CN101852768B - Workpiece flaw identification method based on compound characteristics in magnaflux environment - Google Patents

Workpiece flaw identification method based on compound characteristics in magnaflux environment Download PDF

Info

Publication number
CN101852768B
CN101852768B CN2010101629597A CN201010162959A CN101852768B CN 101852768 B CN101852768 B CN 101852768B CN 2010101629597 A CN2010101629597 A CN 2010101629597A CN 201010162959 A CN201010162959 A CN 201010162959A CN 101852768 B CN101852768 B CN 101852768B
Authority
CN
China
Prior art keywords
scar
suspicious
workpiece
image
zone
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.)
Expired - Fee Related
Application number
CN2010101629597A
Other languages
Chinese (zh)
Other versions
CN101852768A (en
Inventor
周景磊
叶茂
张旭东
赵欣
王波
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
University of Electronic Science and Technology of China
Original Assignee
University of Electronic Science and Technology of China
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by University of Electronic Science and Technology of China filed Critical University of Electronic Science and Technology of China
Priority to CN2010101629597A priority Critical patent/CN101852768B/en
Publication of CN101852768A publication Critical patent/CN101852768A/en
Application granted granted Critical
Publication of CN101852768B publication Critical patent/CN101852768B/en
Expired - Fee Related legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Abstract

The invention relates to a workpiece flaw identification method based on compound characteristics in a magnaflux powder inspection environment, which comprises the following steps of: firstly, preprocessing workpiece images collected in the magnaflux powder inspection environment in the process of flaw automatic detection for the workpiece images, wherein the image preprocessing process comprises an image filtering, image enhancement and image segmentation; after accomplishing the image segmentation, obtaining the foreground regions of the workpieces; performing image morphologic operation on the foreground regions of the workpieces to extract the suspicious flaw regions; performing a surface transverse flaw detection step, extracting the transverse flaw characteristics for the suspicious flaw regions by using a surface transverse flaw detection technology, automatically classifying the flaw characteristics of an SVW classifier obtained by automatically learning the transverse flaw sample set to determine whether the surface transverse flaws in the suspicious flaw regions exist or not. The invention has the advantages of low dependence to the detected objects and convenience for application, thereby realizing the workpiece flaw identification of various parts in the magnaflux powder inspection environment.

