error big.Traditional window sliding algorithm can not be the reason for denoising:
DBCC models have finally used a softmax function, export respectively in this identification process Bridge background bin and bridge
The probability P of crack binbAnd P (x)c(x), then by both probability make comparisons.If only Pb(x)≤Pc(x), it is considered as current
Bridge bin picture be Bridge Crack bin picture, then when view picture Bridge Crack picture recognition, it is possible to by certain
A little bridge noise bins are mistakenly identified as Bridge Crack bin.
The content of the invention
Can not denoising, this big technical problem of detection error, present invention offer in order to solve traditional sliding window algorithm
A kind of window sliding algorithm detected for Bridge Crack.The technical problem to be solved in the present invention is real by the following technical programs
It is existing:
A kind of window sliding algorithm detected for Bridge Crack, comprises the following steps:
Step 1: building data set, data set is divided into after training set and checking collection, instructed in DBCC disaggregated models
Practice；
Step 2: with reference to DBCC disaggregated models, detecting Bridge Crack, the calculating process such as formula (1) of Bridge Crack is detected
It is shown:
Wherein, PbAnd P (x)c(x) be in this identification process, what bridge background bin and Bridge Crack bin were recognized
Probability；F (x) is 0, and it is bridge background bin to represent current bridge bin graphic picture, and f (x) is 1, represents current bridge bin
Image is Bridge Crack bin, and t is that a probability distinguishes threshold value.
A kind of above-mentioned window sliding algorithm detected for Bridge Crack, the value of the t is 0.90~0.99.
A kind of above-mentioned window sliding algorithm detected for Bridge Crack, the step one is specially:
S101. image capture device collection different background texture, a large amount of Bridge Crack pictures of unlike material are utilized；
S102. the picture in step one is divided into artificial amplification data collection and test data set；
S103. using the window of W*H fixed sizes on the picture that artificial amplification data is concentrated it is nonoverlapping enter line slip,
Meanwhile, the small section of the Bridge Crack picture under sliding window is covered is used as a ROI area-of-interest, specific calculating process
As shown in formula (2):
Wherein, W and H is the wide and height of sliding window, imgRoiLx、imgRoiLyFor ROI area-of-interests upper left angle point
Transverse and longitudinal coordinate, imgRoiRx、imgRoiRyFor the transverse and longitudinal coordinate of ROI area-of-interests bottom right angle point, wherein i and j calculating process
As shown in formula (3):
Wherein, srcImgw、srcImghThe respectively wide and height of picture；
S104. by selecting mark, a RGB data collection is constituted, RGB data collection is divided into training set and checking collects；
S105. it is trained using the training set in S104 and checking collection in DBCC disaggregated models.
Compared with prior art, beneficial effects of the present invention:
The present invention is the bridge identified to DBCC identification models by being slided in legacy windows on the basis of algorithm
The probability of crack bin is distinguished threshold value t using probability and further discriminated between.Therefore, the present invention is with directly using traditional window
The Bridge Crack detection for sliding the progress of algorithm combination DBCC models is compared, and the present invention is more accurate for the detection of Bridge Crack
Really, stronger to the rejection ability of bridge facet noise, these are tried to achieve more accurate crack for the Bridge Crack detection later stage and determined
Position position and flaw area be all extremely important basis.
Brief description of the drawings
Fig. 1 is the flow chart of the present invention；
Fig. 2 is the flow chart of Bridge Crack detection positioning；
Fig. 3 is the structural representation of DBCC models.
Embodiment
Further detailed description is done to the present invention with reference to specific embodiment, but embodiments of the present invention are not limited to
This.
