CN109815800A - Object detection method and system based on regression algorithm - Google Patents
Object detection method and system based on regression algorithm Download PDFInfo
- Publication number
- CN109815800A CN109815800A CN201811547328.XA CN201811547328A CN109815800A CN 109815800 A CN109815800 A CN 109815800A CN 201811547328 A CN201811547328 A CN 201811547328A CN 109815800 A CN109815800 A CN 109815800A
- Authority
- CN
- China
- Prior art keywords
- image
- detection model
- detected
- verifying
- target
- 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
Landscapes
- Image Analysis (AREA)
Abstract
The present invention provides a kind of object detection method and system based on regression algorithm, is related to technical field of image detection, and this method includes obtaining device context image to be detected and constructing sample database；Establish the detection model of deep learning；Input detection model is iterated training after image in sample database is carried out grid dividing processing, until reaching preset trained accuracy；The image with measurement equipment to be checked is obtained to form verifying collection, the image input detection model that verifying is concentrated is verified, until reaching preset verifying accuracy；The image with equipment deficiency to be detected is obtained to form test set, the image input detection model in test set is tested, until reaching preset test accuracy；Image to be detected input detection model is detected, to export target defect image.Object detection method provided in an embodiment of the present invention and system can improve the accuracy rate and timeliness of target detection with intelligent screening target object.
Description
Technical field
The present invention relates to technical field of image detection, more particularly, to a kind of object detection method based on regression algorithm and
System.
Background technique
Current power transmission route, which generates the main reason for failure, to be had: artificial origin's damage, inside even from weather transmission line of electricity, defeated
Electric line is contaminated, the transmission line malfunction that bird pest reason generates etc..In order to guarantee the normal operation of power grid, it is necessary to power transmission line
The defect of road is monitored, early warning and protection.For this problem, currently used method is manually to examine on the spot, and discovery is asked
Topic, then got rid of the danger by staff.But the problem of such scheme, is that the huge tour dead angle of workload is more, and there are one
Determine risk, it is difficult to guarantee the accuracy and real-time of monitored results.
In recent years, with the rise of unmanned air vehicle technique, the inspection of transmission line of electricity is started gradually distant by unmanned plane low latitude
Sensing mode substitution manually examines on the spot, and unmanned plane inspection is also because its efficient, accurate, safety advantage has obtained power department
Using.But another problem to be solved is, the picture number of taking photo by plane obtained by unmanned plane is very huge, if still
So large-scale picture is handled using the method for artificial screening, investigation, will certainly encounter and manually examine on the spot similar ask
Topic, as accuracy reduces, it is difficult to guarantee the real-time etc. of defect information.
Summary of the invention
In view of this, the purpose of the present invention is to provide a kind of object detection method and system based on regression algorithm, it can
With intelligent screening target object, the accuracy rate and timeliness of target detection are improved.
In a first aspect, the embodiment of the invention provides a kind of object detection methods based on regression algorithm, comprising: obtain to
Detection device live image, and according to the device context picture construction sample database to be detected；Establish the detection of deep learning
Model, the input of the detection model are the original image in sample database, and the output of the detection model is defect image；It should
Input detection model is iterated training after image in sample database carries out grid dividing processing, until reaching preset instruction
Practice frequency threshold value or reaches preset trained accuracy；The image with measurement equipment to be checked is obtained to form verifying collection, will be verified
The detection model that the image of concentration input after the completion of training after grid dividing processing is verified, and is tested until reaching preset
Card frequency threshold value reaches preset verifying accuracy；Image of the acquisition with equipment deficiency to be detected, will to form test set
The detection model that image in test set input after the completion of verifying after grid dividing processing is tested, until reaching default
Testing time threshold value or reach preset test accuracy；By image to be detected input the detection model after the completion of the test into
Row detection, to export target defect image.
With reference to first aspect, the embodiment of the invention provides the first possible embodiments of first aspect, wherein on
Stating the step of image lattice in the sample database is divided processing includes: to obtain image to be processed；The image is divided
At the grid subgraph of N*N；The target frame of multiple predictions is set for each grid subgraph；Calculate each target frame
Threshold value deletes the lower target window of threshold value.
With reference to first aspect, the embodiment of the invention provides second of possible embodiments of first aspect, wherein on
The step of stating according to the device context picture construction sample database to be detected, comprising: to be checked to this using data enhancing technology
Measurement equipment live image carries out conversion process, the image that obtains that treated；According to the device context image to be detected and the processing
Image afterwards constructs sample database jointly.
The possible embodiment of second with reference to first aspect, the embodiment of the invention provides the third of first aspect
Possible embodiment, wherein data enhancing technology includes rotation transformation, reflection transformation, turning-over changed, scale transformation, puts down
Move one of transformation, change of scale and noise disturbance or a variety of.
With reference to first aspect, the embodiment of the invention provides the 4th kind of possible embodiments of first aspect, wherein
When the detection model is iterated training, verifying and test, further includes: carry out target detection by deep learning technology.
The 4th kind of possible embodiment with reference to first aspect, the embodiment of the invention provides the 5th kind of first aspect
Possible embodiment, wherein the step of carrying out target detection above by deep learning technology, comprising: area is carried out to image
Domain selection, and entire image is traversed using sliding window strategy, selection target region positions the position of target；Extract the target
The characteristic information in region；Classified by classifier to this feature information.
