CN110222609A - A kind of wall body slit intelligent identification Method based on image procossing - Google Patents
A kind of wall body slit intelligent identification Method based on image procossing Download PDFInfo
- Publication number
- CN110222609A CN110222609A CN201910443398.9A CN201910443398A CN110222609A CN 110222609 A CN110222609 A CN 110222609A CN 201910443398 A CN201910443398 A CN 201910443398A CN 110222609 A CN110222609 A CN 110222609A
- Authority
- CN
- China
- Prior art keywords
- wall body
- body slit
- image
- key point
- wall
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
- G06F18/2148—Generating training patterns; Bootstrap methods, e.g. bagging or boosting characterised by the process organisation or structure, e.g. boosting cascade
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
Abstract
A kind of wall body slit intelligent identification Method based on image procossing, comprising: wall images signal is obtained by CCD camera;The gray scale stretching in wall body slit region;Wall body slit graphical rule space is constructed, change of scale is carried out to the wall body slit image after stretching, obtains the expression sequence in wall body slit Image Multiscale space;Key point is chosen;The direction of key point field gradient is counted, maximum value indicates the principal direction of field gradient, forms feature vector;Wall body slit training set of images is trained using weak learning algorithm;T wall body slit Weak Classifier will be obtained after T circulation, finally by the weighted superposition of update to wall body slit strong classifier is obtained together, intelligent recognition is carried out to wall body slit using wall body slit strong classifier.The present invention greatly reduces image data redundancy, significantly improves the accuracy rate of wall body slit image recognition, reduces wall images computing redundancy degree, improve computational efficiency, it can accomplish to identify in real time, be greatly enhanced in terms of calculating speed and real-time, improve recognition efficiency.
Description
Technical field
The present invention relates to image identification technical fields, intelligently know more particularly to a kind of wall body slit based on image procossing
Other method.
Background technique
Building Trade in China obtains the development advanced by leaps and bounds, but since China's building safety mechanism is started late, especially builds
It exposes to wind and rain under the natural environments such as solarization for a long time, will receive the influence of the factors such as temperature stress, between materials for wall can generate
Gap, gap, which may be catalyzed amplification and then generate crack on surface, will cause security risk as crack constantly expands, therefore, must
Palpus effectively fracture is detected, its risk is assessed, potentially hazardous to prevent, at present to the inspection of wall mainly by people
Work inspection, heavy workload are difficult to be timely completed under adverse circumstances, and human subjective's property is strong, and reliability is low, and working efficiency is low,
Property dangerous for skyscraper.
Although existing wall body slit identification division uses image processing techniques, and existing wall body slit recognition methods
Speed is slow, low efficiency, for example, patent CN106651893A, is carried out using image procossing mode or nerve net although solving
Network algorithm still in fact handles image RGB Three-channel data, considerably increases calculation amount, can not reach intelligence certainly
The degree of dynamicization identification;And cause the calculating time more, and the forward method redundancy of the patent is higher, recognition accuracy
It is low;And the wall body slit identifying processing mode of image procossing involved in the prior art is more single, such as patent
In CN106203351A, crack differentiation is carried out just for horizontal, upright projection, characteristic parameter extraction is less, image information benefit
It is low with rate, it can not realize and accurately identify.