Embodiment 1:
A kind of reference picture 1, window sliding algorithm detected for Bridge Crack is specific as follows:
First step utilizes image capture device collection different background texture, a large amount of Bridge Crack pictures of unlike material；
Picture in step one is divided into artificial amplification data collection and test data set by second step；
3rd step nonoverlapping on the picture that artificial amplification data is concentrated is slided using the window of W*H fixed sizes
It is dynamic, meanwhile, the small section of the Bridge Crack picture under sliding window is covered is specific to calculate as a ROI area-of-interest
Shown in process such as formula (1):
Wherein, W and H is the wide and height of sliding window, imgRoiLx、imgRoiLyFor ROI area-of-interests upper left angle point
Transverse and longitudinal coordinate, imgRoiRx、imgRoiRyFor the transverse and longitudinal coordinate of ROI area-of-interests bottom right angle point, wherein i and j calculating process
As shown in formula (2):
Wherein, srcImgw、srcImghThe respectively wide and height of picture；
4th step constitutes a RGB data collection by selecting mark, and RGB data collection is divided into training set and checking collects；
5th step is trained using the training set in step 4 and checking collection in DBCC disaggregated models；
6th step combination DBCC disaggregated models, detect Bridge Crack, detect the calculating process such as formula (3) of Bridge Crack
It is shown:
Wherein, PbAnd P (x)c(x) be in this identification process, the small section recognized be respectively bridge background bin and
The probability of Bridge Crack bin；F (x) is 0, and it is bridge background bin to represent current bridge bin graphic picture, and f (x) is 1, is represented
Current bridge bin graphic picture is Bridge Crack bin, and t is that a probability distinguishes threshold value, and preferably t value is 0.90~0.99.
The present invention is the bridge identified to DBCC identification models by being slided in legacy windows on the basis of algorithm
The probability of crack bin is distinguished threshold value t using probability and further discriminated between.Therefore, the present invention is with directly using traditional window
The Bridge Crack detection for sliding the progress of algorithm combination DBCC models is compared, and the present invention is more accurate for the detection of Bridge Crack
Really, stronger to the rejection ability of bridge facet noise, these are tried to achieve more accurate crack for the Bridge Crack detection later stage and determined
Position position and flaw area be all extremely important basis.
Embodiment 2:
The operating process of reference picture 2, in detail narration Bridge Crack detection positioning:
The first step, the Bridge Crack picture of five kinds of different background textures, unlike material is gathered using image capture device, is adopted
The picture sum of collection is 2000, and all pictures are normalized into the picture of 1024*1024 resolution ratio, by this 2000 figures
Piece is divided into 2 data sets, artificial amplification data collection and test data set, each data set 1000；
Second step, using the window of W*H fixed sizes on 1000 pictures that artificial amplification data is concentrated it is nonoverlapping
Enter line slip, meanwhile, the small section of the Bridge Crack picture under sliding window is covered is used as a ROI area-of-interest.Its
In, the small sectioning image comprising bridge background is referred to as bridge background bin, and the small section comprising Bridge Crack is referred to as Bridge Crack
Bin, detailed process is as shown in following formula:
imgRoiLx=i*W
imgRoiLy=j*H
imgRoiRx=i*W+W
imgRoiRy=j*H+H
Wherein, W and H is the wide and height of sliding window, coordinate (imgRoiLx,imgRoiLy) be ROI region upper left angle point
Coordinate, coordinate (imgRoiRx,imgRoiRy) be ROI region bottom right angular coordinate, wherein i and the j formula for being calculated as follows face
It is shown:
I=srcImgw/W
J=srcImgh/H
Wherein, srcImgw, srcImgh, it is respectively, by the wide and height of the Bridge Crack picture of window sliding, to make srcImgw
=srcImgh=1024pixel；
3rd step, to based on window sliding algorithm expand and come data set select, mark, classify composition one have
The training set for being used to train deep learning model of certain scale and checking collect；
4th step, the data set pre-processed using the 3rd step is trained to DBCC disaggregated models；
5th step, gathers a Bridge Crack picture, picture is normalized to the picture of 1024*1024 resolution ratio, then
Bridge Crack image is sampled using image gaussian pyramid, the Bridge Crack picture of a width low resolution is obtained；
6th step, the low resolution tried to achieve using the DBCC disaggregated models and window sliding algorithm that train in previous step
Detected on Bridge Crack picture, and the transverse and longitudinal of all Bridge Crack bins identified by DBCC disaggregated models is sat
Mark is recorded, and wherein window sliding algorithm is as shown in following formula:
Wherein, PbAnd P (x)c(x) be in this identification process, the small section recognized be respectively bridge background bin and
The probability of Bridge Crack bin；F (x) is 0, and it is bridge background bin to represent current bridge bin graphic picture, and f (x) is 1, is represented
Current bridge bin graphic picture is Bridge Crack bin, and t is that a probability distinguishes threshold value, and general value is 0.90~0.99；
7th step, the transverse and longitudinal coordinate point of all Bridge Crack bins recorded using simple selection sort algorithm to the 6th step
It is not ranked up, minimum transverse and longitudinal coordinate is constituted into a coordinate points, maximum transverse and longitudinal coordinate is also constituted into a coordinate points, and
The two coordinate points are substituted into following formula, two new coordinate points are solved, specific formula is as follows:
Xh=(xl) * 2n
Yh=(yl) * 2n
