CN106951900B - A kind of automatic identifying method of arrester meter reading - Google Patents
A kind of automatic identifying method of arrester meter reading Download PDFInfo
- Publication number
- CN106951900B CN106951900B CN201710239497.6A CN201710239497A CN106951900B CN 106951900 B CN106951900 B CN 106951900B CN 201710239497 A CN201710239497 A CN 201710239497A CN 106951900 B CN106951900 B CN 106951900B
- Authority
- CN
- China
- Prior art keywords
- image
- reading
- connected domain
- area
- rectangle
- 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.)
- Active
Links
Classifications
-
- 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
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
- G06V10/443—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components by matching or filtering
-
- 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
- G06V10/24—Aligning, centring, orientation detection or correction of the image
- G06V10/243—Aligning, centring, orientation detection or correction of the image by compensating for image skew or non-uniform image deformations
-
- 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
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
- G06V10/267—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds
-
- 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
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
Abstract
The invention discloses a kind of automatic identifying methods of arrester meter reading, and the identification of lightning-arrest instruments and meters is divided into region segmentation and Recognition of Reading two parts；Instrument dial plate is divided into multiple regions first with connected domain detection algorithm and rectangle fitting algorithm；Then these regions are detected using priori knowledge, retains the connected domain for representing pointer area and numeric area；Minimum area rectangle fitting is carried out to two connected domains again, obtains two rectangles with deflection angle；Rotation correction is carried out to image according to angle, is then partitioned into pointer and numeric area from correction image；Recognition of Reading is carried out to pointer area and numeric area finally, being utilized respectively preset angle configuration and convolutional neural networks method.The present invention can be completed at the same time total indicator reading identification and digital Recognition of Reading, can effectively be corrected to image, improve the accuracy of reading.
Description
Technical field
The invention belongs to image identification technical field, in particular to a kind of automatic identifying method of arrester meter reading.
Background technique
Arrester is a kind of electrical equipment in substation dedicated for limitation lightning surge or switching overvoltage, internal
Containing a resistance value with the changed valve block of voltage.Under rated voltage, valve block resistance value is very big, is equivalent to an insulation
Body then flows through the current value very little and stabilization of the valve block；After both end voltage is more than threshold value, valve block is switched on, and is had very big
Electric current by valve block, be subsequently poured into the earth, avoid other equipment in parallel of heavy current impact in this way, electricity was completed with this
Pressure protection；When voltage restores, valve block state reverts to insulation, and current value also restores therewith.Valve block is living through voltge surge
Afterwards, it is likely that impaired, it is therefore desirable to the often working condition of detection arrester.
For the ease of detecting the working condition of arrester, a lightning-arrest instruments and meters is configured in equipment.Contain on instrument dial plate
Have pointer and digital two reading areas: the reading of pointer represents the current value for flowing through arrester, and it is lightning-arrest that digital reading represents this
Device is by number of lightning strokes (number for living through voltge surge).The size and arrester unusual condition of current value are directly related, are
The critical data for needing to record when inspection；To ensure that arrester can work normally, after it is by voltge surge, need into
The primary thoroughly manual inspection of row, therefore, number of lightning strokes also needs to record.Currently, this record work is still by artificial complete
At on the one hand this mode is easy to appear missing inspection, on the other hand the data of record and truthful data are easy to appear relatively large deviation.Cause
This, needs a kind of Meter recognition recording process of automation.
Current Meter recognition, handled object are all the instrument with single reading area.For pointer instrument
Identification, common method be by straight line fitting detect dial plate on pointer, then utilize angle information and priori knowledge meter
Calculate reading.This method is suitable for pointer length and dial plate length approaches and the obvious instrument of color of pointer texture；Work as instrument
Compared with timing, total indicator reading is more accurate for position, but when the instrument of shooting tilts, the result of identification often deviation compared with
Greatly.Therefore, be not suitable for the instrument that placement position is different in processing substation, and failure is identified for the pointer for instruments and meters of taking shelter from the thunder.
Identification for digital instrument, method is more, including template matching method, statistic decision method, BP neural network method
Deng.Wherein, template matching method is by the way that the template number in number to be identified and template library to be compared, most with similarity
Number corresponding to that big template is as recognition result；This method effect when identifying press figure is fine, but easily
By noise jamming, be not suitable for identification outdoor meter.Statistic decision method is difficult to reflect the tiny characteristics in number, thus using compared with
It is few.BP neural network method can constantly learn in the training process and modify parameters at different levels, to reach to training sample pole
Good classifying quality；But it is overly dependent upon the selection of input feature value.In addition, the digital instrument studied at present,
The segmentation in progress region is not needed, emphasis is all in Classification and Identification；And the numeric area in lightning-arrest instruments and meters is smaller, needs dividing
It could be handled after cutting, so to be improved to existing method.
Summary of the invention
The purpose of the invention is to provide it is a kind of it is adaptive it is strong, total indicator reading identification can be completed at the same time and number is read
The automatic identifying method of the arrester meter reading of number identification.
