CN109255379A - A kind of goat's horn bow area positioning method combined based on fusion feature and SVM - Google Patents
A kind of goat's horn bow area positioning method combined based on fusion feature and SVM Download PDFInfo
- Publication number
- CN109255379A CN109255379A CN201811009549.1A CN201811009549A CN109255379A CN 109255379 A CN109255379 A CN 109255379A CN 201811009549 A CN201811009549 A CN 201811009549A CN 109255379 A CN109255379 A CN 109255379A
- Authority
- CN
- China
- Prior art keywords
- feature
- lbp
- pixel
- hog
- goat
- 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
-
- 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
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/50—Extraction of image or video features by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/467—Encoded features or binary features, e.g. local binary patterns [LBP]
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of goat's horns combined based on fusion feature and SVM to bend area positioning method.Method are as follows: extract the LBP feature and HOG feature of training sample image first: local binary pattern histogram is constituted by the relationship of the pixel value and its neighborhood that calculate the training sample collected to extract LBP feature；The training sample of collection is split, the unit of several pixels is divided into, direction histogram then is sought to the edge or gradient of pixel each in unit, integrates all direction histograms, constitutes HOG profiler；Then Feature Fusion is used, the LBP feature and HOG feature extracted are merged；It is trained followed by SVM classifier, obtains training parameter；Target image is scanned finally by multi-scale sliding window mouth, goat's horn is extracted and bends region.The present invention has the advantages that locating accuracy is high, easy to implement.
Description
Technical field
The invention belongs to traffic safety field of engineering technology, especially a kind of goat's horn combined based on fusion feature and SVM
Bend area positioning method.
Background technique
Goat's horn bow is located at slide plate two sides, for improving current carrying quality, if goat's horn lacks, will cause pantograph unbalance stress
It is even, so occur slide plate fracture etc. major accidents, goat's horn bow working condition detection to the replacement of pantograph-contact net system,
Maintenance and repair is of great significance.
Foreign countries are more early to the research starting of Contact Line Detection System, by many years exploration practice, have more complete
Fault detection system, troubleshooting system and assessment of fault system.Foreign countries to the research emphasis of Contact Line Detection System respectively not
Identical, France has extensively studied the relevant technologies of bow net dynamic elasticity detection system, and Germany is to detection bow net contact pressure initiation
More perfect detection architecture, Japanese bow net is offline, contact line Abrasion detecting technology is more mature.Is bent to goat's horn by defect for the country
Research it is less, mainly based on artificial detection method.Southwest hands over big Zhu Xiaoheng for the pantograph image of acquisition, using figure
Bend region to goat's horn as two-value method to compare and analyze, it being capable of preliminary judgement defect state.Mo Shengyang team is examined based on 3D vision
Survey technology has carried out the research of abrasion of pantograph pan and has constructed three-dimensional machine using high speed three-dimensional camera, laser emitter
Vision system.On the whole, existing goat's horn bow area positioning method realizes complicated that accuracy rate is not high, needs to seek a kind of letter
Single effective goat's horn bends area positioning method.
Summary of the invention
The purpose of the present invention is to provide a kind of goat's horns combined based on fusion feature and SVM to bend area positioning method, uses
Simply accurately to be positioned to goat's horn bow region.