Description

Under the magnetic powder inspection environment based on the workpiece scar recognition methods of compound characteristics
Technical field
The invention belongs to the Automatic Measurement Technique field, specifically, relate to the workpiece nondestructive testing technique under a kind of magnetic powder inspection environment.
Background technology
Be accompanied by China's rapid development of economy; Market competition fierceness day by day; Before the user outclassed for the concern of product quality, various nondestructiving inspecting equipments were widely used among the manufacturing process of enterprise, but what form sharp contrast therewith is can't carry out automatic scar identification to defectogram picture or industrial flaw detection video that nondestructiving inspecting equipment obtains; A large amount of identification need of work related personnel accomplish; Caused waste of manpower resource, the lifting of cost, the increase of single article detection time and the instability of testing result.
The magnetic powder inspection technology is as a kind of industrial nondestructive testing technique important and that be widely used; Be that workpiece with magnetic material such as iron and steel magnetizes, have the workpiece of surface or near surface flaw to be magnetized after, when defective direction and magnetic direction are angled; Because the variation of the magnetic permeability of fault location; Magnetic line of force effusion surface of the work produces stray field, and the absorption magnetic forms the magnetic trace.Use the magnetic powder detection surface crack, with ultrasonic inspection and radiographic inspection relatively, it is highly sensitive, simple to operate, reliable results, good reproducibility, defective are recognized easily.But this method is only applicable to check the surface and the near surface flaw of ferrimagnet, and in the actual production operation, the magnetic trace identification of its surface of the work also is to be accomplished by the quality aufsichtsrat.Up to the present; The instrument of magnetic powder inspection on the market mainly provides the automatic control and the automatic imaging function of automatic workpiece magnetizing operation, complete machine, but the Automatic Measurement Technique to workpiece scar in the image under the magnetic powder inspection environment or the video of detecting a flaw is not provided.
In the prior art, Southern Yangtze University proposed " based on the bearing defect detecting technique of embedded machine vision " by name on April 13rd, 2009, and number of patent application is China's invention application of CN200910030476.9.This invention application provides a kind of bearing defect detecting technique based on embedded machine vision; The cut, the rust staining that need not manually-operated cut to bearing surface, rust staining, casting skin, outer lead angle, bearing bridge depression, the scarce pearl of bearing, scarce nail and bearing side detect; Successively detect filtration through substep, finally accomplish the filtration of defective bearing.This programme is when scarce nail that carries out bearing or the detection of scarce pearl; At first with circle localization method location area-of-interest; Area-of-interest is carried out spot density and the eccentricity that regions contract, binaryzation operation, region growing, morphological operation obtain area-of-interest; The threshold value that lacks nail, lacks pearl is set with this, and the threshold value of the accuracy of detection that obtains meeting through continuous loop test is used this threshold value to lack to follow closely or the detection of scarce pearl.This programme is when carrying out bearing surface cut, rust staining detection; At first arrive area-of-interest through the circle detection and localization; To area-of-interest carry out regions contract, binaryzation operation, grey level stretching, gray-scale value relatively obtain area-of-interest speck area and girth and with the number percent of annulus area; Be provided with correspondingly to the condition threshold value with this, the threshold value of the accuracy of detection that obtains meeting through continuous loop test uses this threshold value to carry out the bearing surface cut, rust staining detects.This programme is when carrying out the bearing concave surface defects detection; At first arrive area-of-interest through the circle detection and localization; Area-of-interest is carried out minimax distance ratio, area, girth, the centre distance that the operation of regions contract, binaryzation, morphological operation obtain the area-of-interest spot; With this minimax distance ratio, area, girth, centre distance threshold value are set, the threshold value of the accuracy of detection that obtains meeting through continuous loop test uses this threshold value to carry out the bearing concave surface defects detection.This programme is when carrying out bearing side cut, rust staining detection; Adopt independent positioning method to detect area-of-interest; The length and the broadband of area-of-interest are set, and the textural characteristics that obtains area-of-interest comprises average gray value, smoothness and consistance, and the threshold value of texture features parameter is set; The threshold value of the accuracy of detection that obtains meeting through continuous loop test uses this threshold value to carry out bearing side cut, rust staining detects.The bearing defect of this programme detects, and mainly bearing image texture features, gray feature, provincial characteristics is analyzed and is handled.
But the scheme of application of the present invention is mainly paid attention to the detection of particular type bearing scar, does not carry out cutting apart of foreground object, and is for the imaging circumstances complicated situation, not enough like its station-keeping ability of magnetic powder inspection environment.Its characteristic threshold value selected is the mode that adopts loop iteration, occurs the over-fitting phenomenon easily, do not have the autonomous learning mechanism of scar characteristic, and generalization and applicability are limited, so this programme has stronger dependence to testee, can not apply.
In the prior art, Institutes Of Technology Of Tianjin proposed " embedded high-speed on-line machine vision detection method and device " by name on November 17th, 2008, and number of patent application is China's invention application of CN200810153102.1.A kind of embedded high-speed on-line machine vision detection method that this invention application proposes; Techniqueflow is at first to be carried out to picture to the detected object on the production line; Cut apart the sensitizing range of locating easy quality defects; Extract the sensitive image provincial characteristics of easy quality defects, through comparing with standard feature, passing threshold judges whether detected object is qualified.This invention application is when carrying out the On-line Product detection; At first use industrial camera to gather the picture signal of detected object on the production line; Image to gathering carries out denoising, filtering and background compensation, and the image after handling is extracted the sensitizing range that easily goes wrong, and uses wavelet transformation that feature extraction is carried out in the sensitizing range; Characteristic and the standard feature extracted are compared, whether reach threshold value according to both similarity and judge whether detected object is qualified.
But the scheme of application of the present invention only is to use wavelet transformation to extract the frequency domain character of image, and the scar characteristic is single; Only be applicable to the detection of specific scar, the threshold value of detection does not have stability by artificial setting; And be unfavorable for promoting; Applicability is limited, lacks the study mechanism of scar characteristic, and human factor is bigger to the final detection result influence.Therefore, this scheme is not high to the accuracy of scar identification, can not further discern the kind of scar.
Summary of the invention
The objective of the invention is in order to overcome the deficiency of workpiece scar recognition methods under the existing magnetic powder inspection environment, the workpiece scar recognition methods based on compound characteristics is provided under a kind of magnetic powder inspection environment.The scar that this method can be directed to workpiece image or workpiece video respectively detects automatically.
To achieve these goals, the invention provides under a kind of magnetic powder inspection environment, may further comprise the steps based on the workpiece scar recognition methods of compound characteristics:
Step 1: workpiece image is being carried out in the automatic testing process of scar; At first carry out the image pre-service under the magnetic powder inspection environment, collecting workpiece image; The image preprocessing process comprises that image filter makes an uproar, schemes image intensifying and image segmentation; Image segmentation obtains the workpiece foreground area after accomplishing, and the foreground area of workpiece is carried out the morphological image operation, extracts suspicious scar zone;
Step 2: carry out laterally scar detection operation of surface; Utilize the horizontal scar detection technique in surface to the horizontal scar characteristic of suspicious scar extracted region; SVM (the SVMs that use obtains horizontal scar sample set autonomous learning; Support Vector Machines) sorter is classified to the scar characteristic automatically, confirms whether to exist in the suspicious scar zone laterally scar of surface;
Step 3: to not detecting the suspicious scar zone of scar in the step 2; Carry out the surface longitudinal scar and detect operation; Utilize surface longitudinal scar detection technique to the vertical scar characteristic of suspicious scar extracted region; Use is classified to the scar characteristic to the svm classifier device that vertical scar sample set autonomous learning obtains automatically, confirms whether there is the surface longitudinal scar in the suspicious scar zone;
Step 4: to not detecting the suspicious scar zone of scar in the step 3; Carry out surperficial lathe tool line scar and detect operation; Utilize surperficial lathe tool line scar detection technique to suspicious scar extracted region lathe tool line scar characteristic; Use is classified to the scar characteristic to the svm classifier device that lathe tool line scar sample set autonomous learning obtains automatically, confirms whether there is surperficial lathe tool line scar in the suspicious scar zone;
Step 5: carrying out radially to the workpiece cross section, scar detects operation; Utilize radially scar characteristic of suspicious scar extracted region that cross-section radial scar detection technique obtains the workpiece cross section; The svm classifier device that uses scar sample set autonomous learning radially to obtain is classified to the scar characteristic automatically, confirms whether there is the cross-section radial scar in the suspicious scar zone, workpiece cross section.
To collecting to such an extent that the magnetic powder inspection workpiece image carries out the process that image intensifying, image segmentation and suspicious scar extracted region were made an uproar, schemed in the image filter, comprise the steps: in the above-mentioned steps 1
Step 11: image is carried out medium filtering, the spiced salt noise that exists in the filtering image;
Step 12: the image to filtering carries out the gray-scale value stretching;
Step 13: image is carried out vertically and the horizontal edge extraction, obtain the external margin of workpiece according to the positional information of workpiece to be detected;
Step 14:, the external margin that obtains in the step 13 is carried out curve fitting like curvature, length breadth ratio, surface area according to the shape information of workpiece to be detected;
Step 15: the prospect mask according to the result of curve fitting in the step 14 forms workpiece, extract complete workpiece to be detected in the magnetic powder inspection image;
Step 16: the workpiece to be detected that extracts in the step 15 is carried out local binaryzation operation; Bianry image to obtaining carries out image filtering; The magnetic spot that filtering forms at surface of the work owing to spray magnetic keeps suspicious magnetic trace zone, as the suspicious scar zone of workpiece to be detected.
In the above-mentioned steps 2 to extract the suspicious scar zone of workpiece to be detected carry out the surface laterally scar detect operation, comprise the steps:
Step 21: the image overall binary conversion treatment is carried out in the suspicious scar zone to workpiece to be detected;
Step 22: the regional binary image of suspicious scar that obtains in the step 21 is carried out filtering noise and image smoothing operation;