For this purpose, the technical scheme is that a kind of automatic identifying method of arrester meter reading, including following step
It is rapid:
1) image preprocessing: utilizing gaussian pyramid down-sampled images, the Instrument image of input is reduced resolution ratio, then
Color image is converted into gray level image, and gray level image is filtered；
2) region segmentation: the characteristics of according to Meter recognition region, determine that the detailed process of image segmentation is as follows:
A1) edge extracting and Morphological scale-space；Edge detection is carried out using Canny operator, obtains a bianry image, so
The edge of expansion process connection fracture is utilized afterwards；
A2) connected domain detects；Using 8 fields of pixel as syntople, the connected domain in simultaneously tag image, whole picture figure are found
As N number of connected domain will be divided into；
A3 the boundary rectangle of each connected domain) is calculated；The positive rectangle of all pixels point comprising some connected domain is exactly this
The boundary rectangle of a connected domain；
A4 interference connected domain) is filtered out；According to pointer area and numeric area in the relative position information of dial plate, can filter out
Fall the incorrect extraneous rectangles such as some areas are too small, area is excessive, percent information gap is big, namely filters out underproof company
Logical domain；If the resolution ratio of image is w*h, the information of extraneous rectangle are as follows: (xi,yi, width, height), wherein the first two information
The coordinate on rectangle vertex is represented, latter two information represents the width and height of rectangle；When this rectangle meet following (I), (II),
(III) when either condition, the UNICOM domain corresponding to it is deleted；If remaining connected domain > 2, the most work of the percentage of occupancy
To retain item；
If after the UNICOM domain of either condition in satisfaction (I), (II), (III) is filtered out, remaining connected domain number is 2,
Then enter step a5), otherwise, remaining connected domain is calculated as the following formula:Wherein molecules present is connected to
The number of available point in domain, denominator represent the area of extraneous rectangle；Calculated result is ranked up, maximum two is selected
It is a, as remaining two connected domains；
A5 minimum area rectangle fitting) is carried out to remaining two connected domains；Minimum area rectangle is referred to comprising connecting
The smallest rectangle of area of all pixels point in logical domain, and this rectangle is likely to inclined；
If the central point of two rectangles of fitting is respectively p1(x1,y1) and p2(x2,y2), deflection angle is respectively θ1And θ2, meter
It calculates
A6) according to θ correcting colour images and bianry image；School is exactly based on spatial alternation and is mapped as original image newly
Image, original coordinates are converted to new coordinate as the following formula: where two rows three column matrix be known as affine matrix；
Instrument image timing only carries out rotation correction, without translating, therefore affine matrix are as follows:
When carrying out spatial alternation, the position corresponding to input picture is found in turn from the pixel of output image, if mapping
Pixel non-integer, then utilize the single order interpolation calculation pixel value；If the pixel of mapping exceeds the model of input picture
It encloses, then the point is assigned a value of zero；
A7 pointer area P) is partitioned into from the bianry image of correction1, and pointer is partitioned into from the color image of correction
Region P2With numeric area P3；
3) Recognition of Reading: lightning-arrest instruments and meters has pointer and digital two reading areas, using Hough transform detection straight line and
Preset angle configuration identifies total indicator reading；Divide number using sciagraphy, neural network identifies number；The reading for completing entire instrument is known
Not.
Further, a2 in the step 2)) described in connected domain detect specific steps are as follows:
B1) image is progressively scanned, the pixel that pixel value in each row is 255 is formed a whole, and
Line number, the label number, starting point, terminal for marking it, the pixel value of this all pixels point for integrally being included is assigned a value of marking
Number；
B2) from the second row, whether the region for observing this line connection intersects with the region of lastrow, if intersection,
The label number in the region is changed to the label number with his intersecting area, the pixel value of the pixel covered is also changed to label number
Value；Otherwise constant；
B3) after image scanning, the pixel value of all non-zero pixels points is updated；
B4) image is scanned again, regards the identical pixel of pixel value as a connected domain, such entire image
N number of connected domain will be divided into.
Further, a7 in the step 2)) described in be partitioned into pointer area P1, and divide from the color image of correction
Cut out pointer area P2With numeric area P3Specific segmentation step are as follows:
C1) using the bianry image after correcting as object, connected domain therein is detected, the boundary rectangle and meter of connected domain are fitted
Calculate the center point coordinate p (x of rectanglei,yi)；
C2) by the p of all center point coordinates and step a5)1(x1,y1) and p2(x2,y2) be compared, retain coordinate most
Two close rectangles；
C3 it) is partitioned into P respectively from the image of correction1,P2,P3。
Further, specific step is as follows for the identification of pointer described in the step 3):
D1 the boundary line angle of pointer area) is calculated；Utilize Hough transform detection zone P1Two boundary line l1,l2,
Calculate the angle of two lines sectionBecause of P1It is the region divided from bianry image, only includes the apparent line segment in boundary,
There is no the interference of pointer and graduation mark, can more accurately detect the two lines section for representing instrument boundary；
D2) to region P2Carry out gradation conversion, homomorphic filtering processing and piecewise linear transform processing；Homomorphic filtering pass through by
Reflected light and incident light separation, filter out incident light therein, to reduce illumination influence caused by image；It is filtered by homomorphism
Treated that picture contrast is low for wave, using piecewise linear transform using the gray average of image as boundary, improves whole image
Contrast；
D3) image binaryzation；Binarization threshold is calculated using maximum variance between clusters, then obtains binary image；
D4 the deflection angle of pointer) is calculated；First with the line segment in Hough transform detection binary image；Then it screens
Line segment filters out and boundary line l in a plurality of line segment of detection1,l2Line segment similar in position, then length is selected from remaining line segment
Spend maximum one；Finally calculate the angle, θ of this line segmentl；
D5 reading) is calculated；Scale limitation is [Vs,Ve], then pass through calculating
Obtain total indicator reading.