Summary of the invention
In order to solve the above technical problems, the present invention provides a kind of wall body slit intelligent identification Method based on image procossing,
By constructing to wall body slit graphical rule space, key point is chosen, and greatly reduces image data redundancy, and of the invention one
Wall body slit intelligent identification Method of the kind based on image procossing significantly improves the accuracy rate of wall body slit image recognition, reduces
Wall images computing redundancy degree, improve computational efficiency, can accomplish to identify in real time, this method in intelligent control, calculate speed
It is greatly enhanced in terms of degree and real-time, improves recognition efficiency;A kind of wall body slit intelligent recognition side based on image procossing
Method, comprising: wall images signal is obtained by CCD camera;The gray scale stretching of crack area into [c, d] range, then divide
Section stretching conversion formula:
Wherein, f (x, y) is gray value of the wall body slit image at position (x, y), and transformation range uses [0, Mf] indicate, MgTable
Show the minimal gray grade of wall images, MfIndicate the maximum gray scale of wall images, and the tonal range in wall body slit region is
[a,b];Picture contrast after grey linear transformation is enhanced, and wall body slit target area tonal range is expanded
Exhibition;Wall body slit graphical rule space is constructed, change of scale is carried out to the wall body slit image after stretching, obtains wall body slit figure
As the expression sequence of multiscale space;It indicates that sequence key point is chosen, Local Extremum mainly is carried out to wall body slit image
Detection, by comparing the gray value acquiring size extreme point of pixel and its consecutive points, then again in more adjacent scale domain
The gray value size detection extreme point of pixel;Gradient in its local domain and it are calculated to the key point that above step obtains
Directional spreding, determine a direction by being calculated as each key point, the gradient modulus value m (x, y) and direction θ of key point (x,
Y) solution formula is as follows:
θ (x, y)=tan-1((L (x, y+1)-L (x, y-1))/L (x+1, y)-L (x-1, y)))
The scale of scale space where L represents key point in above formula is realized by the size of adjustment parameter σ to key point institute
The weighting of direction histogram in scaled window, wherein adjustment parameter σ is equivalent to weighting coefficient, value 1,1.2,1.4,
1.6,1.8,2, the pixel direction contribution closer apart from key point is bigger;Histogram weighted value size is indicated with R are as follows:
The direction of key point field gradient is counted, maximum value indicates the principal direction of field gradient, i.e., using principal direction as pass
The direction of key point;Key point local domain inside gradient direction in wall body slit image is calculated, with position, scale, three, direction category
Characteristic point is described in property, finally indicates wall body slit image with feature vector;Utilize weak learning algorithm and wall body slit figure
As training set (X1, Y1), (X2, Y2) ..., (XM, YM), wherein XM ∈ X, X represent wall body slit image feature vector, Y table
Show comprising wall body slit or does not include wall body slit, YM ∈ Y={+1, -1 }, when initialization, by specified point of wall body slit training set
With 1/M, i.e., the weight of each wall body slit training sample is 1/M, then, carries out T iteration with weak learning algorithm, every time repeatedly
In generation, all updates the distribution situation of training set according to training result, and gives bigger weight for the training example of failure to train,
Such trained example will more be paid close attention in next iteration, will obtain T wall body slit after T circulation in this way
Weak Classifier finally utilizes wall body slit strong classifier to wall body slit strong classifier is obtained together by the weighted superposition of update
Intelligent recognition is carried out to wall body slit.
Preferably, the building wall body slit graphical rule space is that wall body slit image passes through scale space shift conversion
For several groups of images, one group of image can cover multi-layer image again;Between one group of image adjacent two layers bottom be by upper layer by every
Point sampling generates, and maintains wall images scale space continuity.
Preferably, described to indicate that the selection of sequence key point includes for crack image by choosing 8 in same scale
Field point is compared, then more adjacent totally 18 consecutive points in two adjacent scales up and down again, when a pixel exists
When being maximum value or minimum value in this layer of scale space and adjacent two layers, then it is assumed that the point is the key point under the scale.
Preferably, the direction scope of key point field gradient is 0-360 degree, and defining every ten degree is a side
To, therefore the direction number of a pixel shares 36.
Preferably, after the acquisition wall body slit video signal step by CCD camera further include: to acquisition
Wall body slit image is filtered, and filtering is adjusted according to the local variance of wall body slit image using adaptive wiener filter
The output of device, local variance is bigger, and the smoothing effect of filter is stronger, and wall is made to restore image f ' (x, y) and original wall figure
As the mean square error e of f (x, y)2=E [(f (x, y)-f ' (x, y))2] minimum.
Preferably, the weak learning algorithm is deep neural network algorithm or K-means clustering algorithm.
Preferably, the strong classifier is the weighted average of multiple Weak Classifiers.
It preferably, further include crack segmentation step after just being identified with strong classifier to wall body slit, for wall body slit figure
The average gray of picture, target crack area indicates that its intensity profile standard deviation is indicated with δ, the normal distribution of standard deviation with u
Probability density function p (z)=N [u, δ2] indicate;It is v, wall body slit that the intensity profile of wall body slit background, which has average value,
The intensity profile standard deviation of background indicates that the normal distribution probability function of standard deviation is q (z)=[v, s with s2], target crack accounts for
The ratio of general image is t, at this time the grey level probability density of entire wall body slit image are as follows:
tp(z)+(1-t)q(z)
It is separated now with threshold value T;It is wall body slit background as z < T, it is on the contrary then be target crack, this
When, wall body slit background is mistakenly considered the probability in target crack are as follows:
Target crack is mistakenly considered the probability of wall background are as follows:
So probability of fault discrimination are as follows:
t[1-P(T)]+(1-t)Q(T)
As threshold value when evaluation θ minimum, that is, differentiate and make its zero;
So
(1-t) q (T)-tp (T)=0
For wall body slit image,For
Know, solves threshold value T.