The 5th kind of possible embodiment with reference to first aspect, the embodiment of the invention provides the 6th kind of first aspect
Possible embodiment, wherein the step of the characteristic information of the said extracted target area, comprising: the target area is carried out
Foundation characteristic extracts, which includes contour feature and color characteristic；It is complicated that multilayer is carried out to the foundation characteristic of extraction
Feature extraction, the multilayer complex characteristic include profile layered characteristic and gray level image feature；To the multilayer complex characteristic of extraction
Weight study is carried out, the biggish feature of weight is exported, to predict output result.
The 5th kind of possible embodiment with reference to first aspect, the embodiment of the invention provides the 7th kind of first aspect
Possible embodiment, wherein the classifier is SVM classifier or Adaboost classifier.
With reference to first aspect, the embodiment of the invention provides the 8th kind of possible embodiments of first aspect, wherein should
Frequency of training threshold value is 200,000 times, which is 90%；The verifying frequency threshold value is 120,000 times, which is
95%；The testing time threshold value is 100,000 times, which is 99%.
Second aspect, the embodiment of the invention also provides a kind of object detection systems based on regression algorithm, comprising: sample
Database module, for obtaining device context image to be detected, and according to the device context picture construction sample to be detected
Database；Detection model establishes module, and for establishing the detection model of deep learning, the input of the detection model is sample data
Original image in library, the output of the detection model are defect image；Training module, for by the image in sample database into
Input detection model is iterated training after the processing of row grid dividing, until reaching preset frequency of training threshold value or reaching default
Training accuracy；Authentication module, for obtaining the image for having measurement equipment to be checked to form verifying collection, the figure that verifying is concentrated
The detection model that picture input after the completion of training after grid dividing processing is verified, until reaching preset verifying number threshold
It is worth or reaches preset verifying accuracy；Test module, for obtaining the image for having equipment deficiency to be detected to form test
Collection is tested the detection model that the image in test set input after the completion of verifying after grid dividing processing, until reaching
To preset testing time threshold value or reach preset test accuracy；Output module is detected, for inputting image to be detected
Detection model after the completion of test is detected, to export target defect image.
The embodiment of the present invention bring it is following the utility model has the advantages that
A kind of object detection method and system based on regression algorithm provided in an embodiment of the present invention, this method include obtaining
Device context image to be detected, and according to the device context picture construction sample database to be detected；Establish the inspection of deep learning
Model is surveyed, the input of the detection model is the original image in sample database, and the output of the detection model is defect image；It will
Input detection model is iterated training after image in the sample database carries out grid dividing processing, until reaching preset
Frequency of training threshold value reaches preset trained accuracy；The image with measurement equipment to be checked is obtained to form verifying collection, will be tested
The detection model that the image that card is concentrated input after the completion of training after grid dividing processing is verified, until reaching preset
Verifying frequency threshold value reaches preset verifying accuracy；Obtain with equipment deficiency to be detected image to form test set,
The detection model that image in test set input after the completion of verifying after grid dividing processing is tested, until reaching pre-
If testing time threshold value or reach preset test accuracy；Image to be detected is inputted into the detection model after the completion of the test
It is detected, to export target defect image.Object detection method provided in an embodiment of the present invention based on regression algorithm, passes through
It establishes iconic model and target defect detection is carried out to image, solve that in the prior art mass picture can not be carried out screening investigation
Technical problem, realization mass picture is detected, with intelligent screening target object, can improve target detection accuracy rate and
Timeliness.
Other feature and advantage of the disclosure will illustrate in the following description, alternatively, Partial Feature and advantage can be with
Deduce from specification or unambiguously determine, or by implement the disclosure above-mentioned technology it can be learnt that.
To enable the above objects, features, and advantages of the disclosure to be clearer and more comprehensible, preferred embodiment is cited below particularly, and cooperate
Appended attached drawing, is described in detail below.
Detailed description of the invention
It, below will be to specific in order to illustrate more clearly of the specific embodiment of the invention or technical solution in the prior art
Embodiment or attached drawing needed to be used in the description of the prior art be briefly described, it should be apparent that, it is described below
Attached drawing is some embodiments of the present invention, for those of ordinary skill in the art, before not making the creative labor
It puts, is also possible to obtain other drawings based on these drawings.
Fig. 1 is a kind of flow chart of the object detection method based on regression algorithm provided in an embodiment of the present invention；
Fig. 2 is the original image schematic diagram in a kind of sample database provided in an embodiment of the present invention；
Fig. 3 is a kind of image schematic diagram after image procossing provided in an embodiment of the present invention；
Fig. 4 is provided in an embodiment of the present invention a kind of by gridding treated image schematic diagram；
Fig. 5 is a kind of structural schematic diagram of the object detection system based on regression algorithm provided in an embodiment of the present invention.
Icon:
51- sample database establishes module；52- detection model establishes module；53- training module；54- authentication module；55-
Test module；56- detects output module.
Specific embodiment
In order to make the object, technical scheme and advantages of the embodiment of the invention clearer, below in conjunction with attached drawing to the present invention
Technical solution be clearly and completely described, it is clear that described embodiments are some of the embodiments of the present invention, rather than
Whole embodiments.Based on the embodiments of the present invention, those of ordinary skill in the art are not making creative work premise
Under every other embodiment obtained, shall fall within the protection scope of the present invention.
Currently, being gone forward side by side in the malfunction elimination to transmission line of electricity by the photo site that the mode of taking photo by plane obtains transmission facility
Row detection can alleviate the problem of needing personnel on site to reconnoitre, and still, for the field device image of acquisition, quantity is very huge
Greatly, it if by manually carrying out screening, needs to take considerable time and energy, also not can guarantee the accuracy and in real time of investigation
Property.Based on this, a kind of object detection method and system based on regression algorithm provided in an embodiment of the present invention can be with intelligent screening
Equipment deficiency image is checked out, the efficiency of equipment deficiency investigation is improved.