Step 23: the area that calculates the doubtful scar in the suspicious scar zone in the bianry image that in step 22, obtains;
Step 24: with the bianry image that obtains in the step 23 is the prospect mask; Use Suo Beier operator (Sobeloperator) that the doubtful scar in the suspicious scar zone is carried out rim detection and extracts its gradient characteristic, comprise boundary length, gradient mean, gradient maximal value, gradient minimum value and the gradient variance of doubtful scar;
Step 25:, extract the boundary chain code statistical nature of the doubtful scar in the suspicious scar zone with the regional edge detection results of suspicious scar that obtains in the step 24;
Step 26: the bianry image to obtain in the step 22 is the prospect mask, extracts doubtful scar and the brightness ratio of background in the suspicious scar zone;
Step 27: HOG (the Histograms of OrientedGradients) characteristic of extracting the suspicious scar zone of workpiece to be detected;
Step 28: the scar characteristic that step 23 is extracted in the step 27 forms proper vector, uses the svm classifier device that horizontal scar sample set autonomous learning is obtained that proper vector is made the classification mark, judges the true and false of doubtful scar in the suspicious scar zone;
Step 29: the operation of step 23 to step 28 carried out in all the doubtful scar circulations in the suspicious scar zone, and all the doubtful scars in suspicious scar zone have all carried out the identification of horizontal scar.
In the above-mentioned steps 3 the surface longitudinal scar is carried out in the suspicious scar zone that does not detect horizontal scar and detect operation, comprise the steps:
Step 31: the image overall binary conversion treatment is carried out in the suspicious scar zone to workpiece to be detected;
Step 32: the regional binary image of suspicious scar that obtains in the step 31 is carried out filtering noise and image smoothing operation;
Step 33: with the bianry image that obtains in the step 32 is the prospect mask, uses the sobel operator that the doubtful scar in the suspicious scar zone is extracted its gradient characteristic, comprises gradient mean, gradient maximal value, gradient minimum value and the gradient variance of doubtful scar;
Step 34: carry out Fast Fourier Transform (FFT) to extracting suspicious scar zone, extract the frequency domain character in suspicious scar zone;
Step 35: carry out the horizontal direction projection to extracting suspicious scar zone, extract the average and the variance of the statistic histogram of suspicious scar zone level projection formation;
Step 36: the HOG characteristic of extracting the suspicious scar zone of workpiece to be detected;
Step 37: the scar characteristic that step 33 is extracted in the step 36 forms proper vector, uses the svm classifier device that vertical scar sample set autonomous learning is obtained that proper vector is made the classification mark, judges the true and false of doubtful scar in the suspicious scar zone;
Step 38: the operation of step 33 to step 37 carried out in all the doubtful scar circulations in the suspicious scar zone, and all the doubtful scars in suspicious scar zone have all carried out the identification of vertical scar.
In the above-mentioned steps 4 surperficial lathe tool line scar is carried out in the suspicious scar zone that does not detect vertical scar and detect operation, comprise the steps:
Step 41: the image overall binary conversion treatment is carried out in the suspicious scar zone to workpiece to be detected;
Step 42: the regional binary image of suspicious scar that obtains in the step 41 is carried out filtering noise and image smoothing operation;
Step 43: with the bianry image that obtains in the step 42 is the prospect mask, uses the sobel operator that the doubtful scar in the suspicious scar zone is extracted its gradient characteristic, comprises gradient mean, gradient maximal value, gradient minimum value and the gradient variance of doubtful scar;
Step 44: carry out Fast Fourier Transform (FFT) to extracting suspicious scar zone, extract the frequency domain character in suspicious scar zone;
Step 45: the horizontal direction projection is carried out in the suspicious scar zone to extracting, and extracts the average and the variance of the statistic histogram of suspicious scar zone level projection formation;
Step 46: use the Canny algorithm to extract the vertical border in suspicious scar zone, obtain the vertical boundary number in suspicious scar zone;
Step 47: extract the textural characteristics in suspicious scar zone, comprise angle second moment, moment of inertia, correlativity and entropy statistical nature;
Step 48: HOG feature description that extracts the suspicious scar zone of workpiece to be detected;
Step 49: the scar characteristic that step 43 is extracted in the step 48 forms proper vector, uses the svm classifier device that lathe tool line scar sample set autonomous learning is obtained that proper vector is made the classification mark, judges the true and false of doubtful scar in the suspicious scar zone;
Step 410: the operation of step 43 to step 49 carried out in all the doubtful scar circulations in the suspicious scar zone, and all the doubtful scars in suspicious scar zone have all carried out the identification of lathe tool line scar.
To the detection operation of workpiece cross-section radial scar, comprise the steps: in the above-mentioned steps 5
Step 51: image is carried out medium filtering, the spiced salt noise that exists in the filtering image;
Step 52: the image to filtering carries out the gray-scale value stretching;
Step 53: the image after step 52 processing is carried out local binaryzation operation earlier, then bianry image is carried out morphology and expand, obtain the maximum connected domain of region area as the prospect mask that extracts the workpiece cross section;
Step 54: the prospect mask by obtaining in the step 53 obtains the workpiece cross section, and fixing transverse observation window is set, and watch window is less than the workpiece cross section, and is by the workpiece cross section that the degree rotation is obtained, that the workpiece cross section that gets into watch window is regional as suspicious scar;
Step 55: image binaryzation is carried out in the suspicious scar zone in workpiece to be detected cross section handle;
Step 56: the area that calculates the doubtful scar in the suspicious scar zone in the bianry image that in step 55, obtains;
Step 57: with the bianry image that obtains in the step 55 is the prospect mask; Use the sobel operator that the doubtful scar in the suspicious scar zone is carried out rim detection and extracts its gradient characteristic, comprise boundary length, gradient mean, gradient maximal value, gradient minimum value and the gradient variance of doubtful scar;
Step 58: the bianry image to obtain in the step 55 is the prospect mask, extracts the boundary chain code statistical nature of the doubtful scar in the suspicious scar zone;
Step 59: the bianry image to obtain in the step 55 is the prospect mask, extracts doubtful scar and the brightness ratio of background in the suspicious scar zone;
Step 510: HOG feature description that extracts the suspicious scar zone of workpiece to be detected;
Step 511: the scar characteristic that step 56 is extracted in the step 511 forms proper vector, use pair cross-section radially the svm classifier device that obtains of scar sample set autonomous learning proper vector is made the classification mark, judge the true and false of doubtful scar in the suspicious scar zone;
Step 512: the operation of step 56 to step 512 carried out in all the doubtful scar circulations in the suspicious scar zone, and all the doubtful scars in suspicious scar zone have all carried out the identification of cross-section radial scar;
Step 513: the watch window that rotates is carried out the operation of step 55 to step 513, accomplished the rotation in a week up to the workpiece cross-sectional image.
Above-mentioned steps 2 detects operation, surface longitudinal scar at the horizontal scar in said surface and detects operation, surperficial lathe tool line scar and detect the training that operation and cross-section radial scar detect the svm classifier device that obtains based on specific scar sample set autonomous learning that uses in the operation and comprise the steps: in step 5
Step 61: specific scar highlight regions is carried out the image pattern collection, form the to be selected positive sample set of the specific scar svm classifier device of training;
Step 62: pseudo-scar highlight regions is carried out the image pattern collection, form the negative sample collection to be selected of the specific scar svm classifier device of training;
Step 63: the magnetic trace highlight regions of acquisition testing workpiece (comprising specific scar highlight regions and pseudo-scar highlight regions) forms the test set of specific scar svm classifier device;
Step 64: positive sample set of initial training and initial training negative sample collection extract a sample formation from positive sample set to be selected and negative sample collection to be selected at random;
Step 65: align sample set and negative sample collection and extract specific scar characteristic, form the characteristic data set of training svm classifier device, use sample class mark and characteristic set pair svm classifier device to train, generate initial svm classifier device to specific scar;
Step 66: the initial svm classifier device of use test set pair is tested, and obtains the initial detecting rate;
Step 67: extract a new samples from positive sample set to be selected and negative sample collection to be selected at random and expand positive sample set of training and training negative sample collection, the operation in carry out step 65 generates new svm classifier device;
Step 68: the svm classifier device that the use test set pair is new is tested, and accepts the positive sample set of training that expands if verification and measurement ratio improves and trains the negative sample collection, otherwise reject newly-increased sample, carries out the picked at random exptended sample again;
Step 69: step 67 and step 68 are carried out in circulation, reach actual requirement or converge to a certain than high measurement accuracy up to final svm classifier device accuracy of detection.
In order to realize the object of the invention, the workpiece scar recognition methods based on compound characteristics that is directed against industrial video under the improved magnetic powder inspection environment also further is provided, may further comprise the steps:
Step S1: under the magnetic powder inspection environment workpiece flaw detection video is carried out scar and detect automatically; At first be to read the workpiece video by frame; Upgrade magnetic testing environment background, undergo mutation to confirm leaving of the arrival of new workpiece and workpiece that detection finishes through detection background;
Step S2: for the workpiece that is detecting, each the frame workpiece image that reads is carried out scar detect, comprise step:
Step S21: workpiece image is being carried out in the automatic testing process of scar; At first carry out image filter and make an uproar and scheme image intensifying under the magnetic powder inspection environment, collecting workpiece image; Carry out image segmentation in conjunction with prior imformations such as the shape of workpiece, positions; Obtain the workpiece foreground area, the foreground area of workpiece is carried out the morphological image operation, extract suspicious scar zone;
Step S22: carry out laterally scar detection operation of surface; Utilize the horizontal scar detection technique in surface to the horizontal scar characteristic of suspicious scar extracted region; Use is classified to the scar characteristic to the svm classifier device that horizontal scar sample set autonomous learning obtains automatically, confirms whether to exist in the suspicious scar zone laterally scar of surface;
Step S23: to not detecting the suspicious scar zone of scar among the step S22; Carry out the surface longitudinal scar and detect operation; Utilize surface longitudinal scar detection technique to the vertical scar characteristic of suspicious scar extracted region; Use is classified to the scar characteristic to the svm classifier device that vertical scar sample set autonomous learning obtains automatically, confirms whether there is the surface longitudinal scar in the suspicious scar zone;
Step S24: to not detecting the suspicious scar zone of scar among the step S23; Carry out surperficial lathe tool line scar and detect operation; Utilize surperficial lathe tool line scar detection technique to suspicious scar extracted region lathe tool line scar characteristic; Use is classified to the scar characteristic to the svm classifier device that lathe tool line scar sample set autonomous learning obtains automatically, confirms whether there is surperficial lathe tool line scar in the suspicious scar zone;