Wherein coordinate (xl,yl) it is defined location coordinate, coordinate (x on low resolution pictureh,yh) it is high resolution graphics
Piece defined location coordinate, n representative images gaussian pyramid is to the number of times of down-sampling, and usual n value is 2；
8th step, according to the 7th step solve come two new coordinate points a rectangle is determined in high-resolution pictures
Region, and this rectangular area is set to ROI area-of-interests；
9th step, using the DBCC disaggregated models and window sliding algorithm trained 1024*1024 resolution ratio picture
The upper detection for carrying out Bridge Crack, during Bridge Crack is detected, records the transverse and longitudinal coordinate of all Bridge Crack bins,
, in this course, further the number of Bridge Crack bin graphic picture is counted, and this statistical number is designated as N；
Tenth step, the transverse and longitudinal coordinate of all Bridge Crack bins recorded using simple selection sort algorithm to the 9th step
It is ranked up, and picks out the transverse and longitudinal coordinate and maximum transverse and longitudinal coordinate of minimum and constitute two new position coordinates points, then
A rectangular area is drawn in high-resolution pictures according to the two new position coordinates points, the rectangular area and the two are new
Coordinate points just identify position of the Bridge Crack in Bridge Crack picture；
Determining the method for rectangular area is:Call rectangle () this letter in the computer vision storehouse OpenCv that increases income
Number, it is incoming select after two new coordinate points (i.e. the coordinate in the rectangular area upper left corner and the lower right corner), just can obtain rectangle region
Domain；
11st step, the data N of the Bridge Crack bin come out in Bridge Crack detection process is brought into following
Formula can be to solve the area of Bridge Crack, and specific formula is as follows:
areacrack=N*W*H
Wherein areacrackFinally to solve the area for the Bridge Crack come, N represents the number of Bridge Crack bin, and W is
The width of Bridge Crack bin, H is the height of Bridge Crack bin.
Embodiment 3：
Present embodiment discloses a kind of DBCC disaggregated model construction methods based on CNN deep learnings, comprise the following steps:
(1) all convolution kernels in the original image of input and the first convolutional layer are carried out by convolution summation in convolution mode,
Obtain the Feature Mapping figure of the first convolutional layer；
(2) a Relu activation primitives are added after the first convolutional layer；
(3) addition one is used for local acknowledgement's value normalization layer that picture lightness is corrected, the office after the first convolutional layer
Portion's response normalization layer improves the recognition effect of network；
(4) the Feature Mapping figure of the first convolutional layer is subjected to down-sampling in the first pond layer, reduces resolution ratio and choose
Outstanding feature, is used as the Feature Mapping figure of the first pond layer；
(5) on the second convolutional layer in convolution mode by the institute in the Feature Mapping figure and the second convolutional layer of the first pond layer
There is convolution kernel to carry out convolution summation, obtain the Feature Mapping figure of the second convolutional layer, by the Feature Mapping figure of the second convolutional layer the
Down-sampling is carried out in two pond layers, resolution ratio is reduced and chooses outstanding feature, be used as the Feature Mapping figure of the second pond layer；
(6) added the second pond after the second pond layer after a Relu activation primitives on the 3rd convolutional layer in convolution mode
All convolution kernels changed in the Feature Mapping figure and the 3rd convolutional layer of layer carry out convolution summation, and the feature for obtaining the 3rd convolutional layer is reflected
Figure is penetrated, the Feature Mapping figure of the 3rd convolutional layer is subjected to down-sampling in the 3rd pond layer, resolution ratio is reduced and chooses outstanding
Feature, is used as the Feature Mapping figure of the 3rd pond layer；
(7) added the 3rd pond after the 3rd pond layer after a Relu activation primitives on Volume Four lamination in convolution mode
The Feature Mapping figure and all convolution kernels in Volume Four lamination for changing layer carry out convolution summation, and the feature for obtaining Volume Four lamination is reflected
Penetrate figure；
(8) added after Volume Four lamination after a Relu activation primitives and to export the Feature Mapping figure of Volume Four lamination to
One full articulamentum, and, added after the first full articulamentum for preventing the Dropout layers of over-fitting；
(9) Dropout layers in step 8 obtained Feature Mapping figure are exported to the second full articulamentum, it is complete described second
Using a softmax loss functions as loss function after articulamentum, last layer is that output bridge background bin and bridge split
Stitch the probability of bin generic in this identification process；
Above-mentioned (1), into (9), the specific calculating process of convolutional layer such as formula (31) is shown:
Wherein, X(l-1)ForThe Feature Mapping of layer, W(l)For the convolution kernel of current convolutional layer, b(l)For bias term, f is
Activation primitive, X(l)ForThe Feature Mapping that convolutional layer finally gives；
If current layer is comprising N number of sizeFeature Mapping figure, convolution kernel size be (Kx, Ky), convolution
The sliding step of core in the x and y direction is Sx and Sy, is integer in order to which final result is removed, can be to the Feature Mapping of current layer
The border that figure addition size is pad so that Feature Mapping figure passes through after convolution, and the result of convolution can be entirely fallen within picture
Portion, then the size of Feature Mapping figure is after convolutionShown in specific calculating process such as formula (32):
Wherein, l represents current layer number, and l-1 represents preceding layer.