Further, specific step is as follows for the identification of number described in the step 3):
E1) with region P3To carry out gradation conversion, homomorphic filtering processing and piecewise linear transform processing to picture；Homomorphism filter
Wave is by filtering out incident light therein, to reduce illumination influence caused by image for reflected light and incident light separation；Through
It crosses homomorphic filtering treated that picture contrast is low, using piecewise linear transform using the gray average of image as boundary, improve whole
The contrast of a image；Carry out gray proces, filtering processing, piecewise linear transform, binaryzation；Image binaryzation；Use maximum
Ostu method calculates binarization threshold, then obtains binary image；
E2) individual digit is divided；Assuming that the resolution ratio of image is w*h, the image is progressively scanned, obtains one containing h
The one-dimension array Arr of elementl[h], wherein the value of i-th of element represents the non-zero pixel number of the i-th row；It scans by column, obtains
To an one-dimension array Arr containing w elementv[w], wherein the value of i-th of element represents the non-zero pixel of the i-th column
Number；Seek ArrlTwo local minimum position y of [h]min1,ymin2, retain ymin1And ymin2Interior zone, remaining region view
For boundary；Seek ArrvThe local minimum of [w] can obtain multiple position xm1,xm2...xmn, believed according to position digital in instrument
Breath, filters out four point x for representing digital boundarya,xb,xc,xd；Finally, by following rectangular information segmentation number: (xa,ymin1,
xb-xa,ymin2-ymin1) and (xc,ymin1,xd-xc,ymin2-ymin1)；
E3) the number identification based on convolutional neural networks；The structure of convolutional neural networks uses internal 5 layers of structure, that is, removes
Outside input, output layer, two layers of convolutional layer, two layers of pond layer and one layer of full linking layer are chosen；Wherein, input layer will use grayscale image
As input vector, output layer is classified using the more classification functions of softmax；Training sample directly uses the number of projection localization
Word, the sample size of each number are at least 50；When number identification, will directly it be divided using trained neural network
Class completes the Recognition of Reading of numeric area.
The identification of lightning-arrest instruments and meters is divided into region segmentation and Recognition of Reading two parts by the present invention.It is examined first with connected domain
Instrument dial plate is divided into multiple regions by method of determining and calculating and rectangle fitting algorithm；Then these regions, mistake are detected using priori knowledge
Nonsensical region is filtered, the connected domain for representing pointer area and numeric area is retained.Two connected domains are carried out most again
Small area rectangle fitting obtains two rectangles with deflection angle；Rotation correction is carried out to image according to angle, then from school
Pointer and numeric area are partitioned into positive image.Finally, be utilized respectively preset angle configuration and convolutional neural networks method to pointer area and
Numeric area carries out Recognition of Reading.
The present invention can be completed at the same time total indicator reading identification and digital Recognition of Reading, in addition, the present invention can also overcome room
External environment adverse effect caused by Instrument image has good adaptability；Meanwhile, it is capable to carry out effective school to image
Just, the accuracy of reading is improved.
Detailed description of the invention
It is described in further detail below in conjunction with attached drawing and embodiments of the present invention
Fig. 1 is system block diagram of the invention；
Fig. 2 is region segmentation flow chart of the invention；
Fig. 3 is Recognition of Reading flow chart of the invention.
Specific embodiment
Referring to attached drawing.A kind of automatic identifying method of arrester meter reading, is completed at the same time pointer described in the present embodiment
The recognition methods of Recognition of Reading and digital Recognition of Reading.Two readings are completed using connected domain detection algorithm and rectangle fitting algorithm
Then the segmentation in region is utilized respectively preset angle configuration and convolutional neural networks method and carries out Recognition of Reading to two regions.Overall structure
Figure is as shown in Figure 1.
Specifically includes the following steps:
1) image preprocessing:
The Instrument image of input is usually the color image that resolution ratio is 1920 × 1080.Input picture is pre-processed
Process are as follows:
F1) image down sampling.Using gaussian pyramid down-sampled images, after processing, will obtain a resolution ratio be 960 ×
540 color image；
F2) color image is converted to gray level image；
F3 it) is filtered.It is filtering object with gray level image, progress median filtering, the spiced salt removed in image first is made an uproar
Sound；Then the Gaussian noise in image, while smooth fine edge are removed using gaussian filtering.