For convenient for understanding the present embodiment, first to a kind of based on regression algorithm disclosed in the embodiment of the present invention
Object detection method describes in detail.
Embodiment one:
It is a kind of flow chart of the object detection method based on regression algorithm provided in an embodiment of the present invention referring to Fig. 1, by
Fig. 1 as it can be seen that method includes the following steps:
Step S102: device context image to be detected is obtained, and according to the device context picture construction sample number to be detected
According to library.
Here, measurement equipment to be checked can be any equipment, including millwork equipment, transit equipment, transmission facility etc..
By taking transmission facility as an example, transmission facility live image be on transmission line of electricity each transmission facility in real time or the image of history, in image
Transmission facility be perpetual object, need to its working condition carry out malfunction elimination obtained in time in transmission facility existing defects
Know defect information.Wherein, transmission facility includes overhead line structures, conducting wire, insulator, line hardware, bracing wire, pole and tower foundation, ground connection
Device, breaker etc..Defect include damaged, foreign matter float extension, pollution, non-normal working etc. situations such as.
After obtaining device context image to be detected, in a kind of wherein embodiment, it can increase first with data
Strong technology carries out conversion process to the device context image to be detected, the image that obtains that treated, then, to be detected further according to this
Device context image and should treated that image constructs sample database jointly.In this way, sample database can be effectively increased
Data volume.
Technology is enhanced for above-mentioned data, may include rotation transformation, reflection transformation, turning-over changed, scale transformation, translation
One of transformation, change of scale and noise disturbance are a variety of.Specifically, the concrete mode of various transformation are as follows:
Rotation transformation or reflection transformation: by image Random-Rotation unspecified angle, change the direction of picture material；
It is turning-over changed: along horizontal or vertical direction flipped image；
Scale transformation: zoom in or out image according to a certain percentage；
Translation transformation: on the image plane translating image, and it is specified that random or artificially defined mode can be used
Range of translation and translating step, direction is translated horizontally or vertically, changes the position of picture material；
Change of scale: it to image according to preset scale factor, zooms in or out；Or using preset scale because
Son changes the size or fog-level of picture material to image filtering tectonic scale space；
Noise disturbance: random perturbation is carried out to each pixel RGB of image, common noise pattern is salt-pepper noise and height
This noise.
Step S104: establishing the detection model of deep learning, and the input of the detection model is original in sample database
Image, the output of the detection model are defect image.
The detection model is for detecting defect image, and input is the original image in sample database, and it is scarce for exporting
Fall into image.
Here, deep learning detection model can be based on convolution using R-CNN (Regions with CNN features)
The region method of neural network characteristics or quick R-CNN (Fast R-CNN) method.
Step S106: input detection model is iterated after the image in the sample database is carried out grid dividing processing
Training, until reaching preset frequency of training threshold value or reaching preset trained accuracy.
Here, the step of carrying out grid dividing processing to image is as follows:
Firstly, obtaining image to be processed.
Secondly, the image to be divided into the grid subgraph of N*N.Here, N is natural number, for example, 7*7 can be divided into
Grid subgraph.
Then, the target frame of multiple predictions is set for each grid subgraph.For example, can be by the target frame of prediction
Number is set as 2.
Finally, calculating the threshold value of each target frame, the lower target window of threshold value is deleted.
In a kind of wherein embodiment, which is 200,000 times, which is 90%.Work as detection
When model is iterated trained, while calculating it and training accuracy, namely the correctness of output defect image.
For the example above, when the detection model frequency of training reaches 200,000 times or it trains accuracy to reach 90%
When, training terminates.
Step S108: the image with measurement equipment to be checked is obtained to form verifying collection, the image that verifying is concentrated is subjected to net
The detection model inputted after the completion of training after lattice division processing is verified, until reaching preset verifying frequency threshold value or reaching
Preset verifying accuracy.
In a kind of wherein embodiment, which is 120,000 times, which is 95%.That is,
When detection model verifying number reaches 120,000 times or its verifying accuracy reaches 95%, verifying is completed.
Here, the image for verifying concentration can also be handled by above-mentioned data enhancing technology, also expanded verifying and concentrated
Data volume, to reinforce verifying.
Step S110: obtain with equipment deficiency to be detected image to form test set, by the image in test set into
Row grid dividing processing after input verifying after the completion of detection model tested, until reach preset testing time threshold value or
Reach preset test accuracy.
In a kind of wherein embodiment, which is 100,000 times, which is 99%.That is,
When the testing time of the detection model reaches 100,000 times or its test accuracy reaches 99%, test is completed.
Wherein, for the image in test set, technology can also be enhanced by above-mentioned data and is handled, also expand test
The data volume of concentration, to promote test effect.
Here, when detection model carries out above-mentioned repetitive exercise, verifying and test, which also passes through deep learning
Technology carries out target detection.
Specifically, the step of target detection, comprising:
Firstly, carrying out regional choice to image, and entire image is traversed, selection target region positions the position of target.This
In, in a kind of wherein embodiment, entire image is traversed using sliding window strategy.
Secondly, extracting the characteristic information of the target area.Here, the process for extracting characteristic information includes: to the target area
Domain carries out foundation characteristic extraction, which includes contour feature and color characteristic；The foundation characteristic of extraction is carried out more
Layer complex characteristic is extracted, which includes profile layered characteristic and gray level image feature；It is multiple to the multilayer of extraction
Miscellaneous feature carries out weight study, exports the biggish feature of weight, to predict output result.