Compared with prior art, technical solution of the present invention has the advantages that
Solving can not identify and image algorithm calculates complicated time mistake in real time in traditional wall crack identification technology
Length, redundancy are higher, the unreasonable problem of wall body slit image characteristics extraction;The present invention provides a kind of wall based on image procossing
Body crack intelligent identification Method, by constructing to wall body slit graphical rule space, key point is chosen, and greatly reduces picture number
According to redundancy, a kind of wall body slit intelligent identification Method based on image procossing of the invention significantly improves wall body slit figure
As the accuracy rate of identification, wall images computing redundancy degree is reduced, computational efficiency is improved, can accomplish to identify in real time, the party
Method greatly enhances in terms of intelligent control, calculating speed and real-time, improves recognition efficiency.
Detailed description of the invention
Fig. 1 is a kind of wall body slit intelligent identification Method flow chart based on image procossing of the invention;
Fig. 2 is the expression sequence flow figure in acquisition wall body slit Image Multiscale space of the invention;
Fig. 3 is the grey wall body slit image of the background complexity of acquisition applications of the present invention;
Fig. 4 is the grey wall body slit intermediate image generated in key point extraction process of the present invention;
Fig. 5 is the grey wall body slit image of the background complexity after present invention segmentation;
Fig. 6 is the wall body slit image that the background that the present invention acquires is light color;
Fig. 7 is that the background after present invention segmentation is light wall body slit image.
Specific embodiment
It will be appreciated by those skilled in the art that as described in the background art, can not know in real time in traditional wall crack identification technology
Not and the complicated overlong time of image algorithm calculating, redundancy are higher, the unreasonable problem of wall body slit image characteristics extraction, because
This, the present invention provides a kind of wall body slit intelligent identification Method based on image procossing, by empty to wall body slit graphical rule
Between construct, key point choose, greatly reduce image data redundancy, a kind of wall body slit based on image procossing of the invention
Intelligent identification Method significantly improves the accuracy rate of wall body slit image recognition, reduces wall images computing redundancy degree, mentions
High computational efficiency, can accomplish to identify in real time, this method increases in terms of intelligent control, calculating speed and real-time
By force, recognition efficiency is improved.It is understandable to enable above-mentioned purpose of the invention, feature and beneficial effect to become apparent, it ties below
Attached drawing is closed to be described in detail specific embodiments of the present invention.
Embodiment one:
Fig. 1 shows a kind of wall body slit intelligent identification Method flow chart based on image procossing of the application, Yi Zhongji
In the wall body slit intelligent identification Method of image procossing, comprising: obtain wall images signal by CCD camera;Crack area
The gray scale stretching in domain is then segmented stretching conversion formula into [c, d] range:
Wherein, f (x, y) is gray value of the wall body slit image at position (x, y), and transformation range uses [0, Mf] indicate, MgTable
Show the minimal gray grade of wall images, MfIndicate the maximum gray scale of wall images, and the tonal range in wall body slit region is
[a,b];Picture contrast after grey linear transformation is enhanced, and wall body slit target area tonal range is expanded
Exhibition;As shown in Figure 2, extraction feature point constructs wall body slit graphical rule space, carries out ruler to the wall body slit image after stretching
Degree transformation, obtains the expression sequence in wall body slit Image Multiscale space;It indicates that sequence key point is chosen, mainly wall is split
It stitches image and carries out Local Extremum detection, by comparing the gray value acquiring size extreme point of pixel and its consecutive points, so
Afterwards again in more adjacent scale domain pixel gray value size detection extreme point;The key point obtained to above step calculates it
Gradient and its directional spreding in local domain, determine a direction, the gradient of key point by being calculated as each key point
The solution formula of modulus value m (x, y) and direction θ (x, y) are as follows:
θ (x, y)=tan-1((L (x, y+1)-L (x, y-1))/L (x+1, y)-L (x-1, y)))
The scale of scale space where L represents key point in above formula is realized by the size of adjustment parameter σ to key point institute
The weighting of direction histogram in scaled window, σ value are 1,1.2,1.4,1.6,1.8,2, the pixel closer apart from key point
The contribution of point direction is bigger;Histogram weighted value size is indicated with R are as follows:
The direction of key point field gradient is counted, maximum value indicates the principal direction of field gradient, i.e., using principal direction as pass
The direction of key point;Key point local domain inside gradient direction in wall body slit image is calculated, with position, scale, three, direction category
Characteristic point is described in property, finally indicates wall body slit image with feature vector;Utilize weak learning algorithm and wall body slit figure
As training set (X1, Y1), (X2, Y2) ..., (XM, YM), wherein XM ∈ X, X represent wall body slit image feature vector, Y table
Show comprising wall body slit or does not include wall body slit, YM ∈ Y={+1, -1 }, when initialization, by specified point of wall body slit training set
With 1/M, i.e., the weight of each wall body slit training sample is 1/M, then, carries out T iteration with weak learning algorithm, every time repeatedly
In generation, all updates the distribution situation of training set according to training result, and gives bigger weight for the training example of failure to train,
Such trained example will more be paid close attention in next iteration, will obtain T wall body slit after T circulation in this way
Weak Classifier finally utilizes wall body slit strong classifier to wall body slit strong classifier is obtained together by the weighted superposition of update
Intelligent recognition is carried out to wall body slit.