2) region segmentation
Region segmentation is to carry out later period Recognition of Reading precondition, and the quality of segmentation effect directly affects Recognition of Reading
Effect.Object of the invention is the dual area instrument simultaneously with pointer area and numeric area, according to Meter recognition region
Feature determines that the detailed process of image segmentation is as follows:
A1) edge extracting and Morphological scale-space；Edge detection is carried out using Canny operator, obtains a bianry image, so
The edge of expansion process connection fracture is utilized afterwards；
A2) connected domain detects；Using 8 fields of pixel as syntople, the connected domain in simultaneously tag image is found, it is specific to flow
Journey are as follows:
B1) image is progressively scanned, the pixel that pixel value in each row is 255 is formed a whole, and
Line number, the label number, starting point, terminal for marking it, the pixel value of this all pixels point for integrally being included is assigned a value of marking
Number；
B2) from the second row, whether the region for observing this line connection intersects with the region of lastrow, if intersection,
The label number in the region is changed to the label number with his intersecting area, the pixel value of the pixel covered is also changed to label number
Value；Otherwise constant；
B3) after image scanning, the pixel value of all non-zero pixels points is updated；
B4) image is scanned again, regards the identical pixel of pixel value as a connected domain, such entire image
N number of connected domain will be divided into；
A3 the boundary rectangle of each connected domain) is calculated；The positive rectangle of all pixels point comprising some connected domain is exactly this
The boundary rectangle of a connected domain；
A4 interference connected domain) is filtered out；According to pointer area and numeric area in the relative position information of dial plate, can filter out
Fall the incorrect extraneous rectangles such as some areas are too small, area is excessive, percent information gap is big, namely filters out underproof company
Logical domain；If the resolution ratio of image is w*h, the information of extraneous rectangle are as follows: (xi,yi, width, height), wherein the first two information
The coordinate on rectangle vertex is represented, latter two information represents the width and height of rectangle；When this rectangle meet following (I), (II),
(III) when either condition, the UNICOM domain corresponding to it is deleted；If remaining connected domain > 2, the most work of the percentage of occupancy
To retain item；
If after the UNICOM domain of either condition in satisfaction (I), (II), (III) is filtered out, remaining connected domain number is 2,
Then enter step a5), otherwise, remaining connected domain is calculated as the following formula:Wherein molecules present is connected to
The number of available point in domain, denominator represent the area of extraneous rectangle；Calculated result is ranked up, maximum two is selected
It is a, as remaining two connected domains；
A5 minimum area rectangle fitting) is carried out to remaining two connected domains；Minimum area rectangle is referred to comprising connecting
The smallest rectangle of area of all pixels point in logical domain, and this rectangle is likely to inclined；
If the central point of two rectangles of fitting is respectively p1(x1,y1) and p2(x2,y2), deflection angle is respectively θ1And θ2, meter
It calculates
A6) according to θ correcting colour images and bianry image；School is exactly based on spatial alternation and is mapped as original image newly
Image, original coordinates are converted to new coordinate as the following formula: where two rows three column matrix be known as affine matrix；
Instrument image timing only carries out rotation correction, without translating, therefore affine matrix are as follows:When carrying out spatial alternation, finds and correspond in turn from the pixel of output image
The position of input picture utilizes the single order interpolation calculation pixel value if the pixel non-integer of mapping；If the picture of mapping
Vegetarian refreshments exceeds the range of input picture, then the point is assigned a value of zero；
A7 pointer area P) is partitioned into from the bianry image of correction1, and pointer is partitioned into from the color image of correction
Region P2With numeric area P3；
Specific segmentation step are as follows:
C1) using the bianry image after correcting as object, connected domain therein is detected, the boundary rectangle and meter of connected domain are fitted
Calculate the center point coordinate p (x of rectanglei,yi)；
C2) by the p of all center point coordinates and step a5)1(x1,y1) and p2(x2,y2) be compared, retain coordinate most
Two close rectangles；
C3 it) is partitioned into P respectively from the image of correction1,P2,P3。
3) Recognition of Reading
Lightning-arrest instruments and meters has pointer and digital two reading areas, and the present invention uses Hough transform detection straight line and angle
Method identifies total indicator reading；Divide number using sciagraphy, neural network identifies number；Complete the Recognition of Reading of entire instrument.
Detailed process is as follows.