Then, classified by classifier to this feature information.Wherein, classifier is SVM (Support Vector
Machine, support vector machines) classifier or Adaboost classifier.SVM method is by a Nonlinear Mapping p, sample
Space reflection is into a higher-dimension or even infinite dimensional feature space (space Hilbert), so that in original sample space
The problem of Nonlinear separability, is converted into the problem of linear separability in feature space.And Adaboost is a kind of iterative algorithm,
Its core concept is the classifier (Weak Classifier) different for the training of the same training set, then these weak classifier sets
Get up, constitutes a stronger final classification device (strong classifier).
Step S112: the detection model after the completion of image to be detected input test is detected, to export target defect
Image.
After detection model is completed to test, show that the detection model has reached the expection accuracy requirement of detection.This
When, the detection model after the completion of image to be detected input test is detected, to export target defect image.In this way, being obtained
The target defect image obtained meets accuracy requirement, as a result, believable.
In this way, can be detected automatically to the device context picture to be detected of magnanimity, lacking for measurement equipment to be checked is obtained
Image is fallen into, compared to artificial investigation, cost is lower, and faster, accuracy is higher for speed.
A kind of object detection method based on regression algorithm provided in an embodiment of the present invention, this method include that acquisition is to be detected
Device context image, and according to the device context picture construction sample database to be detected；The detection model of deep learning is established,
The input of the detection model is the original image in sample database, and the output of the detection model is defect image；By the sample
Input detection model is iterated training after image in database carries out grid dividing processing, until reaching preset training time
Number threshold value reaches preset trained accuracy；The image with measurement equipment to be checked is obtained to form verifying collection, verifying is concentrated
Image carry out grid dividing processing after input training after the completion of detection model verified, until reach it is preset verifying time
Number threshold value reaches preset verifying accuracy；The image with equipment deficiency to be detected is obtained to form test set, will be tested
The detection model that the image of concentration input after the completion of verifying after grid dividing processing is tested, until reaching preset survey
Examination frequency threshold value reaches preset test accuracy；The detection model that image to be detected inputs after the completion of the test is examined
It surveys, to export target defect image.This method carries out target defect detection to image by establishing iconic model, solves existing skill
The technical issues of can not carrying out screening investigation in art to mass picture, realization detects mass picture, can be with intelligent screening
Target object improves the accuracy rate and timeliness of target detection.
Embodiment two:
In order to be clearer to understand the object detection method provided by the above embodiment based on regression algorithm, the embodiment of the present invention
It is illustrated in conjunction with concrete application.
There is the image of Bird's Nest to be used as training set firstly, sorting out 164 from existing unmanned plane inspection photo, wherein 18
Opening has the image of Bird's Nest to be used as verifying collection, and 60 images are used as test set, wherein not having Bird's Nest in 2 images.
By the geometric transformation of image, enhance technology using one or more of data splitting to increase input data
Amount:
(1) rotation or reflection transformation (Rotation/reflection): by image Random-Rotation unspecified angle；Change figure
As the direction of content；
(2) turning-over changed (flip): along horizontal or vertical direction flipped image；
(3) scale transformation (zoom): zoom in or out image according to a certain percentage；
(4) translation transformation (shift): image is translated in a certain way on the image plane；It can be used random
Or artificially defined mode specifies range of translation and translating step, direction carries out translation change picture material horizontally or vertically
Position；
(5) it change of scale (scale): to image according to specified scale factor, zooms in or out；Or it combines special
Sign is extracted, and using specified scale factor to image filtering tectonic scale space, changes the size or fog-level of picture material；
(6) contrast variation (contrast): in the hsv color space of image, change saturation degree S and V luminance component, protect
Hold that tone H is constant, the S and V component to each pixel carry out exponent arithmetic, and exponential factor increases illumination and become between 0.25 to 4
Change；
(7) noise disturbance (noise): random perturbation is carried out to each pixel RGB of image, common noise pattern is green pepper
Salt noise and Gaussian noise；
(8) colour switching (color): PCA is carried out in the RGB color of training set pixel value, obtains the 3 of rgb space
A principal direction vector p1, p2, p3；And 3 eigenvalue λs 1, λ 2, λ 3.
In the present embodiment, using by 90,180,270 degree of stranded image rotation of method, increase by 3 times of sample set.
In the present embodiment, detection specific steps are carried out to target are as follows:
(1) regional choice.This step is positioned to the position of target.Since target possibly is present at any of image
Position, and the size of target, Aspect Ratio are not known yet, so the strategy of original adoption sliding window carries out entire image
Traversal, needs to be arranged different scales, different length-width ratios.Although the strategy of this exhaustion contains that target is all to be likely to occur
Position, but disadvantage is also evident from: time complexity is too high, and it is too many to generate redundancy window, seriously affects subsequent characteristics and mentions
The speed and performance for taking and classifying.The length-width ratio of actually due to by time complexity the problem of, sliding window is typically all
Fixed setting is several, so for the biggish multi-class target detection of length-width ratio floating, even sliding window is traversed also not
It can obtain good region.
(2) feature extraction.Due to the Morphological Diversity of target, illumination variation diversity, the factors such as background diversity make
It is not so easy for designing the feature of a robust, however the quality for extracting feature directly influences the accuracy of classification.This
A stage common feature has SIFT, HOG etc..
(3) classifier.The classifier that conventional target detection uses mainly has SVM, Adaboost etc..