Above-mentioned (1), into (9), the main function of pond layer is to carry out down-sampling to Feature Mapping figure, reduces Feature Mapping
The resolution ratio of figure and choose outstanding feature.Pond layer can not only substantially reduce the number of neuron, so that mould
Type has more preferable anti-noise ability.
If sub-sampling function is down (Xl), sub-sampling function generally has two kinds:Maximum pond (Maxim um
Pooling it is) and average pond (Average Pooling), it is specific to calculate as shown in formula (33):
Wherein, aiFor the neuron in pond region, RkFor the set of pond regional neuronal, | Rk| it is pond regional nerve
The total number of member, poolmax(Rk) be all neurons in pond region maximum, poolavg(Rk) it is all nerves in pond region
The average value of the value of member.The calculating of output characteristic mapping graph size is similar with convolutional layer, and specific formula for calculation refers to formula
(32)。
In order to strengthen the ability to express of network, this embodiment introduces continuous nonlinear activation function (Activation
Function).The activation primitive typically used in network has sigmod functions and rectifier (Relu) function.It is specific to calculate
As shown in formula (34):
The explanation on biology is generally acknowledged to due to activation primitive Relu, and Relu has been proved to than sigmod letter
Several fitting effects is more preferable.Therefore, the activation primitive selection in DBCC models uses Relu activation primitives.
Using the Feature Mapping figure of the first convolutional layer as input, by the calculating of Relu activation primitives, obtain strengthening network
First convolutional layer Feature Mapping figure of ability to express, will strengthen the first convolutional layer Feature Mapping figure of network ability to express as defeated
Enter, by local acknowledgement's value normalization layer, the first convolutional layer Feature Mapping figure of the Network Recognition that gets a promotion effect.
Using the Feature Mapping figure of the second pond layer as input, by the calculating of Relu activation primitives, obtain strengthening network
Second pond layer Feature Mapping figure of ability to express；Using the Feature Mapping figure of the 3rd pond layer as input, activated by Relu
The calculating of function, obtains strengthening the 3rd pond layer Feature Mapping figure of network ability to express；By the Feature Mapping of Volume Four lamination
Figure is as input, by the calculating of Relu activation primitives, obtains strengthening the Volume Four lamination Feature Mapping figure of network ability to express；
Using the Feature Mapping figure of the first full articulamentum as input, by the calculating of Relu activation primitives, obtain strengthening net list Danone
First full articulamentum Feature Mapping figure of power.
Using the Feature Mapping figure of the second full articulamentum as input, by the calculating of softmax loss functions, bridge is obtained
The probability of beam background bin and Bridge Crack the bin generic in this identification process.
In Fig. 3, In represents the image data of input, and C represents convolutional layer, and P represents pond layer, and FC represents full articulamentum, S tables
Show softmax functions, Out represents output, and Relu represents that activation primitive Relu, LRN represent local acknowledgement's value normalization, and D is represented
Dropout layers.
The present embodiment also discloses a kind of DBCC disaggregated models based on CNN deep learnings, including 4 layers of convolutional layer, 3 layers of pond
Change layer and 2 layers of full articulamentum, the DBCC disaggregated models are used as loss function using softmax loss functions, it is characterised in that,
First convolutional layer, Volume Four lamination, the second pond layer, the 3rd pond layer, it respectively with the addition of an activation behind the first full articulamentum
Function (RELU), and, local acknowledgement value normalization layer LRN is added behind the first convolutional layer, behind the first full articulamentum
Dropout layers of addition.