The technical solution for realizing the aim of the invention is as follows: a kind of goat's horn bow region combined based on fusion feature and SVM
Localization method, comprising the following steps:
Step 1, training sample are collected: training sample includes positive sample and negative sample, and the picture of positive sample is that goat's horn bends just
Normal state, the picture of negative sample are the unknown objects that goat's horn bow miss status is either cut at random；
Step 2, LBP feature extraction: the pixel value and its neighborhood of each pixel in the training sample that step 1 is collected are calculated
Relationship, constitute local binary pattern histogram, to extract LBP feature；
Step 3, HOG feature extraction: the training sample that step 1 is collected is divided into the unit comprising multiple pixels, so
Direction histogram is sought to the edge or gradient of each pixel in each unit afterwards, integrates all direction histograms, structure
At HOG profiler；
Step 4, Fusion Features: by the LBP feature that step 2 is extracted and the HOG feature that step 3 is extracted, serial fusion is used
Technology is merged, and LBP-HOG fusion feature direction histogram is generated；
Step 5, SVM classifier training: the LBP-HOG fusion feature that step 4 obtains is sent in SVM classifier and is instructed
Practice, obtains training parameter；
Step 6, multi-scale sliding window mouth Scan orientation: target image is scanned using multi-scale sliding window mouth, by step 5
In trained SVM classifier predicted, realize goat's horn bow region positioning.
Further, calculating step 1 described in step 2 collect training sample in each pixel pixel value and its
The relationship of neighborhood constitutes local binary pattern histogram, specific as follows to extract LBP feature:
Step 2.1, the image for collecting step 1 are divided into 4 × 4 regions；
Step 2.2, for each pixel in region, select threshold value of its gray value as LBP operator, by 3 × 3 neighbour
The gray value of other pixels in domain is compared with the threshold value of LBP operator, and the gray value pixel bigger than threshold value is set to 1,
The gray value pixel smaller than threshold value is set to 0；Then by 3 × 3 neighborhoods 8 points generate 8 bits, by two into
Number processed is converted to decimal number to get the LBP value of each pixel is arrived；
Step 2.3, the histogram for calculating each region, i.e., the frequency that each LBP value occurs, obtain 16 LBP histograms；
16 histograms are described step 2.4 with unified histogram；Characteristic pattern size is 80 × 60 pixels, will
It is divided into 4 × 4 pieces, and the vector of generation is 4 × 4 × 256=4096 dimension.
Further, the training sample collected step 1 described in step 3 is divided into the list comprising several pixels
Then member seeks direction histogram to the edge or gradient of each pixel in each unit, integrates all direction Histograms
Figure constitutes HOG profiler, specific as follows:
Step 3.1, the gamma for editing image, formula are as follows:
I (x, y)=I (x, y)gamma (1)
Wherein I (x, y) indicates the pixel value of each pixel, gamma=1/2；
Step 3.2, the gradient direction value for calculating each pixel transverse and longitudinal coordinate in image, and then obtain the ladder of the pixel
Degree；
Set GxFor x direction gradient, GyFor y direction gradient, [- 1,0,1] is center template operator, the ash of pixel (x, y)
Angle value is H (x, y), then the gradient formula of pixel (x, y) are as follows:
Gx(x, y)=H (x+1, y)-H (x-1, y) (2)
Gy(x, y)=H (x, y+1)-H (x, y-1) (3)
The gray scale size that G (x, y) is pixel (x, y) is set, θ is the gray scale direction of pixel (x, y), then:
Entire image is divided into cell by step 3.3, and each cell size is 10 × 10 pixels；Histogram of gradients
Select 9 sections, count the gradient weights of each pixel, draw the histogram of gradients of each cell, obtain 9 dimensional features to
The histogram of gradients of amount；By every 2 × 2 cell spans at a block, the feature vector of each cell is normalized, then each
The feature vector dimension of block is 2 × 2 × 9=36；
Step 3.4 merges the HOG feature of entire image, obtains entire image HOG feature；Due to each unit
The size of lattice is 10 × 10, and a block includes 2 × 2 cells, thus the dimension of the HOG feature of entire image be 9 × 4 ×
(8-1) × (6-1)=1260.