In some embodiments, the building wall body slit graphical rule space is that wall body slit image passes through scale space
Shift conversion is several groups of images, and one group of image can cover multi-layer image again;Bottom is by upper between one group of image adjacent two layers
Layer is sampled by dot interlace and is generated, and wall images scale space continuity is maintained.
In some embodiments, the expression sequence key point selection includes for crack image by choosing same scale
8 interior field points are compared, then more adjacent totally 18 consecutive points in two adjacent scales up and down again, when a picture
When vegetarian refreshments is maximum value or minimum value in this layer of scale space and adjacent two layers, then it is assumed that the point is the key that under the scale
Point.
Embodiment two:
A kind of wall body slit intelligent identification Method based on image procossing, comprising: wall images are obtained by CCD camera
Vision signal;It is made of a series of picture frame, setting given pace is acquired, and each frame image signal is not overlapped, and is not lost
Pierced wall body image section;Wall body slit graphical rule space is constructed, change of scale is carried out to the wall body slit image after stretching, is obtained
Take the expression sequence in wall body slit Image Multiscale space;Indicate sequence key point choose, mainly to wall body slit image into
Then the detection of row Local Extremum is compared again by comparing the gray value acquiring size extreme point of pixel and its consecutive points
The gray value size detection extreme point of pixel in adjacent scale domain;The key point obtained to above step calculates its local domain
Interior gradient and its directional spreding determine a direction by being calculated as each key point, the gradient modulus value m of key point (x,
Y) and the solution formula of direction θ (x, y) is as follows:
θ (x, y)=tan-1((L (x, y+1)-L (x, y-1))/L (x+1, y)-L (x-1, y)))
The scale of scale space where L represents key point in above formula is realized by the size of adjustment parameter σ to key point institute
The weighting of direction histogram in scaled window, σ value are 1,1.2,1.4,1.6,1.8,2, specially when distance is greater than 500
When pixel value, σ value is 1, and when distance is greater than 400 pixel values less than 500 pixel values, σ value is 1.2;When distance is big
When 300 pixel values are less than 400 pixel values, σ value is 1.4, and so on, the pixel direction closer apart from key point
It contributes bigger;Histogram weighted value size is indicated with R are as follows:
The direction of key point field gradient is counted, maximum value indicates the principal direction of field gradient, i.e., using principal direction as pass
The direction of key point;Key point local domain inside gradient direction in wall body slit image is calculated, with position, scale, three, direction category
Characteristic point is described in property, finally indicates wall body slit image with feature vector;Utilize weak learning algorithm and wall body slit figure
As training set (X1, Y1), (X2, Y2) ..., (XM, YM), wherein XM ∈ X, X represent wall body slit image feature vector, Y table
Show comprising wall body slit or does not include wall body slit, YM ∈ Y={+1, -1 }, when initialization, by specified point of wall body slit training set
With 1/M, i.e., the weight of each wall body slit training sample is 1/M, then, carries out T iteration with weak learning algorithm, every time repeatedly
In generation, all updates the distribution situation of training set according to training result, and gives bigger weight for the training example of failure to train,
Such trained example will more be paid close attention in next iteration, will obtain T wall body slit after T circulation in this way
Weak Classifier finally utilizes wall body slit strong classifier to wall body slit strong classifier is obtained together by the weighted superposition of update
Intelligent recognition is carried out to wall body slit.
In some embodiments, the direction scope of key point field gradient is 0-360 degree, and defines every ten degree and be
One direction, therefore the direction number of a pixel shares 36.