Specifically, DBCC is first by 4 layers of convolutional layer (C1~C4), 3 layers of pond layer (P1~P3), 2 layers of full articulamentum
(FC1~FC2), is finally used as loss function using softmax loss functions (S).In C1, C4, P2, P3, behind FC1 respectively plus
One activation primitive (RELU), meanwhile, LRN layers are added behind the first convolutional layer, dropout layers are added behind FC1.Last
Layer output bridge background bin and crack bin this two classes bin, the probable value corresponding to each class.Convolution kernel number is opened from 32
Begin, often by a convolutional layer, the number of convolution kernel is double.Untill 256.Biasing entry value is initialized as 0.1.
In convolutional layer, each convolution kernel can be seen as each width output in a feature extractor, convolutional layer
Feature Mapping figure (Feature Map) can be seen as knot of the input picture after a convolution kernel progress feature extraction
Really, it is not that each convolution kernel can be into but by carrying out visualization contrast to the output result of each convolutional layer
Work(extracts the feature of input picture, so as to obtain effective feature representation (Feature Mapping figure).Therefore, in order to strengthen convolutional layer
Expression ability, the feature to input picture is sufficiently extracted, and is compared with CIFAR10 models, and DBCC models are in each convolutional layer
It all employ more convolution kernels.
LRN completes a kind of " neighbouring to suppress " operation, has carried out normalization operation to local input area and can be used for figure
The correction of piece lightness, and Bridge Crack picture is due to illumination, the problem of picture luminance is uneven occurs in the factor such as shade, because
This, compares with CIFAR10 models, and BDCC models with the addition of local acknowledgement value normalization layer LRN, and LRN layers of addition can lift network
Recognition effect.
Dropout refers in training pattern, random to allow the node of some hidden layers in network not work temporarily, not work
Those nodes made can temporarily not think be network structure a part, but their weight is remained (temporarily not
Update), and when next sample input, random selection, they can may work again.So, sample each time is defeated
Enter, be all trained equivalent to having randomly selected a different network structure, but these different networks are but trained jointly
The weights shared are gone out.Therefore, Dropout can regard a kind of alternative combined between different learning models as, and make
It is to prevent a kind of method of over-fitting again with the different same samples of model training, therefore, Dropout can be prevented effectively
Fitting.Due to training the data set of DBCC models smaller, therefore, compensated using the Dropout of more maximum probability, Dropout
Value 0.55.
Test result indicates that, under certain condition, deeper result is better, therefore, for 16*16pixel for the depth of network
The Bridge Crack bin and bridge background bin picture of size, in order to deepen the depth of network structure, DBCC models as far as possible
The back gauge that size is 2 is with the addition of to Feature Mapping figure, to avoid script size just small Bridge Crack bin and bridge background surfaces
The size of first picture reduces too fast.
In the present embodiment, the quantity difference for the convolution kernel that four layers of each layer of convolutional layer of preferably DBCC disaggregated models are used
For:32、64、128、256.
The DBCC disaggregated models of the present embodiment use full model network structure, more by being used in each convolutional layer
Convolution kernel and addition LRN, network depth is deepened using dropout so that the DBCC disaggregated models are in identification 16*16pixel
During the small picture of resolution ratio, accuracy of identification is high.
Above content is to combine specific preferred embodiment further description made for the present invention, it cannot be assumed that
The specific implementation of the present invention is confined to these explanations.For general technical staff of the technical field of the invention,
On the premise of not departing from present inventive concept, some simple deduction or replace can also be made, should all be considered as belonging to the present invention's
Protection domain.
Claims (3)
1. a kind of window sliding algorithm detected for Bridge Crack, comprises the following steps:
Step 1: building data set, data set is divided into after training set and checking collection, is trained in DBCC disaggregated models；
Step 2: with reference to DBCC disaggregated models, Bridge Crack is detected, shown in the calculating process such as formula (1) for detecting Bridge Crack:
Wherein, PbAnd P (x)c(x) it is that in this identification process, it is general that bridge background bin and Bridge Crack bin are recognized
Rate；F (x) is 0, and it is bridge background bin to represent current bridge bin graphic picture, and f (x) is 1, represents current bridge bin graphic
As being Bridge Crack bin, t is that a probability distinguishes threshold value.