Referring to Fig. 2 and Fig. 3, respectively the original image schematic diagram in sample database and the figure after image procossing
As schematic diagram, due to the image as sample set be derived from different places have different illumination conditions, different resolution ratio with
And different size etc., the characteristic of image are not quite similar.If raw video picture is directly sent into convolutional neural networks,
It certainly will influence whether feature extraction, or even influence the result of target detection.Therefore, neural network is input in image carry out convolution
Before operation, corresponding, effective processing is done to image, can achieve the effect for improving testing result accuracy.At present more
Common image processing method has image gray processing, binaryzation, normalization and data enhancing.
Binaryzation: color image captured by camera has tri- color channels of RGB, each pixel is corresponding with three
Value of a value between [0,255].And so-called gray processing is by the value in three channels by distributing different weights and being added,
It is set only to replace the value of original three color channels with a resulting brightness value is added.And binaryzation is then in gray processing
On the basis of image, by the way that corresponding threshold value is arranged, gray value is divided into two kinds and assigns 0 or 255 two value again.
Normalization: image normalization is that image is changed into system by the means of processing variation to treated image
The image of one form, such as can be uniform sizes by the skimble-scamble image normalization of size of different unmanned plane shootings is derived from.
The method helps speed up the pace of learning of neural network, makes its more rapid convergence.It can determine in an experiment depending on trained convergent
It is fixed whether need using.
Data enhancing: being used as support since deep learning needs huge sample size, otherwise may be due to training
Sample size is insufficient and the phenomenon that over-fitting occurs.Therefore, in order to avoid the generation of this phenomenon, in order to improve being applicable in for training result
Property, need to using concentrate limited image to be stretched original sample, overturn, mirror image, displacement etc. data enhancing method, from
And achieve the purpose that exptended sample collection.Customized data enhancement methods, such as random change background picture also can be used simultaneously
The method of element, partial region in selective erasing image, and use the method for noise filling, realization to the global information feature of image into
Row study, enhances robustness.
Referring to fig. 4, to be a kind of by gridding treated image schematic diagram, here, by the way that object detection task is turned
It changes a regression problem into, greatly accelerates the speed of detection, it can 45 images of processing per second.And since each network is pre-
Full figure information is used when surveying target window, so that false detection rate is greatly reduced.Specifically, including:
(1) input picture is obtained, divides an image into the grid of 7*7 first；
(2) for each grid, 2 frames are all predicted, including each frame be target confidence level and each frame
Probability of the region in multiple classifications；
(3) 7*7*2 target window can be predicted according to previous step, it is relatively low then to remove possibility according to threshold value
Target window removes redundancy window finally by NMS (Non-Maximum Suppression, non-maxima suppression).
Here it is possible to see that whole process is very simple, does not need intermediate candidate region (RegionProposal) and exist
Target is looked for, the judgement for just completing position and classification is directly returned.
Object detection method provided in an embodiment of the present invention based on regression algorithm, by establish iconic model to image into
The technical issues of row target defect detection, solution can not carry out screening investigation to mass picture in the prior art, thus to magnanimity
Picture is detected, intelligent screening investigation, and then realizes the accuracy rate for improving picture screening and the real-time for guaranteeing defect information.
Embodiment three:
The embodiment of the invention also provides a kind of object detection systems based on regression algorithm, referring to Fig. 5, for the system
Structural schematic diagram, as seen from Figure 5, the system include that the sample database being sequentially connected establishes module 51, detection model establishes mould
Block 52, training module 53, authentication module 54, test module 55 and detection output module 56.Wherein, the function of modules is such as
Under:
Sample database establishes module 51, for obtaining transmission facility live image, and according to the transmission facility scene photo
As building sample database；
Detection model establishes module 52, and for establishing the detection model of deep learning, the input of the detection model is sample
Original image in database, the output of the detection model are defect image；
Training module 53 is carried out for input detection model after the image in sample database is carried out grid dividing processing
Repetitive exercise, until reaching preset frequency of training threshold value or reaching preset trained accuracy；
Authentication module 54, for obtain have transmission facility image with formed verifying collection, by verifying concentrate image into
Row grid dividing processing after input training after the completion of detection model verified, until reach preset verifying frequency threshold value or
Reach preset verifying accuracy；
Test module 55, for obtaining the image for having transmission facility defect to form test set, by the figure in test set
The detection model that picture input after the completion of verifying after grid dividing processing is tested, until reaching preset testing time threshold
It is worth or reaches preset test accuracy；
Output module 56 is detected, for detecting the detection model after the completion of image to be detected input test, with defeated
Target defect image out.
The technology of object detection system based on regression algorithm provided by the embodiment of the present invention, realization principle and generation
Effect is identical with the aforementioned object detection method embodiment based on regression algorithm, and to briefly describe, Installation practice part is not mentioned
And place, it can refer to corresponding contents in preceding method embodiment.
It is apparent to those skilled in the art that for convenience and simplicity of description, the system of foregoing description
Specific work process, can refer to corresponding processes in the foregoing method embodiment, details are not described herein.
Unless specifically stated otherwise, the opposite step of the component and step that otherwise illustrate in these embodiments, digital table
It is not limit the scope of the invention up to formula and numerical value.
In all examples being illustrated and described herein, any occurrence should be construed as merely illustratively, without
It is as limitation, therefore, other examples of exemplary embodiment can have different values.