In some embodiments, after the acquisition wall body slit video signal step by CCD camera further include:
The wall body slit image of acquisition is filtered, using adaptive wiener filter according to the local variance of wall body slit image come
The output of filter is adjusted, local variance is bigger, and the smoothing effect of filter is stronger, and wall is made to restore image f ' (x, y) and former
The mean square error e of beginning wall images f (x, y)2=E [(f (x, y)-f ' (x, y))2] minimum.
Embodiment three:
A kind of wall body slit intelligent identification Method based on image procossing, comprising: wall images are obtained by CCD camera
Vision signal;It is made of a series of picture frame, setting given pace is acquired, and each frame image signal is not overlapped, and is not lost
Pierced wall body image section;Key point is chosen, and mainly Local Extremum detection is carried out to wall body slit image, by comparing pixel
Point and it consecutive points gray value acquiring size extreme point, then again in more adjacent scale domain pixel gray value size
Detect extreme point;The key point obtained to above step calculates gradient and its directional spreding in its local domain, passes through meter
Calculating is that each key point determines a direction, and the solution formula of the gradient modulus value m (x, y) and direction θ (x, y) of key point are as follows:
θ (x, y)=tan-1((L (x, y+1)-L (x, y-1))/L (x+1, y)-L (x-1, y)))
The scale of scale space where L represents key point in above formula is realized by the size of adjustment parameter σ to key point institute
The weighting of direction histogram in scaled window, the pixel direction contribution closer apart from key point are bigger;Count key point neck
The direction of domain gradient, maximum value indicates the principal direction of field gradient, i.e., using principal direction as the direction of key point;Wall is calculated to split
Key point local domain inside gradient direction in image is stitched, characteristic point is described with position, scale, three, direction attribute, most
Wall body slit image is indicated with feature vector afterwards;Using weak learning algorithm and wall body slit training set of images (X1, Y1), (X2,
Y2) ..., (XM, YM), wherein XM ∈ X, X represent wall body slit image feature vector, and Y is indicated comprising wall body slit or do not wrapped
Include wall body slit, YM ∈ Y={+1, -1 }, when initialization, by wall body slit training set assignment of allocation 1/M, i.e., each wall body slit
The weight of training sample is all 1/M, then, carries out T iteration with weak learning algorithm, each iteration is all according to training result come more
The distribution situation of new training set, and bigger weight is given for the training example of failure to train, such trained example will be under
An iteration is more paid close attention to, and will obtain T wall body slit Weak Classifier after T circulation in this way, final by update
Weighted superposition to obtaining wall body slit strong classifier together, intelligent knowledge is carried out to wall body slit using wall body slit strong classifier
Not.
In some embodiments, the weak learning algorithm is deep neural network algorithm or K-means clustering algorithm.
In some embodiments, the strong classifier is the weighted average of multiple Weak Classifiers.
In some embodiments, as shown in fig. 3 to 7, image segmentation process, after just being identified with strong classifier to wall body slit
It further include crack segmentation step, for wall body slit image, the average gray of target crack area is indicated with u, its gray scale
Distribution standard deviation indicates with δ, normpdf p (z)=N [u, δ of standard deviation2] indicate;Wall body slit back
It is v that the intensity profile of scape, which has average value, and the intensity profile standard deviation of wall body slit background is indicated with s, the normal state point of standard deviation
Cloth probability function is q (z)=[v, s2], the ratio that target crack accounts for general image is t, at this time the ash of entire wall body slit image
Spending probability density is
tp(z)+(1-t)q(z)
It is separated now with threshold value T;It is wall body slit background as z < T, it is on the contrary then be target crack, at this point, wall is split
Seam background is mistakenly considered the probability in target crack are as follows:
Target crack is mistakenly considered the probability of wall background are as follows:
So probability of fault discrimination are as follows:
t[1-P(T)]+(1-t)Q(T)
As threshold value when evaluation θ minimum, that is, differentiate and make its zero;
So
(1-t) q (T)-tp (T)=0
For wall body slit image,
Be it is known, solve threshold value T.
In some embodiments, wall body slit image segmentation just is carried out with OSTU algorithm.
The present invention provides a kind of wall body slit intelligent identification Method based on image procossing, by wall body slit image ruler
Space building is spent, key point is chosen, and image data redundancy, a kind of wall based on image procossing of the invention are greatly reduced
Crack intelligent identification Method significantly improves the accuracy rate of wall body slit image recognition, reduces wall images computing redundancy
Degree, improves computational efficiency, can accomplish to identify in real time, and this method is big in terms of intelligent control, calculating speed and real-time
Big enhancing, improves recognition efficiency.