Step S25: carrying out radially to the workpiece cross section, scar detects operation; Utilize radially scar characteristic of suspicious scar extracted region that cross-section radial scar detection technique obtains the workpiece cross section; The svm classifier device that uses scar sample set autonomous learning radially to obtain is classified to the scar characteristic automatically, confirms whether there is the cross-section radial scar in the suspicious scar zone, workpiece cross section;
Step S3: the workpiece video is being carried out in the automatic scar testing process; Same workpiece video will form an image sequence; The Multi-SVM sorter that is used to complete training carries out the judgement of the scar true and false according to the testing result of step S2 to image sequence, confirms finally whether this workpiece exists scar.
The training process of the Multi-SVM sorter that uses among the above-mentioned steps S3 comprises the steps:
Step S31: all types of scar highlight regions are carried out the image pattern collection, form the positive sample set of the single characteristic svm classifier device of training;
Step S32: pseudo-scar highlight regions is carried out the image pattern collection, form the negative sample collection of the single characteristic svm classifier device of training;
Step S33: the magnetic trace highlight regions of acquisition testing workpiece forms the test set of single characteristic svm classifier device;
Step S34: align sample set and negative sample collection and extract a certain scar and describe characteristic, form the data set of the single characteristic svm classifier device of training.Use positive sample set and negative sample set pair list characteristic svm classifier device to train, generate single characteristic svm classifier device of describing characteristic to certain scar;
Step S35: single characteristic svm classifier device that the use test set pair generates is tested, if verification and measurement ratio is higher than 0.5, then calculates the weight of this list characteristic svm classifier device, otherwise should the zero setting of list characteristic svm classifier device weight;
Step S36: circulation execution in step S34 describes characteristic up to all scars and has generated corresponding single characteristic svm classifier device and weight to step S35;
Step S37: all single characteristic svm classifier devices of positive sample set test that use single characteristic svm classifier device; To the sample in the positive sample set of each single characteristic svm classifier device; With used single characteristic svm classifier device the respective weights addition is multiply by in the response of this sample and the value that obtains as a sample in the positive sample set of training Multi-SVM sorter; Use all single characteristic svm classifier devices of negative sample collection test of single characteristic svm classifier device; The sample of concentrating to the negative sample of each single characteristic svm classifier device multiply by the respective weights addition to the response of this sample and a sample that the value that obtains is concentrated as the negative sample of training Multi-SVM sorter with used single characteristic svm classifier device;
Step S38: use the positive sample set of Multi-SVM sorter and negative sample training to practice generation Multi-SVM sorter.
The invention has the beneficial effects as follows: in order to satisfy the demand of the automatic detection of scar under the magnetic powder inspection environment; This programme proposes its compound characteristics of feature extraction according to the workpiece scar; The gradient characteristic (like scar region gradient statistic), textural characteristics (like the statistic of the regional gray level co-occurrence matrixes of scar), gray feature (like scar zone and background luminance ratio), edge feature (like edge chain code and edge direction), provincial characteristics (like the area and the girth of scar), the frequency domain character (like the wavelet transformation characteristic statistic in scar zone) that comprise scar; From multi-angle the workpiece scar is described; Through SVMs (SVM; Support vector machines) the scar characteristic is carried out autonomous learning; The svm classifier device that utilizes training to accomplish is discerned to the specific scar image of each type, and the multiclass scar that under unified testing process, same workpiece is walked abreast detects.Therefore the scheme of the relative prior art of this programme has following advantage: 1. the dependence of pair testee is low, is convenient to apply, and realizes the workpiece scar identification of various parts under the magnetic powder inspection environment.2. this scheme is high to the accuracy of scar identification, can further discern the kind of scar, the single product examine that the has improved workpiece rate that tests the speed.3. the present technique scheme detects except carrying out scar to image, and can carry out the scar detection of image sequence to the industrial flaw detection video, has improved the industrial automation level.
Description of drawings
Fig. 1 is the workpiece scar recognition methods main flow chart of the embodiment of the invention 1.
Fig. 2 is the process flow diagram of the image pretreatment process of the embodiment of the invention 1.
Fig. 3 is that the horizontal scar in the surface of the embodiment of the invention 1 detects process flow chart
Fig. 4 is that the surface longitudinal scar of the embodiment of the invention 1 detects process flow chart
Fig. 5 is that the surperficial lathe tool line scar of the embodiment of the invention 1 detects process flow chart
Fig. 6 is that the cross-section radial scar of the embodiment of the invention 1 detects process flow chart
Fig. 7 is the svm classifier device training process flow diagram of the embodiment of the invention 1
Fig. 8 is the workpiece scar recognition methods main flow chart of the embodiment of the invention 2.
Fig. 9 is the Multi-SVM sorter training process flow diagram of the embodiment of the invention 2
Embodiment
Below in conjunction with accompanying drawing and concrete specific embodiment the present invention is done further explanation:
Embodiment 1: as shown in Figure 1, that describes among the figure detects process flow chart automatically for scar in the workpiece image under the magnetic powder inspection environment.This flow chart description under a kind of magnetic powder inspection environment based on the workpiece scar recognition methods of compound characteristics, may further comprise the steps:
Step 1: workpiece image is being carried out in the automatic testing process of scar; At first carry out the image pre-service under the magnetic powder inspection environment, collecting workpiece image; The image preprocessing process comprises that image filter makes an uproar, schemes image intensifying and image segmentation; Image segmentation obtains the workpiece foreground area after accomplishing, and the foreground area of workpiece is carried out the morphological image operation, extracts suspicious scar zone.Because the magnetic powder inspection environment can receive a lot of influence of environmental noise in imaging process and because in order to improve the effect of magnetic powder inspection; Whole flaw detection operating process is among the camera bellows; Therefore collect to such an extent that mostly workpiece image is to receive noise pollution and integral color is dark partially; Can suppress the noise of image through this step, help the scar Feature Extraction, thereby strengthen the accuracy of identification of scar.Because the sprinkling of magnetic is not very accurately with even in the magnetic powder inspection process; For fear of the influence of workpiece background and workpiece sensing operator's console for the scar feature extraction; Thereby through foreground segmentation, make the automatic testing process of whole scar be directed to the prospect workpiece and operate, in this step except accomplishing foreground segmentation; According to stretch forward technical characterstic of magnetic itself; The real-time that scar detects automatically improves because the substep uneven magnetic trace that can form high brightness in magnetic field can be confirmed some suspicious scar zones through these characteristics in the prospect workpiece in the scar place of workpiece to be detected.
Step 2: carry out laterally scar detection operation of surface; Utilize the horizontal scar detection technique in surface to the horizontal scar characteristic of suspicious scar extracted region; SVM (the SVMs that use obtains horizontal scar sample set autonomous learning; Support Vector Machines) sorter is classified to the scar characteristic automatically, confirms whether to exist in the suspicious scar zone laterally scar of surface.
Step 3: to not detecting the suspicious scar zone of scar in the step 2; Carry out the surface longitudinal scar and detect operation; Utilize surface longitudinal scar detection technique to the vertical scar characteristic of suspicious scar extracted region; Use is classified to the scar characteristic to the svm classifier device that vertical scar sample set autonomous learning obtains automatically, confirms whether there is the surface longitudinal scar in the suspicious scar zone.
Step 4: to not detecting the suspicious scar zone of scar in the step 3; Carry out surperficial lathe tool line scar and detect operation; Utilize surperficial lathe tool line scar detection technique to suspicious scar extracted region lathe tool line scar characteristic; Use is classified to the scar characteristic to the svm classifier device that lathe tool line scar sample set autonomous learning obtains automatically, confirms whether there is surperficial lathe tool line scar in the suspicious scar zone.
Step 5: carrying out radially to the workpiece cross section, scar detects operation; Utilize radially scar characteristic of suspicious scar extracted region that cross-section radial scar detection technique obtains the workpiece cross section; The svm classifier device that uses scar sample set autonomous learning radially to obtain is classified to the scar characteristic automatically, confirms whether there is the cross-section radial scar in the suspicious scar zone, workpiece cross section.
Above-mentioned steps 2,3,4 and 5 all be to extract in the step 1 suspicious scar zone extract respectively laterally, the characteristic of vertical, lathe tool line and cross-section radial scar; Use the sorter of the specific scar of training in advance success; Like horizontal scar svm classifier device, vertical scar svm classifier device, lathe tool line scar svm classifier device and cross-section radial scar svm classifier device; To extracting to such an extent that the regional characteristic of suspicious scar is classified, distinguish in the suspicious scar zone whether have true scar with this.
As shown in Figure 2, what describe among the figure is workpiece image pretreatment process process flow diagram under the magnetic powder inspection environment.
Step 11. pair image carries out medium filtering, the spiced salt noise that exists in the filtering image; Magnetic powder inspection image to collecting carries out medium filtering, can the filtering image acquisition process in because the salt-pepper noise that the testing environment instability causes.
The image of step 12. pair filtering carries out gray-scale value and stretches.Image is carried out gray-scale value stretches since in the practical operation of magnetic powder inspection in order to improve the scar visibility, be in the half-light state in the workpiece image gatherer process always, help the detection in suspicious scar zone through the contrast that increases image.
Step 13. pair image carries out vertically and horizontal edge extracts, and obtains the external margin of workpiece according to the positional information of workpiece to be detected.Carry out rim detection at the workpiece image that collects; Because what use in the magnetic powder inspection process is mechanically actuated and IMAQ point relative fixed always; Therefore the picture position relative fixed of workpiece according to the big or small setting search window of workpiece, obtains the exterior contour edge of workpiece.
Step 14. carries out curve fitting to the external margin that obtains in the step 13 like curvature, length breadth ratio, surface area according to the shape information of workpiece to be detected.Exterior contour edge to detected workpiece in the step 13 carries out curve fitting, and obtains the complete edge of work.Because the edge curvature of one type of workpiece generally is fixed value, therefore in the process of carrying out the boundary curve match, can confirm the curvature of workpiece in advance.