Specific step is as follows for pointer identification:
D1 the boundary line angle of pointer area) is calculated；Utilize Hough transform detection zone P1Two boundary line l1,l2,
Calculate the angle of two lines sectionBecause of P1It is the region divided from bianry image, only includes the apparent line segment in boundary,
There is no the interference of pointer and graduation mark, can more accurately detect the two lines section for representing instrument boundary；
D2) to region P2Carry out gradation conversion, homomorphic filtering processing and piecewise linear transform processing；Homomorphic filtering pass through by
Reflected light and incident light separation, filter out incident light therein, to reduce illumination influence caused by image；It is filtered by homomorphism
Treated that picture contrast is low for wave, using piecewise linear transform using the gray average of image as boundary, improves whole image
Contrast；
D3) image binaryzation；Binarization threshold is calculated using maximum variance between clusters, then obtains binary image；
D4 the deflection angle of pointer) is calculated；First with the line segment in Hough transform detection binary image；Then it screens
Line segment filters out and boundary line l in a plurality of line segment of detection1,l2Line segment similar in position, then length is selected from remaining line segment
Spend maximum one；Finally calculate the angle, θ of this line segmentl；
D5 reading) is calculated；Scale limitation is [Vs,Ve], then pass through calculating
Obtain total indicator reading.
Specific step is as follows for number identification:
E1) with P3To carry out gray proces, filtering processing, piecewise linear transform, binaryzation to picture；With pointer identification
Step d2) and step d3)；
E2) individual digit is divided；Assuming that the resolution ratio of image is w*h, the image is progressively scanned, obtains one containing h
The one-dimension array Arr of elementl[h], wherein the value of i-th of element represents the non-zero pixel number of the i-th row；It scans by column, obtains
To an one-dimension array Arr containing w elementv[w], wherein the value of i-th of element represents the non-zero pixel of the i-th column
Number；Seek ArrlTwo local minimum position y of [h]min1,ymin2, retain ymin1And ymin2Interior zone, remaining region view
For boundary；Seek ArrvThe local minimum of [w] can obtain multiple position xm1,xm2...xmn, believed according to position digital in instrument
Breath, filters out four point x for representing digital boundarya,xb,xc,xd；Finally, by following rectangular information segmentation number: (xa,ymin1,
xb-xa,ymin2-ymin1) and (xc,ymin1,xd-xc,ymin2-ymin1)；
E3) the number identification based on convolutional neural networks；The structure of convolutional neural networks uses internal 5 layers of structure, that is, removes
Outside input, output layer, two layers of convolutional layer, two layers of pond layer and one layer of full linking layer are chosen；Wherein, input layer will use grayscale image
As input vector, output layer is classified using the more classification functions of softmax；Training sample directly uses the number of projection localization
Word, the sample size of each number are at least 50；When number identification, will directly it be divided using trained neural network
Class completes the Recognition of Reading of numeric area.
Claims (4)
1. a kind of automatic identifying method of arrester meter reading, it is characterised in that: the following steps are included:
1) image preprocessing: utilizing gaussian pyramid down-sampled images, and the Instrument image of input is reduced resolution ratio, then will be color
Chromatic graph picture is converted to gray level image, and is filtered to gray level image；
2) region segmentation: the characteristics of according to Meter recognition region, determine that the detailed process of image segmentation is as follows:
A1) edge extracting and Morphological scale-space；Edge detection is carried out using Canny operator, obtains a bianry image, it is then sharp
With the edge of expansion process connection fracture；
A2) connected domain detects；Using 8 fields of pixel as syntople, finds and the connected domain in tag image, entire image will
It is divided into N number of connected domain；
A3 the boundary rectangle of each connected domain) is calculated；The positive rectangle of all pixels point comprising some connected domain is exactly this company
The boundary rectangle in logical domain；
A4 interference connected domain) is filtered out；According to pointer area and numeric area in the relative position information of dial plate, one can be filtered out
The incorrect extraneous rectangles such as area is too small, area is excessive, percent information gap is big a bit, namely filter out underproof connection
Domain；If the resolution ratio of image is w*h, the information of extraneous rectangle are as follows: (xi,yi, width, height), wherein the first two information generation
The coordinate on table rectangle vertex, latter two information represent the width and height of rectangle；When this rectangle meet following (I), (II),
(III) when either condition, the UNICOM domain corresponding to it is deleted；If remaining connected domain > 2, the most work of the percentage of occupancy
To retain item；
If will meet (I), (II), after the UNICOM domain of either condition filters out in (III), remaining connected domain number is 2, then into
Enter step a5), otherwise, remaining connected domain is calculated as the following formula:Wherein in molecules present connected domain
The number of available point, denominator represent the area of extraneous rectangle；Calculated result is ranked up, selects maximum two, i.e.,
For remaining two connected domains；
A5 minimum area rectangle fitting) is carried out to remaining two connected domains；Minimum area rectangle is referred to comprising connected domain
The smallest rectangle of area of interior all pixels point, and this rectangle is likely to inclined；
If the central point of two rectangles of fitting is respectively p1(x1,y1) and p2(x2,y2), deflection angle is respectively θ1And θ2, calculate
A6) according to θ correcting colour images and bianry image；School is exactly based on spatial alternation and original image is mapped as to new figure
Original coordinates are converted to new coordinate by picture as the following formula: where the matrix of two rows three column is known as affine matrix；
Instrument image timing only carries out rotation correction, without translating, therefore affine matrix are as follows:
When carrying out spatial alternation, the position corresponding to input picture is found in turn from the pixel of output image, if mapping
Pixel non-integer, then utilize the single order interpolation calculation pixel value；If the pixel of mapping exceeds the model of input picture
It encloses, then the point is assigned a value of zero；
A7 pointer area P) is partitioned into from the bianry image of correction1, and pointer area P is partitioned into from the color image of correction2
With numeric area P3；
Specific segmentation step are as follows:
C1) using the bianry image after correcting as object, connected domain therein is detected, the boundary rectangle of connected domain is fitted and calculates square
Center point coordinate p (the x of shapei,yi)；
C2) by the p of all center point coordinates and step a5)1(x1,y1) and p2(x2,y2) be compared, it is closest to retain coordinate
Two rectangles；
C3 it) is partitioned into P respectively from the image of correction1,P2,P3；
3) Recognition of Reading: lightning-arrest instruments and meters has pointer and digital two reading areas, detects straight line and angle using Hough transform
Method identifies total indicator reading；Divide number using sciagraphy, neural network identifies number；Complete the Recognition of Reading of entire instrument.