Although present disclosure is as above, present invention is not limited to this.Anyone skilled in the art are not departing from this
It in the spirit and scope of invention, can make various changes or modifications, therefore protection scope of the present invention should be with claim institute
Subject to the range of restriction.
Claims (8)
1. a kind of wall body slit intelligent identification Method based on image procossing, which is characterized in that obtain wall by CCD camera
Picture signal;The gray scale stretching in wall body slit region into [c, d] range, then stretching conversion formula is segmented:
Wherein, f (x, y) is gray value of the wall body slit image at position (x, y), and transformation range uses [0, Mf] indicate, MgIndicate wall
The minimal gray grade of body image, MfIndicate wall images maximum gray scale, and the tonal range in wall body slit region be [a,
b];
Wall body slit graphical rule space is constructed, change of scale is carried out to the wall body slit image after stretching, obtains wall body slit
The expression sequence in Image Multiscale space;
Key point is chosen, and Local Extremum detection is carried out to wall body slit image, by comparing pixel and its consecutive points
Gray value acquiring size extreme point, then, then in more adjacent scale domain pixel gray value size detection extreme point;
The key point obtained to above step calculates gradient and directional spreding in its local domain, by being calculated as each key
Point determines a direction, and the solution formula of the gradient modulus value m (x, y) and direction θ (x, y) of key point are as follows:
θ (x, y)=tan-1((L (x, y+1)-L (x, y-1))/L (x+1, y)-L (x-1, y)))
The scale of scale space where L represents key point in above formula is realized by the size of adjustment parameter σ to ruler where key point
The weighting of direction histogram in window is spent, bigger, histogram weighted value size is contributed in the pixel direction closer apart from key point
It is indicated with R are as follows:
The direction of key point field gradient is counted, maximum value indicates the principal direction of field gradient, i.e., using principal direction as key point
Direction;Key point local domain inside gradient direction in wall body slit image is calculated, with position, scale, three, direction attribute pair
Key point is described, and finally indicates wall body slit image with feature vector;
Wall body slit training set of images is trained using weak learning algorithm, (X1, Y1), (X2, Y2) ..., (XM, YM),
In, XM ∈ X, X represent wall body slit image feature vector, and Y is indicated comprising wall body slit or do not included wall body slit, YM ∈ Y=
{+1, -1 }, when initialization, the weight by wall body slit training set assignment of allocation 1/M, i.e., each wall body slit training sample is
Then 1/M carries out T iteration with weak learning algorithm, will obtain T wall body slit Weak Classifier after T circulation, finally
Weighted superposition by update carries out wall body slit using wall body slit strong classifier to wall body slit strong classifier is obtained together
Intelligent recognition.
2. a kind of wall body slit intelligent identification Method based on image procossing according to claim 1, which is characterized in that institute
It by scale space shift conversion is several groups of images, a group picture that state building wall body slit graphical rule space, which be wall body slit image,
As covering multi-layer image;Bottom is to be sampled to generate by dot interlace by upper layer between one group of image adjacent two layers, maintains wall figure
As scale space continuity.
3. a kind of wall body slit intelligent identification Method based on image procossing according to claim 1, which is characterized in that institute
It includes that crack image is compared by choosing 8 field points in same scale that stating, which indicates that sequence key point is chosen, so
More adjacent totally 18 consecutive points in two adjacent scales up and down again afterwards, when a pixel is in this layer of scale space and phase
When being maximum value or minimum value in two layers adjacent, then it is assumed that the point is the key point under the scale.
4. a kind of wall body slit intelligent identification Method based on image procossing according to claim 3, which is characterized in that institute
The direction scope for stating key point field gradient is 0-360 degree, and defining every 10 degree is a direction, i.e., the side of one pixel
36 are shared to number.
5. a kind of wall body slit intelligent identification Method based on image procossing according to claim 1, which is characterized in that institute
It states after obtaining wall body slit video signal step by CCD camera further include: the wall body slit image of acquisition is carried out
Filtering, the output of filter, part side are adjusted using adaptive wiener filter according to the local variance of wall body slit image
Difference is bigger, and the smoothing effect of filter is stronger, and wall is made to restore the mean square error of image f ' (x, y) and original wall images f (x, y)
Poor e2=E [(f (x, y)-f ' (x, y))2] minimum.
6. a kind of wall body slit intelligent identification Method based on image procossing according to claim 1, which is characterized in that institute
Stating weak learning algorithm is deep neural network algorithm or K-means clustering algorithm.