Further, the LBP feature that step 2 is extracted and the HOG feature that step 3 is extracted are used into string described in step 4
Row integration technology is merged, and LBP-HOG fusion feature direction histogram is generated, specific as follows:
LBP feature space is set as A, HOG feature space is B, and there are two feature vector α ∈ A and β ∈ B, using serial
Fusion, obtains:
Wherein, γ is fused feature vector, and θ is weight coefficient；
Set fused matrix dimension as F, then:
F=dim (A)+dim (B) (7)
Goat's horn bow feature is merged using serial integration technology, generates LBP-HOG characteristic direction histogram；By LBP
The dimension 4096 of feature vector, the dimension 1260 of HOG feature vector, can obtain fused LBP-HOG feature vector dimension is
5356。
Further, the LBP-HOG fusion feature that step 4 obtains is sent in SVM classifier described in step 5 and is carried out
Training obtains training parameter, specific as follows:
The quantity of training sample is set as 900,900 LBP-HOG fusion features are sent into SVM classifier and are instructed
Practice, during positive training, 900 samples are divided into two groups respectively, every group includes 300 positive samples and 600 negative samples,
Training stage is as follows:
After first group of 300 positive samples and 600 negative samples are put into SVM training, then with second group of sample set
SVM is tested, test result is obtained, the data that positive sample is mistaken in test result are put into first group of negative sample
In；The data that positive sample is mistaken in test result are put into first group by training and the test process for re-starting SVM classifier
Negative sample in, repeat the above process, obtain a final SVM classifier.
Further, the mouth of use multi-scale sliding window described in step 6 scans target image, passes through training in step 5
Good SVM classifier is predicted, realizes the positioning in goat's horn bow region, specific as follows:
Binaryzation is carried out to target image, obtains 0 and 255 two kind of gray value, since background colour is black, goat's horn bends color
It is white, so the target of detection is the region that gray value is 255；Selecting size is the window of k × l, the picture of image in window
Plain gray value is h (xi, yj), then the accounting that gray value is 255 are as follows:
In formula, η is the threshold value judged in advance, and value is η=0.5；Window region of the accounting greater than 0.5 for being 255 to gray value
Domain carries out LBP-HOG feature extraction, is then predicted by trained SVM classifier, obtains candidate rectangle window；Using
Multi-scale sliding window mouth is detected, and when detecting multiple targets, merges candidate window, finally extracts goat's horn bow region.
Compared with prior art, the present invention its remarkable advantage are as follows: (1) be sent to the fusion feature of LBP feature and HOG feature
SVM classifier is trained, so that the accuracy rate of positioning greatly improves；(2) high, easy to implement with locating accuracy excellent
Point.
Detailed description of the invention
Fig. 1 is the flow diagram based on fusion feature and SVM the goat's horn bow area positioning method combined in the present invention.
Fig. 2 is the positive sample exemplary diagram in the present invention.
Fig. 3 is the negative sample exemplary diagram in the present invention.
Fig. 4 is LBP-HOG characteristic pattern in the present invention.
Fig. 5 is that candidate window merges schematic diagram in the present invention.
Specific embodiment
In conjunction with Fig. 1, area positioning method, packet are bent based on fusion feature and the SVM goat's horn combined proposed in the present invention
Include following steps:
Step 1, training sample are collected: training sample includes positive sample and negative sample, and the picture of positive sample is that goat's horn bends just
Normal state, as shown in Fig. 2, the picture of negative sample is the unknown object that goat's horn bow miss status is either cut at random, such as Fig. 3 institute
Show.
Step 2, LBP feature extraction: the pixel value and its neighborhood of each pixel in the training sample that step 1 is collected are calculated
Relationship, constitute local binary pattern histogram, to extract LBP feature.It is specific as follows:
Step 2.1, the image for collecting step 1 are divided into 4 × 4 regions；
Step 2.2, for each pixel in region, select threshold value of its gray value as LBP operator, by 3 × 3 neighbour
The gray value of other pixels in domain is compared with threshold value, and the gray value pixel bigger than threshold value is set to 1, and gray value compares threshold
It is worth small pixel and is set to 0,8 points in 3 × 3 neighborhoods is then generated into 8 bits, binary number is converted to
Decimal number is to get the LBP value for arriving each pixel；
Step 2.3, the histogram for calculating each region, i.e., the frequency that each LBP value occurs, obtain 16 LBP histograms；
16 histograms are described step 2.4 with unified histogram.