The flow chart and block diagram in the drawings show the system of multiple embodiments according to the present invention, method and computer journeys
The architecture, function and operation in the cards of sequence product.In this regard, each box in flowchart or block diagram can generation
A part of one module, section or code of table, a part of the module, section or code include one or more use
The executable instruction of the logic function as defined in realizing.It should also be noted that in some implementations as replacements, being marked in box
The function of note can also occur in a different order than that indicated in the drawings.For example, two continuous boxes can actually base
Originally it is performed in parallel, they can also be executed in the opposite order sometimes, and this depends on the function involved.It is also noted that
It is the combination of each box in block diagram and or flow chart and the box in block diagram and or flow chart, can uses and execute rule
The dedicated hardware based system of fixed function or movement is realized, or can use the group of specialized hardware and computer instruction
It closes to realize.
In addition, in the description of the embodiment of the present invention unless specifically defined or limited otherwise, term " installation ", " phase
Even ", " connection " shall be understood in a broad sense, for example, it may be being fixedly connected, may be a detachable connection, or be integrally connected；It can
To be mechanical connection, it is also possible to be electrically connected；It can be directly connected, can also can be indirectly connected through an intermediary
Connection inside two elements.For the ordinary skill in the art, above-mentioned term can be understood at this with concrete condition
Concrete meaning in invention.
In the description of the present invention, it should be noted that term " center ", "upper", "lower", "left", "right", "vertical",
The orientation or positional relationship of the instructions such as "horizontal", "inner", "outside" be based on the orientation or positional relationship shown in the drawings, merely to
Convenient for description the present invention and simplify description, rather than the device or element of indication or suggestion meaning must have a particular orientation,
It is constructed and operated in a specific orientation, therefore is not considered as limiting the invention.In addition, term " first ", " second ",
" third " is used for descriptive purposes only and cannot be understood as indicating or suggesting relative importance.
The computer program product of the object detection method based on regression algorithm is carried out provided by the embodiment of the present invention, is wrapped
The computer readable storage medium for storing the executable non-volatile program code of processor is included, what said program code included
Instruction can be used for executing previous methods method as described in the examples, and specific implementation can be found in embodiment of the method, no longer superfluous herein
It states.
It is apparent to those skilled in the art that for convenience and simplicity of description, the system of foregoing description,
The specific work process of device and unit, can refer to corresponding processes in the foregoing method embodiment, and details are not described herein.
In several embodiments provided herein, it should be understood that disclosed systems, devices and methods, it can be with
It realizes by another way.The apparatus embodiments described above are merely exemplary, for example, the division of the unit,
Only a kind of logical function partition, there may be another division manner in actual implementation, in another example, multiple units or components can
To combine or be desirably integrated into another system, or some features can be ignored or not executed.Another point, it is shown or beg for
The mutual coupling, direct-coupling or communication connection of opinion can be through some communication interfaces, device or unit it is indirect
Coupling or communication connection can be electrical property, mechanical or other forms.
The unit as illustrated by the separation member may or may not be physically separated, aobvious as unit
The component shown may or may not be physical unit, it can and it is in one place, or may be distributed over multiple
In network unit.It can select some or all of unit therein according to the actual needs to realize the mesh of this embodiment scheme
's.
It, can also be in addition, the functional units in various embodiments of the present invention may be integrated into one processing unit
It is that each unit physically exists alone, can also be integrated in one unit with two or more units.
It, can be with if the function is realized in the form of SFU software functional unit and when sold or used as an independent product
It is stored in the executable non-volatile computer-readable storage medium of a processor.Based on this understanding, of the invention
Technical solution substantially the part of the part that contributes to existing technology or the technical solution can be with software in other words
The form of product embodies, which is stored in a storage medium, including some instructions use so that
One computer equipment (can be personal computer, server or the network equipment etc.) executes each embodiment institute of the present invention
State all or part of the steps of method.And storage medium above-mentioned includes: USB flash disk, mobile hard disk, read-only memory (ROM, Read-
Only Memory), random access memory (RAM, Random Access Memory), magnetic or disk etc. are various can be with
Store the medium of program code.
Finally, it should be noted that embodiment described above, only a specific embodiment of the invention, to illustrate the present invention
Technical solution, rather than its limitations, scope of protection of the present invention is not limited thereto, although with reference to the foregoing embodiments to this hair
It is bright to be described in detail, those skilled in the art should understand that: anyone skilled in the art
In the technical scope disclosed by the present invention, it can still modify to technical solution documented by previous embodiment or can be light
It is readily conceivable that variation or equivalent replacement of some of the technical features；And these modifications, variation or replacement, do not make
The essence of corresponding technical solution is detached from the spirit and scope of technical solution of the embodiment of the present invention, should all cover in protection of the invention
Within the scope of.Therefore, protection scope of the present invention should be based on the protection scope of the described claims.
Claims (10)
1. a kind of object detection method based on regression algorithm characterized by comprising
Device context image to be detected is obtained, and according to the device context picture construction sample database to be detected；
The detection model of deep learning is established, the input of the detection model is the original image in the sample database, institute
The output for stating detection model is defect image；
It is iterated training by the detection model is inputted after the image progress grid dividing processing in the sample database, directly
To reaching preset frequency of training threshold value or reach preset trained accuracy；
The image with measurement equipment to be checked is obtained to form verifying collection, the image that the verifying is concentrated is subjected to grid dividing processing
The detection model inputted after the completion of the training afterwards is verified, until reaching preset verifying frequency threshold value or reaching preset
Verify accuracy；
The image with equipment deficiency to be detected is obtained to form test set, the image in the test set is subjected to grid dividing
The detection model inputted after the completion of the verifying after processing is tested, until reaching preset testing time threshold value or reaching pre-
If test accuracy；
The detection model that image to be detected inputs after the completion of the test is detected, to export target defect image.