7. a kind of wall body slit intelligent identification Method based on image procossing according to claim 1, which is characterized in that institute
State the weighted average that strong classifier is multiple Weak Classifiers.
8. a kind of wall body slit intelligent identification Method based on image procossing according to claim 1, which is characterized in that right
Wall body slit further includes crack segmentation step after just being identified with strong classifier, for wall body slit image, target crack area
Average gray indicates that its intensity profile standard deviation is indicated with δ with u, the normpdf p of standard deviation
(z)=N [u, δ2] indicate;It is v, the intensity profile mark of wall body slit background that the intensity profile of wall body slit background, which has average value,
Quasi- difference indicates that the normal distribution probability function of standard deviation is q (z)=[v, s with s2], the ratio that target crack accounts for general image is
T, the grey level probability density of entire wall body slit image is tp (z)+(1-t) q (z) at this time
It is separated now with threshold value T;It is wall body slit background as z < T, it is on the contrary then be target crack, at this point, wall body slit is carried on the back
Scape is mistakenly considered the probability in target crack are as follows:
Target crack is mistakenly considered the probability of wall background are as follows:
So probability of fault discrimination are as follows:
t[1-P(T)]+(1-t)Q(T)
As threshold value when evaluation θ minimum, that is, differentiate and make its zero;
So
(1-t) q (T)-tp (T)=0
For wall body slit image,
Be it is known, solve threshold value T.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910443398.9A CN110222609A (en) | 2019-05-24 | 2019-05-24 | A kind of wall body slit intelligent identification Method based on image procossing |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910443398.9A CN110222609A (en) | 2019-05-24 | 2019-05-24 | A kind of wall body slit intelligent identification Method based on image procossing |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110222609A true CN110222609A (en) | 2019-09-10 |
Family
ID=67818390
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910443398.9A Pending CN110222609A (en) | 2019-05-24 | 2019-05-24 | A kind of wall body slit intelligent identification Method based on image procossing |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110222609A (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110929565A (en) * | 2019-10-15 | 2020-03-27 | 平安科技(深圳)有限公司 | Risk monitoring method and device based on machine learning, storage medium and electronic equipment |
CN113066248A (en) * | 2021-03-25 | 2021-07-02 | 武汉畅途网络科技有限公司 | Intelligent community construction security monitoring intelligent management system based on video image processing |
CN116664960A (en) * | 2023-07-28 | 2023-08-29 | 深圳弗瑞奇科技有限公司 | Cat stool identification method based on image processing |
CN116756835A (en) * | 2023-08-15 | 2023-09-15 | 深圳市前海荣群铝业科技有限公司 | Template combination design method, device, equipment and storage medium |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20110106459A1 (en) * | 2009-10-29 | 2011-05-05 | Northrop Grumman Corporation | In-situ optical crack measurement using a dot pattern |
CN104005325A (en) * | 2014-06-17 | 2014-08-27 | 武汉武大卓越科技有限责任公司 | Pavement crack detecting device and method based on depth and gray level images |
CN106327494A (en) * | 2016-08-24 | 2017-01-11 | 张家利 | Pavement crack image automatic detection method |
CN107392911A (en) * | 2017-07-15 | 2017-11-24 | 江西理工大学 | The shaddock image partition method of application adaptability gaussian sum sonar surveillance system rope |
CN108985363A (en) * | 2018-07-03 | 2018-12-11 | 长安大学 | A kind of cracks in reinforced concrete bridge classifying identification method based on RBPNN |
CN109580657A (en) * | 2019-01-23 | 2019-04-05 | 郑州工程技术学院 | A kind of crack detection method in bridge quality testing |
CN109615616A (en) * | 2018-11-27 | 2019-04-12 | 北京联合大学 | A kind of crack identification method and system based on ABC-PCNN |
CN109754368A (en) * | 2019-01-23 | 2019-05-14 | 郑州工程技术学院 | A kind of crack joining method in bridge quality testing |
-
2019
- 2019-05-24 CN CN201910443398.9A patent/CN110222609A/en active Pending
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20110106459A1 (en) * | 2009-10-29 | 2011-05-05 | Northrop Grumman Corporation | In-situ optical crack measurement using a dot pattern |
CN104005325A (en) * | 2014-06-17 | 2014-08-27 | 武汉武大卓越科技有限责任公司 | Pavement crack detecting device and method based on depth and gray level images |
CN106327494A (en) * | 2016-08-24 | 2017-01-11 | 张家利 | Pavement crack image automatic detection method |