2. a kind of window sliding algorithm detected for Bridge Crack according to claim 1, it is characterised in that, the t
Value be 0.90~0.99.
3. a kind of window sliding algorithm detected for Bridge Crack according to claim 1 or 2, it is characterised in that, institute
Stating step one is specially:
S101. image capture device collection different background texture, a large amount of Bridge Crack pictures of unlike material are utilized；
S102. the picture in step one is divided into artificial amplification data collection and test data set；
S103. using the window of W*H fixed sizes on the picture that artificial amplification data is concentrated it is nonoverlapping enter line slip, together
When, the small section of the Bridge Crack picture under sliding window is covered is as a ROI area-of-interest, and specific calculating process is such as
Shown in formula (2):
<mrow>
<mtable>
<mtr>
<mtd>
<mrow>
<msub>
<mi>imgRoiL</mi>
<mi>x</mi>
</msub>
<mo>=</mo>
<mi>i</mi>
<mo>*</mo>
<mi>W</mi>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<msub>
<mi>imgRoiL</mi>
<mi>y</mi>
</msub>
<mo>=</mo>
<mi>j</mi>
<mo>*</mo>
<mi>H</mi>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<msub>
<mi>imgRoiR</mi>
<mi>x</mi>
</msub>
<mo>=</mo>
<mi>i</mi>
<mo>*</mo>
<mi>W</mi>
<mo>+</mo>
<mi>W</mi>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<msub>
<mi>imgRoiR</mi>
<mi>y</mi>
</msub>
<mo>=</mo>
<mi>j</mi>
<mo>*</mo>
<mi>H</mi>
<mo>+</mo>
<mi>H</mi>
</mrow>
</mtd>
</mtr>
</mtable>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>2</mn>
<mo>)</mo>
</mrow>
</mrow>
Wherein, W and H is the wide and height of sliding window, imgRoiLx、imgRoiLyFor the transverse and longitudinal of ROI area-of-interests upper left angle point
Coordinate, imgRoiRx、imgRoiRyFor the transverse and longitudinal coordinate of ROI area-of-interests bottom right angle point, wherein i and j calculating process such as public affairs
Shown in formula (3):
<mrow>
<mtable>
<mtr>
<mtd>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mi>s</mi>
<mi>r</mi>
<mi>c</mi>
<mi> </mi>
<mi>Im</mi>
<mi> </mi>
<msub>
<mi>g</mi>
<mi>w</mi>
</msub>
<mo>/</mo>
<mi>W</mi>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<mi>j</mi>
<mo>=</mo>
<mi>s</mi>
<mi>r</mi>
<mi>c</mi>
<mi> </mi>
<mi>Im</mi>
<mi> </mi>
<msub>
<mi>g</mi>
<mi>h</mi>
</msub>
<mo>/</mo>
<mi>H</mi>
</mrow>
</mtd>
</mtr>
</mtable>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>3</mn>
<mo>)</mo>
</mrow>
</mrow>
Wherein, srcImgw、srcImghThe respectively wide and height of picture；
S104. by selecting mark, a RGB data collection is constituted, RGB data collection is divided into training set and checking collects；
S105. it is trained using the training set in S104 and checking collection in DBCC disaggregated models.