2. a kind of automatic identifying method of arrester meter reading as described in claim 1, it is characterised in that: the step 2)
Middle a2) described in connected domain detect specific steps are as follows:
B1) image is progressively scanned, the pixel that pixel value in each row is 255 is formed a whole, and marks
The pixel value of this all pixels point for integrally being included is assigned a value of label number by its line number, label number, starting point, terminal；
B2) from the second row, whether the region for observing this line connection intersects with the region of lastrow, should if intersection
The label number in region is changed to the label number with his intersecting area, and the pixel value of the pixel covered is also changed to the value of label number；
Otherwise constant；
B3) after image scanning, the pixel value of all non-zero pixels points is updated；
B4) image is scanned again, regards the identical pixel of pixel value as a connected domain, such entire image will be by
It is divided into N number of connected domain.
3. a kind of automatic identifying method of arrester meter reading as described in claim 1, it is characterised in that: the step 3)
Described in pointer identification specific step is as follows:
D1 the boundary line angle of pointer area) is calculated；Utilize Hough transform detection zone P1Two boundary line l1,l2, calculate
The angle of two lines sectionBecause of P1It is the region divided from bianry image, only includes the apparent line segment in boundary, do not have
The interference of pointer and graduation mark can more accurately detect the two lines section for representing instrument boundary；
D2) to region P2Carry out gradation conversion, homomorphic filtering processing and piecewise linear transform processing；Homomorphic filtering will be by that will reflect
Light and incident light separation, filter out incident light therein, to reduce illumination influence caused by image；At homomorphic filtering
Picture contrast after reason is low, using piecewise linear transform using the gray average of image as boundary, improves the comparison of whole image
Degree；
D3) image binaryzation；Binarization threshold is calculated using maximum variance between clusters, then obtains binary image；
D4 the deflection angle of pointer) is calculated；First with the line segment in Hough transform detection binary image；Then line segment is screened,
In a plurality of line segment of detection, filter out and boundary line l1,l2Line segment similar in position, then length is selected most from remaining line segment
Big one；Finally calculate the angle, θ of this line segmentl；
D5 reading) is calculated；Scale limitation is [Vs,Ve], then pass through calculating
Obtain total indicator reading.
4. a kind of automatic identifying method of arrester meter reading as described in claim 1, it is characterised in that: the step 3)
Described in number identification specific step is as follows:
E1) with region P3To carry out gradation conversion, homomorphic filtering processing and piecewise linear transform processing to picture；Homomorphic filtering passes through
Reflected light and incident light are separated, incident light therein is filtered out, to reduce illumination influence caused by image；By homomorphism
Picture contrast after filtering processing is low, using piecewise linear transform using the gray average of image as boundary, improves whole image
Contrast；Carry out gray proces, filtering processing, piecewise linear transform, binaryzation；Image binaryzation；Use side between maximum kind
Poor method calculates binarization threshold, then obtains binary image；
E2) individual digit is divided；Assuming that the resolution ratio of image is w*h, the image is progressively scanned, one is obtained and contains h element
One-dimension array Arrl[h], wherein the value of i-th of element represents the non-zero pixel number of the i-th row；It scans by column, obtains one
A one-dimension array Arr containing w elementv[w], wherein the value of i-th of element represents the non-zero pixel number of the i-th column；It asks
ArrlTwo local minimum position y of [h]min1,ymin2, retain ymin1And ymin2Interior zone, remaining region are considered as side
Boundary；Seek ArrvThe local minimum of [w] can obtain multiple position xm1,xm2...xmn, according to location information digital in instrument,
Filter out four point x for representing digital boundarya,xb,xc,xd；Finally, by following rectangular information segmentation number:
(xa,ymin1,xb-xa,ymin2-ymin1) and (xc,ymin1,xd-xc,ymin2-ymin1)；
E3) the number identification based on convolutional neural networks；The structure of convolutional neural networks uses internal 5 layers of structure, that is, except defeated
Enter, outside output layer, chooses two layers of convolutional layer, two layers of pond layer and one layer of full linking layer；Wherein, input layer will be made using grayscale image
For input vector, output layer is classified using the more classification functions of softmax；Training sample directly uses the number of projection localization