2. the object detection method according to claim 1 based on regression algorithm, which is characterized in that described by the sample
Image lattice in database divides the step of handling
Obtain image to be processed；
Divide the image into the grid subgraph of N*N；
The target frame of multiple predictions is set for each grid subgraph；
The threshold value of each target frame is calculated, the lower target window of threshold value is deleted.
3. the object detection method according to claim 1 based on regression algorithm, which is characterized in that it is described according to
Detection device live image constructs the step of sample database, comprising:
Conversion process is carried out to the device context image to be detected using data enhancing technology, the image that obtains that treated；
According to the device context image to be detected and described treated that image constructs sample database jointly.
4. the object detection method according to claim 3 based on regression algorithm, which is characterized in that the data enhance skill
Art includes one of rotation transformation, reflection transformation, turning-over changed, scale transformation, translation transformation, change of scale and noise disturbance
Or it is a variety of.
5. the object detection method according to claim 1 based on regression algorithm, which is characterized in that in the detection model
When being iterated training, verifying and test, further includes:
Target detection is carried out by deep learning technology.
6. the object detection method according to claim 5 based on regression algorithm, which is characterized in that described by depth
Habit technology carries out the step of target detection, comprising:
Regional choice is carried out to image, and entire image is traversed using sliding window strategy, selection target region positions target
Position；
Extract the characteristic information of the target area；
Classified by classifier to the characteristic information.
7. the object detection method according to claim 6 based on regression algorithm, which is characterized in that described to extract the mesh
The step of marking the characteristic information in region, comprising:
Foundation characteristic extraction is carried out to the target area, the foundation characteristic includes contour feature and color characteristic；
The extraction of multilayer complex characteristic is carried out to the foundation characteristic of extraction, the multilayer complex characteristic includes profile layered characteristic
With gray level image feature；
Weight study is carried out to the multilayer complex characteristic of extraction, exports the biggish feature of weight, to predict output result.
8. the object detection method according to claim 6 based on regression algorithm, which is characterized in that the classifier is
SVM classifier or Adaboost classifier.
9. the object detection method according to claim 1 based on regression algorithm, which is characterized in that the frequency of training threshold
Value is 200,000 times, and the trained accuracy is 90%；
The verifying frequency threshold value is 120,000 times, and the verifying accuracy is 95%；
The testing time threshold value is 100,000 times, and the test accuracy is 99%.
10. a kind of object detection system based on regression algorithm characterized by comprising
Sample database establishes module, for obtaining device context image to be detected, and according to the device context figure to be detected
As building sample database；
Detection model establishes module, and for establishing the detection model of deep learning, the input of the detection model is the sample
Original image in database, the output of the detection model are defect image；
Training module, for image in the sample database is carried out input after grid dividing processing the detection model into
Row iteration training, until reaching preset frequency of training threshold value or reaching preset trained accuracy；
Authentication module, for obtain have measurement equipment to be checked image with formed verifying collection, by it is described verifying concentrate image into
The detection model that row grid dividing is inputted after handling after the completion of the training is verified, until reaching preset verifying number threshold
It is worth or reaches preset verifying accuracy；
Test module, for obtaining the image for having equipment deficiency to be detected to form test set, by the figure in the test set
As the detection model input after the completion of the verifying after grid dividing processing is tested, until reaching preset test time
Number threshold value reaches preset test accuracy；
Output module is detected, the detection model for inputting image to be detected after the completion of the test detects, with output
Target defect image.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811547328.XA CN109815800A (en) | 2018-12-17 | 2018-12-17 | Object detection method and system based on regression algorithm |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811547328.XA CN109815800A (en) | 2018-12-17 | 2018-12-17 | Object detection method and system based on regression algorithm |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109815800A true CN109815800A (en) | 2019-05-28 |
Family
ID=66601736
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811547328.XA Pending CN109815800A (en) | 2018-12-17 | 2018-12-17 | Object detection method and system based on regression algorithm |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109815800A (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110200619A (en) * | 2019-06-06 | 2019-09-06 | 中山大学孙逸仙纪念医院 | It is a kind of to establish major cardiovascular disease risks Early-warning Model method and device |
| CN110287822A (en) * | 2019-06-10 | 2019-09-27 | 浙江大学城市学院 | The snail pest control method of view-based access control model target detection in a kind of officinal dendrobium stem plantation |
| CN110503689A (en) * | 2019-08-30 | 2019-11-26 | 清华大学 | Attitude prediction method, model training method and device |
| CN110738644A (en) * | 2019-10-10 | 2020-01-31 | 清华大学 | automobile coating surface defect detection method and system based on deep learning |
| CN110807405A (en) * | 2019-10-29 | 2020-02-18 | 维沃移动通信有限公司 | Detection method of candid camera device and electronic equipment |
| CN112435210A (en) * | 2019-08-26 | 2021-03-02 | 银河水滴科技（北京）有限公司 | Quality detection method and device for annular device |