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710028216 | 2017-01-13 | ||
| CN2017100282162 | 2017-01-13 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107123109A true CN107123109A (en) | 2017-09-01 |
Family
ID=59724637
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710207983.XA Pending CN107123109A (en) | 2017-01-13 | 2017-03-31 | A kind of window sliding algorithm detected for Bridge Crack |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107123109A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107967685A (en) * | 2017-12-11 | 2018-04-27 | 中交第二公路勘察设计研究院有限公司 | A kind of bridge pier and tower crack harmless quantitative detection method based on unmanned aerial vehicle remote sensing |
| CN109389615A (en) * | 2018-09-29 | 2019-02-26 | 佳都新太科技股份有限公司 | Coin discriminating method and processing terminal based on deep learning convolutional neural networks |
| CN111612747A (en) * | 2020-04-30 | 2020-09-01 | 重庆见芒信息技术咨询服务有限公司 | Method and system for rapidly detecting surface cracks of product |
| CN114820611A (en) * | 2022-06-29 | 2022-07-29 | 南通恒强轧辊有限公司 | Mechanical part quality evaluation method and system based on artificial intelligence |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105975972A (en) * | 2016-04-27 | 2016-09-28 | 湖南桥康智能科技有限公司 | Bridge crack detection and characteristic extraction method based on image |
-
2017
- 2017-03-31 CN CN201710207983.XA patent/CN107123109A/en active Pending
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105975972A (en) * | 2016-04-27 | 2016-09-28 | 湖南桥康智能科技有限公司 | Bridge crack detection and characteristic extraction method based on image |
Non-Patent Citations (4)
| Title |
|---|
| HENRY G.R.GOUK ETC.: ""Fast Sliding Window Classification with Convolutional Neural Networks"", 《ACM》 * |
| 刘洪公等: ""基于卷积神经网络的桥梁裂缝检测与识别"", 《河北科技大学学报》 * |
| 汪虹等: ""基于后验概率SVM的交通标志识别研究"", 《西南科技大学学报》 * |
| 陈才扣等: ""基于泛滑动窗与2DLDA的单样本人脸识别"", 《计算机应用》 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107967685A (en) * | 2017-12-11 | 2018-04-27 | 中交第二公路勘察设计研究院有限公司 | A kind of bridge pier and tower crack harmless quantitative detection method based on unmanned aerial vehicle remote sensing |
| CN109389615A (en) * | 2018-09-29 | 2019-02-26 | 佳都新太科技股份有限公司 | Coin discriminating method and processing terminal based on deep learning convolutional neural networks |
| CN111612747A (en) * | 2020-04-30 | 2020-09-01 | 重庆见芒信息技术咨询服务有限公司 | Method and system for rapidly detecting surface cracks of product |
| CN111612747B (en) * | 2020-04-30 | 2023-10-20 | 湖北煌朝智能自动化装备有限公司 | Rapid detection method and detection system for product surface cracks |
| CN114820611A (en) * | 2022-06-29 | 2022-07-29 | 南通恒强轧辊有限公司 | Mechanical part quality evaluation method and system based on artificial intelligence |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106910185B (en) | A kind of DBCC disaggregated model construction method based on CNN deep learning | |
| CN106910186A (en) | A kind of Bridge Crack detection localization method based on CNN deep learnings | |
| CN107123109A (en) | A kind of window sliding algorithm detected for Bridge Crack | |
| CN109508710A (en) | Based on the unmanned vehicle night-environment cognitive method for improving YOLOv3 network | |
| CN107742099A (en) | A kind of crowd density estimation based on full convolutional network, the method for demographics | |
| CN104036323B (en) | A kind of vehicle checking method based on convolutional neural networks | |
| CN104077577A (en) | Trademark detection method based on convolutional neural network | |
| CN108009518A (en) | A kind of stratification traffic mark recognition methods based on quick two points of convolutional neural networks | |
| CN107392130A (en) | Classification of Multispectral Images method based on threshold adaptive and convolutional neural networks | |
| CN107230202A (en) | The automatic identifying method and system of pavement disease image | |
| CN107945153A (en) | A kind of road surface crack detection method based on deep learning | |
| CN106910187A (en) | A kind of artificial amplification method of image data set for Bridge Crack detection | |
| CN106529398A (en) | Quick and accurate face detection method based on cascade structure | |
| CN110378232B (en) | Improved test room examinee position rapid detection method of SSD dual-network | |
| CN109978882A (en) | A kind of medical imaging object detection method based on multi-modal fusion | |
| CN100418469C (en) | Computerized disease sign analysis system based on tongue picture characteristics | |
| CN108596892A (en) | A kind of identification of Weld Defects based on improvement LeNet-5 models | |
| CN111191628B (en) | Remote sensing image earthquake damage building identification method based on decision tree and feature optimization | |
| CN107808376A (en) | A kind of detection method of raising one's hand based on deep learning | |
| CN109740485B (en) | Reservoir or small reservoir identification method based on spectral analysis and deep convolutional neural network | |
| CN104778713B (en) | A kind of image processing method | |
| CN107784308A (en) | Conspicuousness object detection method based on the multiple dimensioned full convolutional network of chain type | |
| CN106373146A (en) | Target tracking method based on fuzzy learning | |
| CN110097090A (en) | A kind of image fine granularity recognition methods based on multi-scale feature fusion | |
| CN112215815A (en) | Bare soil coverage automatic detection method for construction site |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20170901 |