CN107392911A (en) * | 2017-07-15 | 2017-11-24 | 江西理工大学 | The shaddock image partition method of application adaptability gaussian sum sonar surveillance system rope |
CN108985363A (en) * | 2018-07-03 | 2018-12-11 | 长安大学 | A kind of cracks in reinforced concrete bridge classifying identification method based on RBPNN |
CN109615616A (en) * | 2018-11-27 | 2019-04-12 | 北京联合大学 | A kind of crack identification method and system based on ABC-PCNN |
CN109580657A (en) * | 2019-01-23 | 2019-04-05 | 郑州工程技术学院 | A kind of crack detection method in bridge quality testing |
CN109754368A (en) * | 2019-01-23 | 2019-05-14 | 郑州工程技术学院 | A kind of crack joining method in bridge quality testing |
Non-Patent Citations (2)
Title |
---|
孙朝云等: "沥青混合料图像阈值分割算法的实现", 《长安大学学报(自然科学版)》 * |
王琳等: "单板表面裂缝图像处理方法研究", 《山东工业技术》 * |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110929565A (en) * | 2019-10-15 | 2020-03-27 | 平安科技(深圳)有限公司 | Risk monitoring method and device based on machine learning, storage medium and electronic equipment |
CN110929565B (en) * | 2019-10-15 | 2023-12-22 | 平安科技(深圳)有限公司 | Risk monitoring method and device based on machine learning, storage medium and electronic equipment |
CN113066248A (en) * | 2021-03-25 | 2021-07-02 | 武汉畅途网络科技有限公司 | Intelligent community construction security monitoring intelligent management system based on video image processing |
CN113066248B (en) * | 2021-03-25 | 2021-11-05 | 福建省凯特科技有限公司 | Intelligent community construction security monitoring intelligent management system based on video image processing |
CN116664960A (en) * | 2023-07-28 | 2023-08-29 | 深圳弗瑞奇科技有限公司 | Cat stool identification method based on image processing |
CN116756835A (en) * | 2023-08-15 | 2023-09-15 | 深圳市前海荣群铝业科技有限公司 | Template combination design method, device, equipment and storage medium |
CN116756835B (en) * | 2023-08-15 | 2023-12-01 | 深圳市前海荣群铝业科技有限公司 | Template combination design method, device, equipment and storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110348319B (en) | Face anti-counterfeiting method based on face depth information and edge image fusion | |
CN110222609A (en) | A kind of wall body slit intelligent identification Method based on image procossing | |
CN104008370B (en) | A kind of video face identification method | |
CN109522857B (en) | People number estimation method based on generation type confrontation network model | |
CN105518709B (en) | The method, system and computer program product of face for identification | |
CN107016357B (en) | Video pedestrian detection method based on time domain convolutional neural network | |
CN108520216B (en) | Gait image-based identity recognition method | |
CN105182350B (en) | A kind of multibeam sonar object detection method of application signature tracking | |
CN109241913A (en) | In conjunction with the ship detection method and system of conspicuousness detection and deep learning | |
CN110119728A (en) | Remote sensing images cloud detection method of optic based on Multiscale Fusion semantic segmentation network | |
CN108389220B (en) | Remote sensing video image motion target real-time intelligent cognitive method and its device | |
CN109508710A (en) | Based on the unmanned vehicle night-environment cognitive method for improving YOLOv3 network | |
CN110175504A (en) | A kind of target detection and alignment schemes based on multitask concatenated convolutional network | |
CN106980852B (en) | Based on Corner Detection and the medicine identifying system matched and its recognition methods | |
CN109800629A (en) | A kind of Remote Sensing Target detection method based on convolutional neural networks | |
CN105160310A (en) | 3D (three-dimensional) convolutional neural network based human body behavior recognition method | |
CN108734171A (en) | A kind of SAR remote sensing image ocean floating raft recognition methods of depth collaboration sparse coding network | |
JP2015191334A (en) | Information processor and information processing method | |
CN106909902A (en) | A kind of remote sensing target detection method based on the notable model of improved stratification | |
CN110060273B (en) | Remote sensing image landslide mapping method based on deep neural network | |
CN106611423B (en) | SAR image segmentation method based on ridge ripple filter and deconvolution structural model | |
CN104616006B (en) | A kind of beard method for detecting human face towards monitor video | |
CN106169073A (en) | A kind of expression recognition method and system | |
CN109506628A (en) | Object distance measuring method under a kind of truck environment based on deep learning | |
CN105138983B (en) | The pedestrian detection method divided based on weighting block model and selective search |
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 |
Application publication date: 20190910 |
|
RJ01 | Rejection of invention patent application after publication |