Step 15. is extracted complete workpiece to be detected according to the prospect mask that the result of curve fitting in the step 14 forms workpiece in the magnetic powder inspection image.Utilize the complete edge of work that step 14 obtains prospect mask, from collecting to such an extent that obtain the prospect workpiece the magnetic powder inspection image as workpiece.
The workpiece to be detected that extracts in the step 16. pair step 15 carries out local binaryzation operation; Bianry image to obtaining carries out image filtering; The magnetic spot that filtering forms at surface of the work owing to spray magnetic keeps suspicious magnetic trace zone, as the suspicious scar zone of workpiece to be detected.Because magnetic powder inspection technology is if exist scar according to these characteristics the workpiece foreground segmentation in the step 15 to be become a series of suspicious scar zones that have the magnetic trace because the disturbance of the magnetic line of force forms the magnetic trace on the surface of the work.
As shown in Figure 3, what describe among the figure is that the horizontal scar in workpiece image surface detects process flow chart under the magnetic powder inspection environment.After the workpiece image of gathering is finished the image pre-service, can carry out the feature extraction of horizontal scar to extracting suspicious scar zone, and then accomplish automatic detection horizontal scar, comprise the steps:
The image overall binary conversion treatment is carried out in the suspicious scar zone of step 21. pair workpiece to be detected.The result images of its binaryzation will be used for the extraction of scar area features.
The binary image in the suspicious scar zone that obtains in the step 22. pair step 21 carries out filtering noise and image smoothing operation.Because except the magnetic trace of doubtful scar, also there are some a spot of inhomogeneous magnetics in suspicious scar zone, these magnetics can impact the scar Feature Extraction, therefore make an uproar through the image filter and eliminate the influence of the magnetic that is scattered with smooth operation.
Calculate the area of the doubtful scar in the suspicious scar zone in the bianry image that step 23. obtains in step 22.The binary image that obtains with step 22 calculates the area in the bright scar of doubtful height zone as the prospect mask in suspicious scar zone, as one of characteristic of the horizontal scar in surface.
Step 24. is the prospect mask with the bianry image that obtains in the step 23; Use Suo Beier operator (Sobeloperator) that the doubtful scar in the suspicious scar zone is carried out rim detection and extracts its gradient characteristic, comprise boundary length, gradient mean, gradient maximal value, gradient minimum value and the gradient variance of doubtful scar.And with boundary length, gradient mean, gradient maximal value, gradient minimum value and the gradient variance of scar in the suspicious scar zone as one of characteristic of the horizontal scar in surface.
Step 25. is extracted the boundary chain code statistical nature of the doubtful scar in the suspicious scar zone with the edge detection results in the suspicious scar zone that obtains in the step 24.To the edge in suspicious scar zone use from all directions to chain code describe, with the chain code value at the edge in suspicious scar zone as one of characteristic of the horizontal scar in surface.
Step 26. is the prospect mask with the bianry image that obtains in the step 22, and the brightness ratio of extracting doubtful scar and background in the suspicious scar zone is as one of characteristic of the horizontal scar in surface.
Step 27. is extracted HOG (the Histograms of OrientedGradients) characteristic in the suspicious scar zone of workpiece to be detected.The HOG characteristic in suspicious scar zone; In fact based on the statistic of partial gradient characteristic; Initial HOG characteristic is applied to pedestrian detection, in order to improve the applicability of HOG characteristic to workpiece scar in the magnetic powder inspection, when extracting the HOG characteristic, has enlarged the Local Search window; Make the HOG characteristic be more suitable for representing the direction of workpiece scar, with extract the HOG characteristic in suspicious scar zone as one of characteristic of the horizontal scar in surface.
The scar characteristic that step 28. is extracted step 23 in the step 27 forms proper vector, uses the svm classifier device that horizontal scar sample set autonomous learning is obtained that proper vector is made the classification mark, judges the true and false of doubtful scar in the suspicious scar zone;
The operation of step 23 to step 28 carried out in the doubtful scar circulation of in the step 29. pair suspicious scar zone all, and all the doubtful scars in suspicious scar zone have all carried out the identification of horizontal scar.
As shown in Figure 4, what describe among the figure is that workpiece image surface longitudinal scar detects process flow chart under the magnetic powder inspection environment.After the horizontal scar of gathering in workpiece image surface is detected operation, can carry out the feature extraction of vertical scar to the suspicious scar zone of extracting, and then accomplish automatic detection vertical scar, comprise the steps:
The image overall binary conversion treatment is carried out in the suspicious scar zone of step 31. pair workpiece to be detected.
The binary image in the suspicious scar zone that obtains in the step 32. pair step 31 carries out filtering noise and image smoothing operation.
Step 33. is the prospect mask with the bianry image that obtains in the step 32, uses the sobel operator that the doubtful scar in the suspicious scar zone is extracted its gradient characteristic, comprises gradient mean, gradient maximal value, gradient minimum value and the gradient variance of doubtful scar.
Step 34. pair is extracted suspicious scar zone and is carried out Fast Fourier Transform (FFT), and the frequency domain character that extracts suspicious scar zone is as one of characteristic of surface longitudinal scar.
The horizontal direction projection is carried out in the step 35. pair suspicious scar of extraction zone, extracts the average and the variance of the statistic histogram of suspicious scar zone level projection formation.Horizontal projection is carried out in suspicious scar zone, if having vertical scar in the suspicious scar zone, then the histogram that forms of projection operation will obvious peak, therefore with histogrammic statistic as one of surface longitudinal scar characteristic.
Step 36. is extracted the HOG characteristic in the suspicious scar zone of workpiece to be detected.
The scar characteristic that step 37. is extracted step 33 in the step 36 forms proper vector, uses the svm classifier device that vertical scar sample set autonomous learning is obtained that proper vector is made the classification mark, judges the true and false of doubtful scar in the suspicious scar zone.
The operation of step 33 to step 37 carried out in the doubtful scar circulation of in the step 38. pair suspicious scar zone all, and all the doubtful scars in suspicious scar zone have all carried out the identification of vertical scar.
Above-mentioned workpiece image surface longitudinal scar detects operation, and except that step 34 and step 35, the principle of all the other steps is identical with the step that the horizontal scar in workpiece image surface detects operation, therefore is not described in detail.
As shown in Figure 5, what describe among the figure is that workpiece image surface lathe tool line scar detects process flow chart under the magnetic powder inspection environment.After the workpiece image surface longitudinal scar of gathering is detected operation, can carry out the feature extraction of lathe tool line scar to the suspicious scar zone of extracting, and then accomplish automatic detection lathe tool line scar, comprise the steps:
The image overall binary conversion treatment is carried out in the suspicious scar zone of step 41. pair workpiece to be detected.
The binary image in the suspicious scar zone that obtains in the step 42. pair step 41 carries out filtering noise and image smoothing operation.
Step 43. is the prospect mask with the bianry image that obtains in the step 42, uses the sobel operator that the doubtful scar in the suspicious scar zone is extracted its gradient characteristic, comprises gradient mean, gradient maximal value, gradient minimum value and the gradient variance of doubtful scar.
Fast Fourier Transform (FFT) is carried out in the step 44. pair suspicious scar of extraction zone, extracts the frequency domain character in suspicious scar zone.
The horizontal direction projection is carried out in the suspicious scar zone of step 45. pair extraction, extracts the average and the variance of the statistic histogram of suspicious scar zone level projection formation.
Step 46. uses the Canny algorithm to extract the vertical border in suspicious scar zone, and the vertical boundary number of obtaining suspicious scar zone is as one of characteristic of surperficial lathe tool line scar.
Step 47. is extracted the textural characteristics in suspicious scar zone, comprises angle second moment, moment of inertia, correlativity and entropy statistical nature.Through obtaining the gray level co-occurrence matrixes in suspicious scar zone, so through gray level co-occurrence matrixes calculate as angle second moment, moment of inertia, correlativity and entropy or the like statistical nature as the regional textural characteristics of suspicious scar, be used for the automatic detection of the scar of lathe tool line.
Step 48. is extracted the HOG characteristic in the suspicious scar zone of workpiece to be detected.
The scar characteristic that step 49. is extracted step 43 in the step 48 forms proper vector, uses the svm classifier device that lathe tool line scar sample set autonomous learning is obtained that proper vector is made the classification mark, judges the true and false of doubtful scar in the suspicious scar zone.
The operation of step 43 to step 49 carried out in the doubtful scar circulation of in the step 410. pair suspicious scar zone all, and all the doubtful scars in suspicious scar zone have all carried out the identification of lathe tool line scar.
Above-mentioned workpiece image surface lathe tool line scar detects operation, and except that step 46 and step 47, the principle of all the other steps is identical with the step that workpiece image surface longitudinal scar detects operation, therefore is not described in detail.
Detection process flow chart for workpiece cross-section radial scar under the magnetic powder inspection environment as shown in Figure 6, as to describe among the figure.This operation comprises the steps:
Step 51. pair image carries out medium filtering, the spiced salt noise that exists in the filtering image.
The image of step 52. pair filtering carries out gray-scale value and stretches.
Image after step 53. pair step 52 is handled carries out local binaryzation operation earlier, then bianry image is carried out morphology and expands, and obtains the maximum connected domain of region area as the prospect mask that extracts the workpiece cross section.Because a large amount of magnetics can be sprayed in the workpiece cross section when carrying out magnetic powder inspection; And magnetic can form local high-brightness region in the workpiece cross-sectional image that finally collects; Therefore use the method for local binaryzation; Can be that the plurality of discrete speck appears in the workpiece cross section, and these specks can connect together and formed the prospect mask in workpiece cross section after morphology expands.
Step 54. is obtained the workpiece cross section by the prospect mask that obtains in the step 53, and fixing transverse observation window is set, and watch window is less than the workpiece cross section, and is by the workpiece cross section that the degree rotation is obtained, that the workpiece cross section that gets into watch window is regional as suspicious scar.
The suspicious scar zone in step 55. pair workpiece to be detected cross section is carried out image binaryzation and is handled.
Calculate the area of the doubtful scar in the suspicious scar zone in the bianry image that step 56. obtains in step 55.
Step 57. is the prospect mask with the bianry image that obtains in the step 55; Use the sobel operator that the doubtful scar in the suspicious scar zone is carried out rim detection and extracts its gradient characteristic, comprise boundary length, gradient mean, gradient maximal value, gradient minimum value and the gradient variance of doubtful scar.
Step 58. is the prospect mask with the bianry image that obtains in the step 55, extracts the boundary chain code statistical nature of the doubtful scar in the suspicious scar zone.
Step 59. is the prospect mask with the bianry image that obtains in the step 55, extracts doubtful scar and the brightness ratio of background in the suspicious scar zone.
Step 510. is extracted the HOG characteristic in the suspicious scar zone of workpiece to be detected.
The scar characteristic that step 511. is extracted step 56 in the step 511 forms proper vector, use pair cross-section radially the svm classifier device that obtains of scar sample set autonomous learning proper vector is made the classification mark, judge the true and false of doubtful scar in the suspicious scar zone.
Step 512: the operation of step 56 to step 512 carried out in all the doubtful scar circulations in the suspicious scar zone, and all the doubtful scars in suspicious scar zone have all carried out the identification of cross-section radial scar.
Step 513: the watch window that rotates is carried out the operation of step 55 to step 513, accomplished the rotation in a week up to the workpiece cross-sectional image.By the rotational workpieces cross section prospect of degree, bring in constant renewal in suspicious scar zone, covered whole workpiece cross section prospects until the watch window circulation.
Above-mentioned cross-section radial scar detects in the operation, and step 57 is identical with the principle that horizontal scar detects the operation corresponding steps to the scar feature extracting method of step 511, therefore is not described in detail.
As shown in Figure 7; The svm classifier device training process flow diagram of describing among the figure for the horizontal scar detection of magnetic powder inspection environment lower surface operation, surface longitudinal scar detection operation, surperficial lathe tool line scar detection operation and cross-section radial scar detection operation; The svm classifier device training flow process of above-mentioned operation is consistent, so this unified description.In the above-mentioned operation, the training of the svm classifier device that obtains based on specific scar sample set autonomous learning of use comprises the steps:
Step 61: specific scar (like the horizontal scar in surface, surface longitudinal scar, surperficial lathe tool line scar and cross-section radial scar) highlight regions is carried out the image pattern collection, form the to be selected positive sample set of the specific scar svm classifier device of training.
Step 62: pseudo-scar highlight regions is carried out the image pattern collection, form the negative sample collection to be selected of the specific scar svm classifier device of training.Collection has formed the magnetic trace but has had no the pseudo-scar area image of workpiece of scar, forms the negative sample collection to be selected that training classifier uses.
Step 63: the magnetic trace highlight regions of acquisition testing workpiece (comprising specific scar highlight regions and pseudo-scar highlight regions) forms the test set of specific scar svm classifier device.
Step 64: positive sample set of initial training and initial training negative sample collection extract a sample formation from positive sample set to be selected and negative sample collection to be selected at random.
Step 65: align sample set and negative sample collection and extract specific scar characteristic, form the characteristic data set of training svm classifier device, use sample class mark and characteristic set pair svm classifier device to train, generate initial svm classifier device to specific scar.
Step 66: use test collection (comprising true scar zone and pseudo-scar zone) is tested initial svm classifier device, obtains the initial detecting rate.
Step 67: extract a new samples from positive sample set to be selected and negative sample collection to be selected at random and expand positive sample set of training and training negative sample collection, the operation in carry out step 65 generates new svm classifier device.
Step 68: the svm classifier device that the use test set pair is new is tested, and accepts the positive sample set of training that expands if verification and measurement ratio improves and trains the negative sample collection, otherwise reject newly-increased sample, carries out the picked at random exptended sample again.
Step 69: step 67 and step 68 are carried out in circulation, reach actual requirement or converge to a certain than high measurement accuracy up to final svm classifier device accuracy of detection.
Embodiment 2: embodiment 1 is to scar automatic testing method in the image under the magnetic powder inspection environment.On the basis of embodiment 1, the invention allows for automatic testing method to scar in the workpiece video under the magnetic powder inspection environment.Automatic detection for scar in the workpiece video; Can regard each frame of video as a width of cloth workpiece image; But in the video scar detect with image in the maximum difference that detects of scar be that same workpiece can form a continuous images sequence in video, and existingly in this image sequence differentiated for there being the frame of video of scar, have also that to be differentiated be the frame of video that does not have scar; Therefore the identification of scar is aimed at image sequence, but not individual picture.As shown in Figure 8, described to the workpiece scar recognition methods in the workpiece video under the magnetic powder inspection environment, comprise step:
Step S1: under the magnetic powder inspection environment workpiece flaw detection video is carried out scar and detect automatically; At first be to read the workpiece video by frame; Upgrade magnetic testing environment background, undergo mutation to confirm leaving of the arrival of new workpiece and workpiece that detection finishes through detection background.The workpiece video is pursued reading of frame, be used to judge for every frame statistical color histogram whether workpiece arrives.If workpiece arrives, then can change the distribution of the color histogram of present frame, according to color histogram whether sudden change having taken place confirms to carry out the time that scar detects.
Step S2. carries out scar to each the frame workpiece image that reads and detects for the workpiece that is detecting, and comprises step:
Step S21: workpiece image is being carried out in the automatic testing process of scar; At first carry out image filter and make an uproar and scheme image intensifying under the magnetic powder inspection environment, collecting workpiece image; Carry out image segmentation in conjunction with prior imformations such as the shape of workpiece, positions; Obtain the workpiece foreground area, the foreground area of workpiece is carried out the morphological image operation, extract suspicious scar zone;
Step S22: carry out laterally scar detection operation of surface; Utilize the horizontal scar detection technique in surface to the horizontal scar characteristic of suspicious scar extracted region; SVM (the SVMs that use obtains horizontal scar sample set autonomous learning; Support Vector Machines) sorter is classified to the scar characteristic automatically, confirms whether to exist in the suspicious scar zone laterally scar of surface;
Step S23: to not detecting the suspicious scar zone of scar among the step S22; Carry out the surface longitudinal scar and detect operation; Utilize surface longitudinal scar detection technique to the vertical scar characteristic of suspicious scar extracted region; Use is classified to the scar characteristic to the svm classifier device that vertical scar sample set autonomous learning obtains automatically, confirms whether there is the surface longitudinal scar in the suspicious scar zone;
Step S24: to not detecting the suspicious scar zone of scar among the step S23; Carry out surperficial lathe tool line scar and detect operation; Utilize surperficial lathe tool line scar detection technique to suspicious scar extracted region lathe tool line scar characteristic; Use is classified to the scar characteristic to the svm classifier device that lathe tool line scar sample set autonomous learning obtains automatically, confirms whether there is surperficial lathe tool line scar in the suspicious scar zone;
Step S25: carrying out radially to the workpiece cross section, scar detects operation; Utilize radially scar characteristic of suspicious scar extracted region that cross-section radial scar detection technique obtains the workpiece cross section; The svm classifier device that uses scar sample set autonomous learning radially to obtain is classified to the scar characteristic automatically, confirms whether there is the cross-section radial scar in the suspicious scar zone, workpiece cross section.(all utilized the svm classifier device to carry out autonomous learning, this process is the innovation place)
This step S2 has quoted the complete processing procedure of embodiment 1 fully, its concrete steps can reference implementation example 1, therefore be not described in detail.
Step S3: the workpiece video is being carried out in the automatic scar testing process; Same workpiece video will form an image sequence; The Multi-SVM sorter that is used to complete training carries out the judgement of the scar true and false according to the testing result of step S2 to image sequence, confirms finally whether this workpiece exists scar.
As shown in Figure 9, the training process of having described the Multi-SVM sorter that uses among the step S3 comprises the steps:
Step S31: all types of scar highlight regions are carried out the image pattern collection, form the positive sample set of the single characteristic svm classifier device of training.
Step S32: pseudo-scar highlight regions is carried out the image pattern collection, form the negative sample collection of the single characteristic svm classifier device of training.
Step S33: the magnetic trace highlight regions of acquisition testing workpiece forms the test set of single characteristic svm classifier device.
Step S34: to the negative sample collection of the positive sample set of single tagsort device and single tagsort device extract successively used sample to the data of single characteristic (like scar gradient characteristic, textural characteristics, frequency domain character or the like) as the single tagsort device of training; Use positive sample set and negative sample set pair list characteristic svm classifier device to train, generate single characteristic svm classifier device to certain scar characteristic.
Step S35: single characteristic svm classifier device that the use test set pair generates is tested, if verification and measurement ratio is higher than 0.5, then uses the F-score method to calculate the weight of sorter, otherwise should the zero setting of list characteristic svm classifier device weight.
Step S36: circulation execution in step S34 has generated corresponding single characteristic svm classifier device and weight to step S35 up to all scar characteristics.
Step S37: all single characteristic svm classifier devices of positive sample set test that use single characteristic svm classifier device; To the sample in the positive sample set of each single characteristic svm classifier device; With used single characteristic svm classifier device the respective weights addition is multiply by in the response of this sample and the value that obtains as a sample in the positive sample set of training Multi-SVM sorter; Use all single characteristic svm classifier devices of negative sample collection test of single characteristic svm classifier device; The sample of concentrating to the negative sample of each single characteristic svm classifier device multiply by the respective weights addition to the response of this sample and a sample that the value that obtains is concentrated as the negative sample of training Multi-SVM sorter with used single characteristic svm classifier device.
Step S38: use the positive sample set of Multi-SVM sorter and negative sample training to practice generation Multi-SVM sorter.
Embodiment 1 and embodiment 2 utilize compound characteristics the workpiece scar under the magnetic powder inspection environment to be carried out the description of multi-angle; Relying on computer vision technique provides the unified solution, particularly embodiment 2 that the polytype scar is discerned automatically not only to support the automatic identification of static workpiece scar for image but also has supported the automatic identification of dynamic workpiece scar for video.
Those of ordinary skill in the art will appreciate that embodiment described here is in order to help reader understanding's principle of the present invention, should to be understood that protection scope of the present invention is not limited to such special statement and embodiment.Those of ordinary skill in the art can make various other various concrete distortion and combinations that do not break away from essence of the present invention according to these teachings disclosed by the invention, and these distortion and combination are still in protection scope of the present invention.

Claims (9)

  1. Under the magnetic powder inspection environment based on the workpiece scar recognition methods of compound characteristics, it is characterized in that, may further comprise the steps:
    Step 1: workpiece image is being carried out in the automatic testing process of scar; At first carry out the image pre-service under the magnetic powder inspection environment, collecting workpiece image; The image preprocessing process comprises that image filter makes an uproar, schemes image intensifying and image segmentation; Image segmentation obtains the workpiece foreground area after accomplishing, and the foreground area of workpiece is carried out the morphological image operation, extracts suspicious scar zone;
    Step 2: carry out laterally scar detection operation of surface; Utilize the horizontal scar detection technique in surface to the horizontal scar characteristic of suspicious scar extracted region; SVM (the SVMs that use obtains horizontal scar sample set autonomous learning; Support Vector Machines) sorter is classified to the scar characteristic automatically, confirms whether to exist in the suspicious scar zone laterally scar of surface;
    Step 3: to not detecting the suspicious scar zone of scar in the step 2; Carry out the surface longitudinal scar and detect operation; Utilize surface longitudinal scar detection technique to the vertical scar characteristic of suspicious scar extracted region; Use is classified to the scar characteristic to the svm classifier device that vertical scar sample set autonomous learning obtains automatically, confirms whether there is the surface longitudinal scar in the suspicious scar zone;
    Step 4: to not detecting the suspicious scar zone of scar in the step 3; Carry out surperficial lathe tool line scar and detect operation; Utilize surperficial lathe tool line scar detection technique to suspicious scar extracted region lathe tool line scar characteristic; Use is classified to the scar characteristic to the svm classifier device that lathe tool line scar sample set autonomous learning obtains automatically, confirms whether there is surperficial lathe tool line scar in the suspicious scar zone;
    Step 5: carrying out radially to the workpiece cross section, scar detects operation; Utilize radially scar characteristic of suspicious scar extracted region that cross-section radial scar detection technique obtains the workpiece cross section; The svm classifier device that uses scar sample set autonomous learning radially to obtain is classified to the scar characteristic automatically, confirms whether there is the cross-section radial scar in the suspicious scar zone, workpiece cross section.
  2. 2. under the magnetic powder inspection environment according to claim 1 based on the workpiece scar recognition methods of compound characteristics; It is characterized in that; To collecting to such an extent that the magnetic powder inspection workpiece image carries out the process that image intensifying, image segmentation and suspicious scar extracted region were made an uproar, schemed in the image filter, comprise the steps: in the step 1
    Step 11: image is carried out medium filtering, the spiced salt noise that exists in the filtering image;
    Step 12: the image to filtering carries out the gray-scale value stretching;
    Step 13: image is carried out vertically and the horizontal edge extraction, obtain the external margin of workpiece according to the positional information of workpiece to be detected;
    Step 14: according to the shape information of workpiece to be detected, comprise curvature, length breadth ratio, surface area, the external margin that obtains in the step 13 is carried out curve fitting;
    Step 15: the prospect mask according to the result of curve fitting in the step 14 forms workpiece, extract complete workpiece to be detected in the magnetic powder inspection image;
    Step 16: the workpiece to be detected that extracts in the step 15 is carried out local binaryzation operation; Bianry image to obtaining carries out image filtering; The magnetic spot that filtering forms at surface of the work owing to spray magnetic keeps suspicious magnetic trace zone, as the suspicious scar zone of workpiece to be detected.
  3. 3. based on the workpiece scar recognition methods of compound characteristics, it is characterized in that under the magnetic powder inspection environment according to claim 1, in the step 2 to extract the suspicious scar zone of workpiece to be detected carry out the surface laterally scar detect operation, comprise the steps:
    Step 21: the image overall binary conversion treatment is carried out in the suspicious scar zone to workpiece to be detected;
    Step 22: the regional binary image of suspicious scar that obtains in the step 21 is carried out filtering noise and image smoothing operation;
    Step 23: the area that calculates the doubtful scar in the suspicious scar zone in the bianry image that in step 22, obtains;
    Step 24: with the bianry image that obtains in the step 23 is the prospect mask; Use Suo Beier operator (Sobel operator) that the doubtful scar in the suspicious scar zone is carried out rim detection and extracts its gradient characteristic, comprise boundary length, gradient mean, gradient maximal value, gradient minimum value and the gradient variance of doubtful scar;
    Step 25:, extract the boundary chain code statistical nature of the doubtful scar in the suspicious scar zone with the regional edge detection results of suspicious scar that obtains in the step 24;
    Step 26: the bianry image to obtain in the step 22 is the prospect mask, extracts doubtful scar and the brightness ratio of background in the suspicious scar zone;
    Step 27: HOG (the Histograms of Oriented Gradients) characteristic of extracting the suspicious scar zone of workpiece to be detected;
    Step 28: the scar characteristic that step 23 is extracted in the step 27 forms proper vector, uses the svm classifier device that horizontal scar sample set autonomous learning is obtained that proper vector is made the classification mark, judges the true and false of doubtful scar in the suspicious scar zone;
    Step 29: the operation of step 23 to step 28 carried out in all the doubtful scar circulations in the suspicious scar zone, and all the doubtful scars in suspicious scar zone have all carried out the identification of horizontal scar.
  4. 4. based on the workpiece scar recognition methods of compound characteristics, it is characterized in that under the magnetic powder inspection environment according to claim 1, in the step 3 the surface longitudinal scar is carried out in the suspicious scar zone that does not detect horizontal scar and detect operation, comprise the steps:
    Step 31: the image overall binary conversion treatment is carried out in the suspicious scar zone to workpiece to be detected;
    Step 32: the regional binary image of suspicious scar that obtains in the step 31 is carried out filtering noise and image smoothing operation;
    Step 33: with the bianry image that obtains in the step 32 is the prospect mask, uses the sobel operator that the doubtful scar in the suspicious scar zone is extracted its gradient characteristic, comprises gradient mean, gradient maximal value, gradient minimum value and the gradient variance of doubtful scar;
    Step 34: carry out Fast Fourier Transform (FFT) to extracting suspicious scar zone, extract the frequency domain character in suspicious scar zone;
    Step 35: carry out the horizontal direction projection to extracting suspicious scar zone, extract the average and the variance of the statistic histogram of suspicious scar zone level projection formation;
    Step 36: the HOG characteristic of extracting the suspicious scar zone of workpiece to be detected;
    Step 37: the scar characteristic that step 33 is extracted in the step 36 forms proper vector, uses the svm classifier device that vertical scar sample set autonomous learning is obtained that proper vector is made the classification mark, judges the true and false of doubtful scar in the suspicious scar zone;
    Step 38: the operation of step 33 to step 37 carried out in all the doubtful scar circulations in the suspicious scar zone, and all the doubtful scars in suspicious scar zone have all carried out the identification of vertical scar.
  5. 5. based on the workpiece scar recognition methods of compound characteristics, it is characterized in that under the magnetic powder inspection environment according to claim 1, in the step 4 surperficial lathe tool line scar is carried out in the suspicious scar zone that does not detect vertical scar and detect operation, comprise the steps:
    Step 41: the image overall binary conversion treatment is carried out in the suspicious scar zone to workpiece to be detected;
    Step 42: the regional binary image of suspicious scar that obtains in the step 41 is carried out filtering noise and image smoothing operation;
    Step 43: with the bianry image that obtains in the step 42 is the prospect mask, uses the sobel operator that the doubtful scar in the suspicious scar zone is extracted its gradient characteristic, comprises gradient mean, gradient maximal value, gradient minimum value and the gradient variance of doubtful scar;
    Step 44: carry out Fast Fourier Transform (FFT) to extracting suspicious scar zone, extract the frequency domain character in suspicious scar zone;
    Step 45: the horizontal direction projection is carried out in the suspicious scar zone to extracting, and extracts the average and the variance of the statistic histogram of suspicious scar zone level projection formation;
    Step 46: use the Canny algorithm to extract the vertical border in suspicious scar zone, obtain the vertical boundary number in suspicious scar zone;
    Step 47: extract the textural characteristics in suspicious scar zone, comprise angle second moment, moment of inertia, correlativity and entropy statistical nature;
    Step 48: HOG feature description that extracts the suspicious scar zone of workpiece to be detected;
    Step 49: the scar characteristic that step 43 is extracted in the step 48 forms proper vector, uses the svm classifier device that lathe tool line scar sample set autonomous learning is obtained that proper vector is made the classification mark, judges the true and false of doubtful scar in the suspicious scar zone;
    Step 410: the operation of step 43 to step 49 carried out in all the doubtful scar circulations in the suspicious scar zone, and all the doubtful scars in suspicious scar zone have all carried out the identification of lathe tool line scar.
  6. 6. based on the workpiece scar recognition methods of compound characteristics, it is characterized in that under the magnetic powder inspection environment according to claim 1, to the detection operation of workpiece cross-section radial scar, comprise the steps: in the step 5
    Step 51: image is carried out medium filtering, the spiced salt noise that exists in the filtering image;
    Step 52: the image to filtering carries out the gray-scale value stretching;
    Step 53: the image after step 52 processing is carried out local binaryzation operation earlier, then bianry image is carried out morphology and expand, obtain the maximum connected domain of region area as the prospect mask that extracts the workpiece cross section;
    Step 54: the prospect mask by obtaining in the step 53 obtains the workpiece cross section, and fixing transverse observation window is set, and watch window is less than the workpiece cross section, and is by the workpiece cross section that the degree rotation is obtained, that the workpiece cross section that gets into watch window is regional as suspicious scar;
    Step 55: image binaryzation is carried out in the suspicious scar zone in workpiece to be detected cross section handle;
    Step 56: the area that calculates the doubtful scar in the suspicious scar zone in the bianry image that in step 55, obtains;
    Step 57: with the bianry image that obtains in the step 55 is the prospect mask; Use the sobel operator that the doubtful scar in the suspicious scar zone is carried out rim detection and extracts its gradient characteristic, comprise boundary length, gradient mean, gradient maximal value, gradient minimum value and the gradient variance of doubtful scar;
    Step 58: the bianry image to obtain in the step 55 is the prospect mask, extracts the boundary chain code statistical nature of the doubtful scar in the suspicious scar zone;
    Step 59: the bianry image to obtain in the step 55 is the prospect mask, extracts doubtful scar and the brightness ratio of background in the suspicious scar zone;
    Step 510: HOG feature description that extracts the suspicious scar zone of workpiece to be detected;
    Step 511: the scar characteristic that step 56 is extracted in the step 510 forms proper vector, use pair cross-section radially the svm classifier device that obtains of scar sample set autonomous learning proper vector is made the classification mark, judge the true and false of doubtful scar in the suspicious scar zone;
    Step 512: the operation of step 56 to step 511 carried out in all the doubtful scar circulations in the suspicious scar zone, and all the doubtful scars in suspicious scar zone have all carried out the identification of cross-section radial scar;
    Step 513: the watch window that rotates is carried out the operation of step 55 to step 512, accomplished the rotation in a week up to the workpiece cross-sectional image.
  7. According under any described magnetic powder inspection environment in the claim 1 to 5 based on the workpiece scar recognition methods of compound characteristics; It is characterized in that, detect operation, surface longitudinal scar at the horizontal scar in said surface and detect operation, surperficial lathe tool line scar and detect the training that operation and cross-section radial scar detect the svm classifier device that obtains based on specific scar sample set autonomous learning that uses in the operation and comprise the steps:
    Step 61: specific scar highlight regions is carried out the image pattern collection, form the to be selected positive sample set of the specific scar svm classifier device of training;
    Step 62: pseudo-scar highlight regions is carried out the image pattern collection, form the negative sample collection to be selected of the specific scar svm classifier device of training;
    Step 63: the magnetic trace highlight regions of acquisition testing workpiece, comprise specific scar highlight regions and pseudo-scar highlight regions, form the test set of specific scar svm classifier device;
    Step 64: positive sample set of initial training and initial training negative sample collection extract a sample formation from positive sample set to be selected and negative sample collection to be selected at random;
    Step 65: align sample set and negative sample collection and extract specific scar characteristic, form the characteristic data set of training svm classifier device, use sample class mark and characteristic set pair svm classifier device to train, generate initial svm classifier device to specific scar;
    Step 66: the initial svm classifier device of use test set pair is tested, and obtains the initial detecting rate;
    Step 67: extract a new samples from positive sample set to be selected and negative sample collection to be selected at random and expand positive sample set of training and training negative sample collection, the operation in carry out step 65 generates new svm classifier device;
    Step 68: the svm classifier device that the use test set pair is new is tested, and accepts the positive sample set of training that expands if verification and measurement ratio improves and trains the negative sample collection, otherwise reject newly-increased sample, carries out the picked at random exptended sample again;
    Step 69: step 67 and step 68 are carried out in circulation, reach actual requirement or converge to a certain than high measurement accuracy up to final svm classifier device accuracy of detection.
  8. Under the magnetic powder inspection environment based on the workpiece scar recognition methods of compound characteristics, it is characterized in that, may further comprise the steps:
    Step S1: under the magnetic powder inspection environment workpiece flaw detection video is carried out scar and detect automatically; At first be to read the workpiece video by frame; Upgrade magnetic testing environment background, undergo mutation to confirm leaving of the arrival of new workpiece and workpiece that detection finishes through detection background;
    Step S2: for the workpiece that is detecting, each the frame workpiece image that reads is carried out scar detect, comprise step:
    Step S21: workpiece image is being carried out in the automatic testing process of scar; At first carry out image filter and make an uproar and scheme image intensifying under the magnetic powder inspection environment, collecting workpiece image; Prior imformation in conjunction with the shape that comprises workpiece, position is carried out image segmentation; Obtain the workpiece foreground area, the foreground area of workpiece is carried out the morphological image operation, extract suspicious scar zone;
    Step S22: carry out laterally scar detection operation of surface; Utilize the horizontal scar detection technique in surface to the horizontal scar characteristic of suspicious scar extracted region; Use is classified to the scar characteristic to the svm classifier device that horizontal scar sample set autonomous learning obtains automatically, confirms whether to exist in the suspicious scar zone laterally scar of surface;
    Step S23: to not detecting the suspicious scar zone of scar among the step S22; Carry out the surface longitudinal scar and detect operation; Utilize surface longitudinal scar detection technique to the vertical scar characteristic of suspicious scar extracted region; Use is classified to the scar characteristic to the svm classifier device that vertical scar sample set autonomous learning obtains automatically, confirms whether there is the surface longitudinal scar in the suspicious scar zone;
    Step S24: to not detecting the suspicious scar zone of scar among the step S23; Carry out surperficial lathe tool line scar and detect operation; Utilize surperficial lathe tool line scar detection technique to suspicious scar extracted region lathe tool line scar characteristic; Use is classified to the scar characteristic to the svm classifier device that lathe tool line scar sample set autonomous learning obtains automatically, confirms whether there is surperficial lathe tool line scar in the suspicious scar zone;
    Step S25: carrying out radially to the workpiece cross section, scar detects operation; Utilize radially scar characteristic of suspicious scar extracted region that cross-section radial scar detection technique obtains the workpiece cross section; The svm classifier device that uses scar sample set autonomous learning radially to obtain is classified to the scar characteristic automatically, confirms whether there is the cross-section radial scar in the suspicious scar zone, workpiece cross section;
    Step S3: the workpiece video is being carried out in the automatic scar testing process; Same workpiece video will form an image sequence; The Multi-SVM sorter that is used to complete training carries out the judgement of the scar true and false according to the testing result of step S2 to image sequence, confirms finally whether this workpiece exists scar.
  9. 9. based on the workpiece scar recognition methods of compound characteristics, it is characterized in that under the magnetic powder inspection environment according to claim 8 that the training process of the Multi-SVM sorter that uses among the said step S3 comprises the steps:
    Step S31: all types of scar highlight regions are carried out the image pattern collection, form the positive sample set of the single characteristic svm classifier device of training;
    Step S32: pseudo-scar highlight regions is carried out the image pattern collection, form the negative sample collection of the single characteristic svm classifier device of training;
    Step S33: the magnetic trace highlight regions of acquisition testing workpiece forms the test set of single characteristic svm classifier device;
    Step S34: align sample set and negative sample collection and extract a certain scar and describe characteristic, form the data set of the single characteristic svm classifier device of training; Use sample class mark and positive sample set and negative sample set pair list characteristic svm classifier device to train, generate single characteristic svm classifier device of describing characteristic to certain scar;
    Step S35: single characteristic svm classifier device that the use test set pair generates is tested, if verification and measurement ratio is higher than 0.5, then calculates the weight of this list characteristic svm classifier device, otherwise should the zero setting of list characteristic svm classifier device weight;
    Step S36: circulation execution in step S34 describes characteristic up to all scars and has generated corresponding single characteristic svm classifier device and weight to step S35;
    Step S37: all single characteristic svm classifier devices of positive sample set test that use single characteristic svm classifier device; To the sample in the positive sample set of each single characteristic svm classifier device; With used single characteristic svm classifier device the respective weights addition is multiply by in the response of this sample and the value that obtains as a sample in the positive sample set of training Multi-SVM sorter; Use all single characteristic svm classifier devices of negative sample collection test of single characteristic svm classifier device; The sample of concentrating to the negative sample of each single characteristic svm classifier device multiply by the respective weights addition to the response of this sample and a sample that the value that obtains is concentrated as the negative sample of training Multi-SVM sorter with used single characteristic svm classifier device;
    Step S38: use the positive sample set of Multi-SVM sorter and negative sample training to practice generation Multi-SVM sorter.
CN2010101629597A 2010-05-05 2010-05-05 Workpiece flaw identification method based on compound characteristics in magnaflux environment Expired - Fee Related CN101852768B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN2010101629597A CN101852768B (en) 2010-05-05 2010-05-05 Workpiece flaw identification method based on compound characteristics in magnaflux environment

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN2010101629597A CN101852768B (en) 2010-05-05 2010-05-05 Workpiece flaw identification method based on compound characteristics in magnaflux environment

Publications (2)

Publication Number Publication Date
CN101852768A CN101852768A (en) 2010-10-06
CN101852768B true CN101852768B (en) 2012-02-22

Family

ID=42804354

Family Applications (1)

Application Number Title Priority Date Filing Date
CN2010101629597A Expired - Fee Related CN101852768B (en) 2010-05-05 2010-05-05 Workpiece flaw identification method based on compound characteristics in magnaflux environment

Country Status (1)

Country Link
CN (1) CN101852768B (en)

Families Citing this family (18)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102590330A (en) * 2011-12-29 2012-07-18 南京理工大学常熟研究院有限公司 Image processing-based magnetic particle inspection defect intelligent identification detection system
CN102854191B (en) * 2012-07-18 2014-09-10 湖南大学 Real-time visual detection and identification method for high speed rail surface defect
CN102854193A (en) * 2012-08-30 2013-01-02 苏州天准精密技术有限公司 Detection method and detection system used for image defect detection
CN103395433B (en) * 2013-07-30 2016-04-27 广州电力机车有限公司 A kind of Locomotive body maintenance method
CN103745234B (en) * 2014-01-23 2017-01-25 东北大学 Band steel surface defect feature extraction and classification method
CN103984951B (en) * 2014-04-25 2017-12-08 西南科技大学 A kind of Magnetic testing defect inspection method and system
CN104297335A (en) * 2014-09-25 2015-01-21 武汉重工铸锻有限责任公司 Method applied to quality detection of surface of hot extrusion pipe nozzle of main steam super pipeline
FR3030042A1 (en) * 2014-12-15 2016-06-17 Michelin & Cie METHOD OF DETECTING DEFECT ON A PNEUMATIC SURFACE
CN104463180B (en) * 2014-12-31 2018-01-12 深圳市华星光电技术有限公司 The method for identifying conducting particles press mark
CN104951795B (en) * 2015-05-26 2019-07-05 重庆金山科技(集团)有限公司 Image classification identifies judgment method
CN105388162B (en) * 2015-10-28 2017-12-01 镇江苏仪德科技有限公司 Raw material silicon chip surface scratch detection method based on machine vision
CN106485231A (en) * 2016-10-21 2017-03-08 上海未来伙伴机器人有限公司 A kind of target color lump extracting method and device
CN107145896A (en) * 2017-03-14 2017-09-08 西南科技大学 Dysnusia identifying system based on fluorescentmagnetic particle(powder)
CN107862689A (en) * 2017-11-21 2018-03-30 广东工业大学 Leather surface substantially damaged automatic identifying method and computer-readable recording medium
CN108460763B (en) * 2018-03-26 2021-03-30 上海交通大学 Automatic detection and identification method for magnetic powder inspection image
CN108876769B (en) * 2018-05-31 2020-11-03 厦门大学 Left auricle CT image segmentation method
CN109118471A (en) * 2018-06-26 2019-01-01 广东工业大学 A kind of polishing workpiece, defect detection method suitable under complex environment
WO2020140198A1 (en) * 2019-01-02 2020-07-09 深圳市邻友通科技发展有限公司 Fingernail image segmentation method, apparatus and device, and storage medium

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101122585A (en) * 2007-09-12 2008-02-13 天津大学 Automatic identification method for supersonic phased array for detecting oil gas pipeline girth weld defect type
CN101403602A (en) * 2008-11-17 2009-04-08 天津理工大学 Embedded high-speed on-line machine vision detection method and device
CN101526484A (en) * 2009-04-13 2009-09-09 江南大学 Bearing defect detecting technique based on embedded-type machine vision

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8108324B2 (en) * 2008-05-15 2012-01-31 Intel Corporation Forward feature selection for support vector machines

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101122585A (en) * 2007-09-12 2008-02-13 天津大学 Automatic identification method for supersonic phased array for detecting oil gas pipeline girth weld defect type
CN101403602A (en) * 2008-11-17 2009-04-08 天津理工大学 Embedded high-speed on-line machine vision detection method and device
CN101526484A (en) * 2009-04-13 2009-09-09 江南大学 Bearing defect detecting technique based on embedded-type machine vision

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
朱凌云 等.基于支持向量机的缺陷识别方法.《重庆大学学报(自然科学版)》.2002,第25卷(第6期),42-54. *

Also Published As

Publication number Publication date
CN101852768A (en) 2010-10-06

Similar Documents

Publication Publication Date Title
CN101852768B (en) Workpiece flaw identification method based on compound characteristics in magnaflux environment
CN107389701A (en) A kind of PCB visual defects automatic checkout system and method based on image
CN102854191B (en) Real-time visual detection and identification method for high speed rail surface defect
Zhang et al. Matched filtering algorithm for pavement cracking detection
CN104021574B (en) Pavement disease automatic identifying method
CN103871053B (en) Vision conspicuousness-based cloth flaw detection method
CN106093066A (en) A kind of magnetic tile surface defect detection method based on the machine vision attention mechanism improved
CN104215639A (en) Pear surface defect detection method based on machine vision
Dubey et al. Maximally stable extremal region marking-based railway track surface defect sensing
CN107782733A (en) Image recognition the cannot-harm-detection device and method of cracks of metal surface
CN106250936A (en) Multiple features multithreading safety check contraband automatic identifying method based on machine learning
CN107876425A (en) A kind of bearing defect detecting system device of view-based access control model
CN107610104A (en) Crack detecting method at a kind of elevator compensation chain R based on machine vision
CN110211101A (en) A kind of rail surface defect rapid detection system and method
CN109785301A (en) A kind of rail corrugation automatic identifying method and appraisal procedure based on image procossing
CN110009618B (en) Shaft part surface quality detection method and device
Alimohamadi et al. Defect detection in textiles using morphological analysis of optimal Gabor wavelet filter response
CN106914428A (en) A kind of New Algorithm of the steel ball surface defect Differential Detection based on machine vision
CN109724984A (en) A kind of defects detection identification device and method based on deep learning algorithm
CN108985170A (en) Transmission line of electricity hanger recognition methods based on Three image difference and deep learning
CN108416774A (en) A kind of fabric types recognition methods based on fine granularity neural network
CN107392884A (en) A kind of identification of solid coloured cloth defect regions based on image procossing and extracting method
CN103984951A (en) Automatic defect recognition method and system for magnetic particle testing
CN107145896A (en) Dysnusia identifying system based on fluorescentmagnetic particle(powder)
CN103258218B (en) Masking-out detects skeleton generating method, device, defect inspection method and device

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
C14 Grant of patent or utility model
GR01 Patent grant
C17 Cessation of patent right
CF01 Termination of patent right due to non-payment of annual fee

Granted publication date: 20120222

Termination date: 20140505