Word, the sample size of each number are at least 50；When number identification, will directly it be divided using trained neural network
Class completes the Recognition of Reading of numeric area.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710239497.6A CN106951900B (en) | 2017-04-13 | 2017-04-13 | A kind of automatic identifying method of arrester meter reading |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710239497.6A CN106951900B (en) | 2017-04-13 | 2017-04-13 | A kind of automatic identifying method of arrester meter reading |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106951900A CN106951900A (en) | 2017-07-14 |
| CN106951900B true CN106951900B (en) | 2019-10-22 |
Family
ID=59475911
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710239497.6A Active CN106951900B (en) | 2017-04-13 | 2017-04-13 | A kind of automatic identifying method of arrester meter reading |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106951900B (en) |
Families Citing this family (32)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107766836A (en) * | 2017-11-07 | 2018-03-06 | 国网黑龙江省电力有限公司检修公司 | A kind of circular deflection readings of pointer type meters antidote |
| CN108009535A (en) * | 2017-11-21 | 2018-05-08 | 武汉中元华电科技股份有限公司 | A kind of simple pointer meter reading method based on machine vision |
| CN108388894A (en) * | 2017-12-26 | 2018-08-10 | 新智数字科技有限公司 | A kind of recognition methods, device and the equipment of number meter reading |
| CN108345889B (en) * | 2018-02-27 | 2022-02-11 | 国网上海市电力公司 | Application method for performing reading identification on communication cabinet by utilizing raspberry party |
| JP6730379B2 (en) * | 2018-03-23 | 2020-07-29 | 株式会社東芝 | Reading system, reading method, program, and storage medium |
| CN108491842A (en) * | 2018-03-27 | 2018-09-04 | 康体佳智能科技（深圳）有限公司 | A kind of dial plate identifying system and recognition methods based on neural network |
| CN108573261A (en) * | 2018-04-17 | 2018-09-25 | 国家电网公司 | A kind of read out instrument recognition methods suitable for Intelligent Mobile Robot |
| CN108562821B (en) * | 2018-05-08 | 2021-09-28 | 中国电力科学研究院有限公司 | Method and system for determining single-phase earth fault line selection of power distribution network based on Softmax |
| CN108764234B (en) * | 2018-05-10 | 2021-10-12 | 浙江理工大学 | Liquid level meter reading identification method based on inspection robot |
| CN108764253A (en) * | 2018-05-15 | 2018-11-06 | 杭州电子科技大学 | Pointer instrument digitizing solution |
| CN109376749B (en) * | 2018-05-22 | 2021-03-19 | 国网山东省电力公司电力科学研究院 | Power transmission and transformation equipment infrared image temperature wide range identification method based on deep learning |
| CN108921203B (en) * | 2018-06-13 | 2022-04-01 | 深圳市云识科技有限公司 | Detection and identification method for pointer type water meter |
| CN109146806A (en) * | 2018-07-29 | 2019-01-04 | 国网上海市电力公司 | Gauge pointer position detection recognition methods based on shadow removing optimization in remote monitoriong of electric power |
| CN109389150B (en) * | 2018-08-28 | 2022-04-05 | 东软集团股份有限公司 | Image consistency comparison method and device, storage medium and electronic equipment |
| CN109271985A (en) * | 2018-09-07 | 2019-01-25 | 广东中粤电力科技有限公司 | A kind of digital instrument reading image-recognizing method and system |
| CN109977944B (en) * | 2019-02-21 | 2023-08-01 | 杭州朗阳科技有限公司 | Digital water meter reading identification method |
| CN110135420B (en) * | 2019-05-16 | 2021-11-16 | 北京灵汐科技有限公司 | Dial plate state identification method and device, readable storage medium and electronic equipment |
| CN110110697B (en) * | 2019-05-17 | 2021-03-12 | 山东省计算中心（国家超级计算济南中心） | Multi-fingerprint segmentation extraction method, system, device and medium based on direction correction |
| CN110210477B (en) * | 2019-05-24 | 2023-03-24 | 四川阿泰因机器人智能装备有限公司 | Digital instrument reading identification method |
| CN110414510B (en) * | 2019-07-26 | 2021-10-08 | 华中科技大学 | Reading correction method for pointer instrument |
| CN111091100A (en) * | 2019-12-21 | 2020-05-01 | 河海大学 | Concrete crack identification method based on complex noise image deep learning |
| CN111325164B (en) * | 2020-02-25 | 2023-11-21 | 北京眸视科技有限公司 | Pointer representation number identification method and device and electronic equipment |
| CN111444781B (en) * | 2020-03-09 | 2023-08-29 | 武汉理工大学 | Water meter reading identification method, device and storage medium |
| CN111476787A (en) * | 2020-04-23 | 2020-07-31 | 中科开创（广州）智能科技发展有限公司 | Automatic reading method and device for adaptive distortion of pointer meter |
| CN111598094B (en) * | 2020-05-27 | 2023-08-18 | 深圳市铁越电气有限公司 | Angle regression instrument reading identification method, equipment and system based on deep learning |
| CN112668578B (en) * | 2020-12-31 | 2023-11-07 | 中广核研究院有限公司 | Pointer type instrument reading method, pointer type instrument reading device, computer equipment and storage medium |