| CN116385336A (en) * | 2022-12-14 | 2023-07-04 | 广州市斯睿特智能科技有限公司 | Deep learning-based weld joint detection method, system, device and storage medium |
| CN117689258A (en) * | 2023-12-12 | 2024-03-12 | 广东电网有限责任公司广州供电局 | Knife switch performance evaluation supervision system and method based on machine vision |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107742093A (en) * | 2017-09-01 | 2018-02-27 | 国网山东省电力公司电力科学研究院 | A kind of infrared image power equipment component real-time detection method, server and system |
| WO2018053340A1 (en) * | 2016-09-15 | 2018-03-22 | Twitter, Inc. | Super resolution using a generative adversarial network |
| CN108805093A (en) * | 2018-06-19 | 2018-11-13 | 华南理工大学 | Escalator passenger based on deep learning falls down detection algorithm |
| CN108875595A (en) * | 2018-05-29 | 2018-11-23 | 重庆大学 | A kind of Driving Scene object detection method merged based on deep learning and multilayer feature |
-
2018
- 2018-12-17 CN CN201811547328.XA patent/CN109815800A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018053340A1 (en) * | 2016-09-15 | 2018-03-22 | Twitter, Inc. | Super resolution using a generative adversarial network |
| CN107742093A (en) * | 2017-09-01 | 2018-02-27 | 国网山东省电力公司电力科学研究院 | A kind of infrared image power equipment component real-time detection method, server and system |
| CN108875595A (en) * | 2018-05-29 | 2018-11-23 | 重庆大学 | A kind of Driving Scene object detection method merged based on deep learning and multilayer feature |
| CN108805093A (en) * | 2018-06-19 | 2018-11-13 | 华南理工大学 | Escalator passenger based on deep learning falls down detection algorithm |
Non-Patent Citations (1)
| Title |
|---|
| 刘士波: ""基于数字图像的输电线故障识别与定位方法研究"", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110200619A (en) * | 2019-06-06 | 2019-09-06 | 中山大学孙逸仙纪念医院 | It is a kind of to establish major cardiovascular disease risks Early-warning Model method and device |
| CN110287822A (en) * | 2019-06-10 | 2019-09-27 | 浙江大学城市学院 | The snail pest control method of view-based access control model target detection in a kind of officinal dendrobium stem plantation |
| CN112435210A (en) * | 2019-08-26 | 2021-03-02 | 银河水滴科技（北京）有限公司 | Quality detection method and device for annular device |
| CN112435210B (en) * | 2019-08-26 | 2024-03-01 | 银河水滴科技（北京）有限公司 | Quality detection method and device for annular device |
| CN110503689A (en) * | 2019-08-30 | 2019-11-26 | 清华大学 | Attitude prediction method, model training method and device |
| US11461925B2 (en) | 2019-08-30 | 2022-10-04 | Tsinghua University | Pose prediction method and apparatus, and model training method and apparatus |
| CN110738644A (en) * | 2019-10-10 | 2020-01-31 | 清华大学 | automobile coating surface defect detection method and system based on deep learning |
| CN110807405A (en) * | 2019-10-29 | 2020-02-18 | 维沃移动通信有限公司 | Detection method of candid camera device and electronic equipment |
| CN116385336A (en) * | 2022-12-14 | 2023-07-04 | 广州市斯睿特智能科技有限公司 | Deep learning-based weld joint detection method, system, device and storage medium |
| CN116385336B (en) * | 2022-12-14 | 2024-04-12 | 广州市斯睿特智能科技有限公司 | Deep learning-based weld joint detection method, system, device and storage medium |
| CN117689258A (en) * | 2023-12-12 | 2024-03-12 | 广东电网有限责任公司广州供电局 | Knife switch performance evaluation supervision system and method based on machine vision |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109815800A (en) | Object detection method and system based on regression algorithm | |
| CN112183788B (en) | Domain adaptive equipment operation detection system and method | |
| CN109829881A (en) | Bird's Nest detection method and system based on deep learning | |
| CN111784685B (en) | Power transmission line defect image identification method based on cloud edge cooperative detection | |
| CN112380952B (en) | Power equipment infrared image real-time detection and identification method based on artificial intelligence | |
| CN109785288A (en) | Transmission facility defect inspection method and system based on deep learning | |
| CN109815798A (en) | Unmanned plane image processing method and system | |
| Sohn et al. | Automatic powerline scene classification and reconstruction using airborne lidar data | |
| CN108596055B (en) | Airport target detection method of high-resolution remote sensing image under complex background | |
| CN109902806A (en) | Method is determined based on the noise image object boundary frame of convolutional neural networks | |
| CN110163213B (en) | Remote sensing image segmentation method based on disparity map and multi-scale depth network model | |
| CN108038846A (en) | Transmission line equipment image defect detection method and system based on multilayer convolutional neural networks | |
| CN108710913A (en) | A kind of switchgear presentation switch state automatic identification method based on deep learning | |
| CN110288586A (en) | A kind of multiple dimensioned transmission line of electricity defect inspection method based on visible images data | |
| CN112233073A (en) | Real-time detection method for infrared thermal imaging abnormity of power transformation equipment | |
| CN115731164A (en) | Insulator defect detection method based on improved YOLOv7 | |
| CN114743119B (en) | High-speed rail contact net hanger nut defect detection method based on unmanned aerial vehicle | |
| CN112560634B (en) | Power line insulator rapid detection and positioning method and system based on field image | |
| CN115294473A (en) | Insulator fault identification method and system based on target detection and instance segmentation | |
| CN113807450A (en) | Unmanned aerial vehicle power line patrol fault detection method based on ultrahigh resolution picture | |
| CN113569981A (en) | Power inspection bird nest detection method based on single-stage target detection network | |
| CN114187505A (en) | Detection method and device for falling-off of damper of power transmission line, medium and terminal equipment | |
| Kumar et al. | Feature relevance analysis for 3D point cloud classification using deep learning | |
| CN115937736A (en) | Small target detection method based on attention and context awareness | |
| Chen et al. | Insulator recognition method for distribution network overhead transmission lines based on modified yolov3 |
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: 20190528 |
|
| RJ01 | Rejection of invention patent application after publication |