Further, the characteristic pattern size that this patent uses is 80 × 60, is divided into 4 × 4 pieces, the vector of generation is 4 × 4 ×
256=4096 dimension.
Step 3, HOG feature extraction: the training sample that step 1 is collected is divided into the unit comprising several pixels, so
Direction histogram is sought to the edge or gradient of each pixel in each unit afterwards, integrates all direction histograms, structure
At HOG profiler.It is specific as follows:
Step 3.1, the color space standards in order to realize input picture edit the gamma of image, reduce illumination
Change the influence to image space, preferably enhancing picture contrast.It is as follows that Gamma compresses formula:
I (x, y)=I (x, y)gamma (1)
Wherein I (x, y) indicates the pixel value of each pixel, gamma=1/2；
Step 3.2, the gradient direction value for calculating each pixel transverse and longitudinal coordinate in image, and then obtain the ladder of the pixel
Degree；
Set GxFor x direction gradient, GyFor y direction gradient, [- 1,0,1] is center template operator, the ash of pixel (x, y)
Angle value is H (x, y), then the gradient formula of pixel (x, y) are as follows:
Gx(x, y)=H (x+1, y)-H (x-1, y) (2)
Gy(x, y)=H (x, y+1)-H (x, y-1) (3)
The gray scale size that G (x, y) is pixel (x, y) is set, θ is the gray scale direction of pixel (x, y), then:
Entire image is divided into cell by step 3.3, and each cell size is 10 × 10 pixels；Histogram of gradients
Select 9 sections, count the gradient weights of each pixel, draw the histogram of gradients of each cell, obtain 9 dimensional features to
The histogram of gradients of amount；By every 2 × 2 cell spans at a block, the feature vector of each cell is normalized, then each
The feature vector dimension of block is 2 × 2 × 9=36；
Step 3.4 merges the HOG feature of entire image, obtains entire image HOG feature；
Further, the cell size that this patent is selected is 10 × 10, and a block includes 2 × 2 cells, then block
It merges, the dimension of the HOG feature of final image is 9 × 4 × (8-1) × (6-1)=1260.
Step 4, Fusion Features: by the LBP feature that step 2 is extracted and the HOG feature that step 3 is extracted, serial fusion is used
Technology is merged, and LBP-HOG fusion feature direction histogram is generated.It is specific as follows:
LBP feature space is set as A, HOG feature space is B, and there are two feature vector α ∈ A and β ∈ B, using serial
Fusion, obtains:
Wherein, γ is fused feature vector, and θ is weight coefficient；
Set fused matrix dimension as F, then:
F=dim (A)+dim (B) (7)
Goat's horn bow feature is merged using serial integration technology, the LBP-HOG characteristic direction histogram process of generation
As shown in Figure 4.
Further, the size of camera acquisition is 80 × 60 pictures, selects the cell of 10 × 10 sizes, cell span
It is 2 × 2 block at size, the dimension of LBP feature vector is 4 × 4 × 256=4096, the dimension of HOG feature vector is 9 × 4 ×
7 × 5=1260, can obtain fused LBP-HOG feature vector dimension is 5356 dimensions.
Step 5, SVM classifier training: the LBP-HOG fusion feature that step 4 obtains is sent in SVM classifier, is carried out
Training obtains training parameter.It is specific as follows:
The quantity of training sample is set as 900,900 LBP-HOG fusion features are sent into SVM classifiers and are trained,
During positive training, 900 samples are divided into two groups respectively, every group includes 300 positive samples and 600 negative samples, instruction
It is as follows to practice the stage:
After first group of 300 positive samples and 600 negative samples are put into SVM training, then with second group of sample set
SVM is tested, test result is obtained, the data that positive sample is mistaken in test result are put into first group of negative sample
In；The data that positive sample is mistaken in test result are put into first group by training and the test process for re-starting SVM classifier
Negative sample in, repeat the above process, finally obtain an accurate SVM classifier.
Step 6, multi-scale sliding window mouth Scan orientation: target image is scanned using multi-scale sliding window mouth, by step 5
In trained SVM classifier predicted, realize goat's horn bow region positioning.It is specific as follows:
To reduce extra calculating, binaryzation is carried out to target image, 0 and 255 two kind of gray value are obtained, since background colour is
Black, goat's horn bow color is white, so the target of detection is the region that gray value is 255；Selecting size is the window of k × l,
The grey scale pixel value of image in window is h (xi, yj), then the accounting that gray value is 255 are as follows:
In formula, η is the threshold value judged in advance, and value is η=0.5；Window region of the accounting greater than 0.5 for being 255 to gray value
Domain carries out LBP-HOG feature extraction, is then predicted by trained SVM classifier, obtains candidate rectangle window；Using
Multi-scale sliding window mouth is detected, and when detecting multiple targets, merges candidate window, as shown in figure 5, RED sector is to close
Window after and finally extracts goat's horn bow region.
Embodiment 1
The present embodiment is that the goat's horn combined based on fusion feature and SVM bends area positioning method.
1 HOG feature recognition result table of table
In conjunction with table 1, it can be found that the step sizes of sliding window movement, cell during HOG feature extraction
(cell) size, region unit (block) size, can have an impact detection accuracy, as a result after many experiments, find 2 × 2 sizes
Region unit (block), the cell (cell) of 10 × 10 sizes, moving step length be Width/4 when, HOG feature recognition effect
Relatively good, recognition result is as shown in table 1, and wherein false detection rate is picture number/picture sum of detection mistake, and omission factor is missing inspection
Picture number/picture sum.
2 different characteristic extracting method testing result contrast table of table
In conjunction with table 2,5660 pictures are acquired, use LBP feature extracting method, HOG feature extracting method and the present invention respectively
The fixation and recognition of goat's horn bow, 2 institute of recognition result table are carried out based on the goat's horn bow area positioning method that fusion feature and SVM are combined
Show.As can be seen that it is special that the present invention is based on fusions compared with two kinds of algorithms of LBP feature extracting method and HOG feature extracting method
The goat's horn bow area positioning method accuracy rate that the SVM that seeks peace is combined is higher.
Claims (6)
1. a kind of goat's horn combined based on fusion feature and SVM bends area positioning method, which comprises the following steps:
Step 1, training sample are collected: training sample includes positive sample and negative sample, and the picture of positive sample is that goat's horn bends normal shape
State, the picture of negative sample are the unknown objects that goat's horn bow miss status is either cut at random；
Step 2, LBP feature extraction: the pass of the pixel value of each pixel and its neighborhood in the training sample that step 1 is collected is calculated
System constitutes local binary pattern histogram, to extract LBP feature；
Step 3, HOG feature extraction: the training sample that step 1 is collected is divided into the unit comprising multiple pixels, then right
The edge of each pixel or gradient seek direction histogram in each unit, integrate all direction histograms, constitute HOG
Profiler；
Step 4, Fusion Features: by the LBP feature that step 2 is extracted and the HOG feature that step 3 is extracted, serial integration technology is used
It is merged, generates LBP-HOG fusion feature direction histogram；
Step 5, SVM classifier training: the LBP-HOG fusion feature that step 4 obtains being sent in SVM classifier and is trained,
Obtain training parameter；
Step 6, multi-scale sliding window mouth Scan orientation: target image is scanned using multi-scale sliding window mouth, by instructing in step 5
The SVM classifier perfected is predicted, realizes the positioning in goat's horn bow region.
2. the goat's horn according to claim 1 combined based on fusion feature and SVM bends area positioning method, feature exists
In the relationship of the pixel value of each pixel and its neighborhood, structure in the training sample that calculating step 1 described in step 2 is collected
It is specific as follows to extract LBP feature at local binary pattern histogram:
Step 2.1, the image for collecting step 1 are divided into 4 × 4 regions；
Step 2.2, for each pixel in region, select threshold value of its gray value as LBP operator, will be in 3 × 3 neighborhoods
The gray values of other pixels be compared with the threshold value of LBP operator, the gray value pixel bigger than threshold value is set to 1, gray scale
The value pixel smaller than threshold value is set to 0；Then 8 points in 3 × 3 neighborhoods are generated into 8 bits, by binary number
Decimal number is converted to get the LBP value of each pixel is arrived；
Step 2.3, the histogram for calculating each region, i.e., the frequency that each LBP value occurs, obtain 16 LBP histograms；
16 histograms are described step 2.4 with unified histogram；Characteristic pattern size is 80 × 60 pixels, by its point
At 4 × 4 pieces, the vector of generation is 4 × 4 × 256=4096 dimension.
3. the goat's horn according to claim 1 combined based on fusion feature and SVM bends area positioning method, feature exists
In the training sample for collecting step 1 described in step 3 is divided into the unit comprising several pixels, then to each list
The edge of each pixel or gradient seek direction histogram in member, integrate all direction histograms, constitute HOG feature and retouch
Device is stated, specific as follows:
Step 3.1, the gamma for editing image, formula are as follows:
I (x, y)=I (x, y)gamma (1)
Wherein I (x, y) indicates the pixel value of each pixel, gamma=1/2；
Step 3.2, the gradient direction value for calculating each pixel transverse and longitudinal coordinate in image, and then obtain the gradient of the pixel；
Set GxFor x direction gradient, GyFor y direction gradient, [- 1,0,1] is center template operator, the gray value of pixel (x, y)
For H (x, y), then the gradient formula of pixel (x, y) are as follows:
Gx(x, y)=H (x+1, y)-H (x-1, y) (2)
Gy(x, y)=H (x, y+1)-H (x, y-1) (3)
The gray scale size that G (x, y) is pixel (x, y) is set, θ is the gray scale direction of pixel (x, y), then:
Entire image is divided into cell by step 3.3, and each cell size is 10 × 10 pixels；Histogram of gradients selects 9
A section counts the gradient weights of each pixel, draws the histogram of gradients of each cell, obtains 9 dimensional feature vectors
Histogram of gradients；By every 2 × 2 cell spans at a block, normalize the feature vector of each cell, then each piece
Feature vector dimension is 2 × 2 × 9=36；
Step 3.4 merges the HOG feature of entire image, obtains entire image HOG feature；Due to each cell
Size is 10 × 10, and a block includes 2 × 2 cells, so the dimension of the HOG feature of entire image is 9 × 4 × (8-1)
× (6-1)=1260.
4. the goat's horn according to claim 1 combined based on fusion feature and SVM bends area positioning method, feature exists
In being carried out using serial integration technology by the LBP feature that step 2 is extracted and the HOG feature that step 3 is extracted described in step 4
Fusion generates LBP-HOG fusion feature direction histogram, specific as follows:
LBP feature space is set as A, HOG feature space is B, there are two feature vector α ∈ A and β ∈ B, using serially melting
It closes, obtains:
Wherein, γ is fused feature vector, and θ is weight coefficient；
Set fused matrix dimension as F, then:
F=dim (A)+dim (B) (7)
Goat's horn bow feature is merged using serial integration technology, generates LBP-HOG characteristic direction histogram；By LBP feature
The dimension 4096 of vector, the dimension 1260 of HOG feature vector, can obtain fused LBP-HOG feature vector dimension is 5356.
5. the goat's horn according to claim 1 combined based on fusion feature and SVM bends area positioning method, feature exists
In, the LBP-HOG fusion feature that step 4 obtains is sent in SVM classifier described in step 5 and is trained, acquisition training
Parameter, specific as follows:
The quantity of training sample is set as 900,900 LBP-HOG fusion features are sent into SVM classifier and are trained,
During positive training, 900 samples are divided into two groups respectively, every group includes 300 positive samples and 600 negative samples, training
Stage is as follows:
After first group of 300 positive samples and 600 negative samples are put into SVM training, then with second group of sample set to SVM
It is tested, obtains test result, the data that positive sample is mistaken in test result are put into first group of negative sample；Again
Training and the test process for carrying out SVM classifier, are put into the data that positive sample is mistaken in test result in first group of negative sample
In this, repeats the above process, obtain a final SVM classifier.
6. the goat's horn according to claim 1 combined based on fusion feature and SVM bends area positioning method, feature exists
Scan target image in, use multi-scale sliding window mouth described in step 6, by SVM classifier trained in step 5 into
The positioning in goat's horn bow region is realized in row prediction, specific as follows:
Binaryzation is carried out to target image, obtains 0 and 255 two kind of gray value, since background colour is black, it is white that goat's horn, which bends color,
Color, so the target of detection is the region that gray value is 255；Selecting size is the window of k × l, the pixel ash of image in window
Angle value is h (xi, yj), then the accounting that gray value is 255 are as follows:
In formula, η is the threshold value judged in advance, and value is η=0.5；To gray value be 255 accounting greater than 0.5 window area into
Row LBP-HOG feature extraction, is then predicted by trained SVM classifier, and candidate rectangle window is obtained；Using more rulers
Degree sliding window is detected, and when detecting multiple targets, merges candidate window, finally extracts goat's horn bow region.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811009549.1A CN109255379A (en) | 2018-08-31 | 2018-08-31 | A kind of goat's horn bow area positioning method combined based on fusion feature and SVM |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811009549.1A CN109255379A (en) | 2018-08-31 | 2018-08-31 | A kind of goat's horn bow area positioning method combined based on fusion feature and SVM |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109255379A true CN109255379A (en) | 2019-01-22 |
Family
ID=65049814
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811009549.1A Pending CN109255379A (en) | 2018-08-31 | 2018-08-31 | A kind of goat's horn bow area positioning method combined based on fusion feature and SVM |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109255379A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110006907A (en) * | 2019-04-10 | 2019-07-12 | 清华大学深圳研究生院 | A kind of die casting detection method of surface flaw and system based on machine vision |
| CN110163161A (en) * | 2019-05-24 | 2019-08-23 | 西安电子科技大学 | Multiple features fusion pedestrian detection method based on Scale invariant |
| CN112232215A (en) * | 2020-10-16 | 2021-01-15 | 哈尔滨市科佳通用机电股份有限公司 | Railway wagon coupler yoke key joist falling fault detection method |
| CN112287919A (en) * | 2020-07-07 | 2021-01-29 | 国网江苏省电力有限公司常州供电分公司 | Power equipment identification method and system based on infrared image |
| CN112800968A (en) * | 2021-01-29 | 2021-05-14 | 江苏大学 | Method for identifying identity of pig in drinking area based on feature histogram fusion of HOG blocks |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20130195321A1 (en) * | 2010-09-22 | 2013-08-01 | Matteo Sacchi | Pantograph monitoring system and method |
| CN108062568A (en) * | 2017-12-15 | 2018-05-22 | 成都唐源电气股份有限公司 | A kind of locator base openings pin lacks recognition methods |
-
2018
- 2018-08-31 CN CN201811009549.1A patent/CN109255379A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20130195321A1 (en) * | 2010-09-22 | 2013-08-01 | Matteo Sacchi | Pantograph monitoring system and method |
| CN108062568A (en) * | 2017-12-15 | 2018-05-22 | 成都唐源电气股份有限公司 | A kind of locator base openings pin lacks recognition methods |
Non-Patent Citations (1)
| Title |
|---|
| 徐丹: "基于OpenCV的接触网旋转双耳缺陷自动识别", 《中国优秀硕士学位论文全文数据库》 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110006907A (en) * | 2019-04-10 | 2019-07-12 | 清华大学深圳研究生院 | A kind of die casting detection method of surface flaw and system based on machine vision |
| CN110163161A (en) * | 2019-05-24 | 2019-08-23 | 西安电子科技大学 | Multiple features fusion pedestrian detection method based on Scale invariant |
| CN112287919A (en) * | 2020-07-07 | 2021-01-29 | 国网江苏省电力有限公司常州供电分公司 | Power equipment identification method and system based on infrared image |
| CN112287919B (en) * | 2020-07-07 | 2022-08-30 | 国网江苏省电力有限公司常州供电分公司 | Power equipment identification method and system based on infrared image |
| CN112232215A (en) * | 2020-10-16 | 2021-01-15 | 哈尔滨市科佳通用机电股份有限公司 | Railway wagon coupler yoke key joist falling fault detection method |
| CN112232215B (en) * | 2020-10-16 | 2021-04-06 | 哈尔滨市科佳通用机电股份有限公司 | Railway wagon coupler yoke key joist falling fault detection method |
| CN112800968A (en) * | 2021-01-29 | 2021-05-14 | 江苏大学 | Method for identifying identity of pig in drinking area based on feature histogram fusion of HOG blocks |
| CN112800968B (en) * | 2021-01-29 | 2024-05-14 | 江苏大学 | HOG blocking-based feature histogram fusion method for identifying identity of pigs in drinking area |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109255379A (en) | A kind of goat's horn bow area positioning method combined based on fusion feature and SVM | |
| CN112434695B (en) | Upper pull rod fault detection method based on deep learning | |
| Schmugge et al. | Detection of cracks in nuclear power plant using spatial-temporal grouping of local patches | |
| CN108985170A (en) | Transmission line of electricity hanger recognition methods based on Three image difference and deep learning | |
| CN106290388A (en) | A kind of insulator breakdown automatic testing method | |
| CN113324864B (en) | Pantograph carbon slide plate abrasion detection method based on deep learning target detection | |
| CN104268505A (en) | Automatic cloth defect point detection and recognition device and method based on machine vision | |
| CN105044122A (en) | Copper part surface defect visual inspection system and inspection method based on semi-supervised learning model | |
| CN106886216A (en) | Robot automatic tracking method and system based on RGBD Face datections | |
| CN109087286A (en) | A kind of detection method and application based on Computer Image Processing and pattern-recognition | |
| CN113947731B (en) | Foreign matter identification method and system based on contact net safety inspection | |
| CN109376740A (en) | A kind of water gauge reading detection method based on video | |
| CN108765402A (en) | Non-woven fabrics defects detection and sorting technique | |
| CN109685766A (en) | A kind of Fabric Defect detection method based on region fusion feature | |
| CN104198497A (en) | Surface defect detection method based on visual saliency map and support vector machine | |
| CN109886170B (en) | Intelligent detection, identification and statistics system for oncomelania | |
| CN107798293A (en) | A kind of crack on road detection means | |
| CN109598681B (en) | No-reference quality evaluation method for image after repairing of symmetrical Thangka | |
| CN112967255A (en) | Shield segment defect type identification and positioning system and method based on deep learning | |
| CN110728269B (en) | High-speed rail contact net support pole number plate identification method based on C2 detection data | |
| CN113487570A (en) | High-temperature continuous casting billet surface defect detection method based on improved yolov5x network model | |
| CN106290379A (en) | Rail surface defects based on Surface scan camera detection device and method | |
| CN111242899A (en) | Image-based flaw detection method and computer-readable storage medium | |
| CN108508023B (en) | Defect detection system for contact end jacking bolt in railway contact network | |
| Gou et al. | Pavement crack detection based on the improved faster-rcnn |
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: 20190122 |
|
| RJ01 | Rejection of invention patent application after publication |