| CN112906602B (en) * | 2021-03-04 | 2023-08-25 | 杭州电力设备制造有限公司 | Automatic identification device and identification method for electric quantity meter of power distribution cabinet based on image processing |
| CN113159027B (en) * | 2021-04-13 | 2024-02-09 | 杭州电子科技大学 | Seven-segment digital display instrument identification method based on minimum external rectangular variant |
| CN113469162B (en) * | 2021-06-02 | 2023-09-26 | 广东白云学院 | Pointer instrument identification method, device, equipment and medium based on double-scale segmentation |
| CN113642437B (en) * | 2021-08-03 | 2022-05-31 | 中国地质大学（北京） | Quantitative calculation method for content and radius of different components in coal |
| CN116612118B (en) * | 2023-07-19 | 2023-10-03 | 中建五局第三建设有限公司 | Artificial intelligence-based quality detection and evaluation method for building lightning arrester |
| CN117095246A (en) * | 2023-10-20 | 2023-11-21 | 国网江西省电力有限公司超高压分公司 | Polarization imaging-based deep learning pointer instrument reading identification method |
Family Cites Families (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102975826A (en) * | 2012-12-03 | 2013-03-20 | 上海海事大学 | Portable ship water gauge automatic detection and identification method based on machine vision |
| CN103914843B (en) * | 2014-04-04 | 2018-04-03 | 上海交通大学 | The image partition method marked based on watershed algorithm and morphology |
| CN103955694B (en) * | 2014-04-09 | 2017-05-24 | 广州邦讯信息系统有限公司 | Image recognition meter reading system and method |
| CN105203148B (en) * | 2015-09-30 | 2018-03-23 | 湖北工业大学 | A kind of visible detection method of automobile combination meter |
| CN105303168A (en) * | 2015-10-14 | 2016-02-03 | 南京第五十五所技术开发有限公司 | Multi-view pointer type instrument identification method and device |
| CN105740829A (en) * | 2016-02-02 | 2016-07-06 | 暨南大学 | Scanning line processing based automatic reading method for pointer instrument |
| CN106529519A (en) * | 2016-09-19 | 2017-03-22 | 国家电网公司 | Automatic number identification method and system of power pointer type instrument |
-
2017
- 2017-04-13 CN CN201710239497.6A patent/CN106951900B/en active Active
Non-Patent Citations (1)
| Title |
|---|
| 基于图像的换流站用避雷器仪表识别;崔家瑞等;《电工技术学报》;20151231;第30卷(第增刊1期);第377-382页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106951900A (en) | 2017-07-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106951900B (en) | A kind of automatic identifying method of arrester meter reading | |
| CN110363182B (en) | Deep learning-based lane line detection method | |
| CN112699876B (en) | Automatic reading method for various meters of gas collecting station | |
| CN111598098B (en) | Water gauge water line detection and effectiveness identification method based on full convolution neural network | |
| CN110533064B (en) | Partial discharge map mode identification method and system of GIS (geographic information System) equipment | |
| CN109636784A (en) | Saliency object detection method based on maximum neighborhood and super-pixel segmentation | |
| CN108764234B (en) | Liquid level meter reading identification method based on inspection robot | |
| CN109409205A (en) | Video road driveway line detecting method of taking photo by plane based on line pitch characteristics point cluster | |
| CN109858480A (en) | A kind of digital instrument recognition methods | |
| CN109086763B (en) | Pointer instrument reading method and device | |
| Chen et al. | Extracting contour lines from common-conditioned topographic maps | |
| CN108629347A (en) | The wide tuning drive gear formula instrument automatic identifying method of substation | |
| CN111126253A (en) | Knife switch state detection method based on image recognition | |
| CN115359239A (en) | Wind power blade defect detection and positioning method and device, storage medium and electronic equipment | |
| CN112132131A (en) | Measuring cylinder liquid level identification method and device | |
| CN111291818B (en) | Non-uniform class sample equalization method for cloud mask | |
| KR101874968B1 (en) | Visibility measuring system base on image information and method for using the same | |
| CN110070545A (en) | A kind of method that textural characteristics density in cities and towns automatically extracts cities and towns built-up areas | |
| CN116188756A (en) | Instrument angle correction and indication recognition method based on deep learning | |
| CN114120094A (en) | Water pollution identification method and system based on artificial intelligence | |
| CN112734729A (en) | Water gauge water level line image detection method and device suitable for night light supplement condition and storage medium | |
| CN113592839B (en) | Distribution network line typical defect diagnosis method and system based on improved fast RCNN | |
| CN114266947A (en) | Classification method and device based on fusion of laser point cloud and visible light image | |
| CN112036391B (en) | Electronic inspection method and system based on shale gas field production station | |
| CN108694415A (en) | Image characteristic extracting method, device and water source image classification method, device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |