Study On Crack Detection System Of Straddle Monorail Traffic PC Beam Surface Based On Support Vector Machine

With the rapid increase of urban population density and travel frequency,the urban traffic system is facing great challenges.At present,the rail transit is an effective way to alleviate traffic pressure.Chongqing City has carried out a lot of explorations,and is establishing the rail transit network,dominated by the straddle type monorail transportation.The defects such as cracks on the surface of the PC track beam,serving as the supporting part of straddle type monorail transportation,will threaten the operation of the whole traffic line.Thus,the detection on cracks is of great concern.Currently,the common detection method is manual.However,the manual method is inefficient and large error.Therefore,the development of effective automatic detection system is hanging over our head.In this paper,the author commences on the automatic crack detection system,and studies several aspects,including the pretreatment of the crack image,the rapid identification,and the crack segmentation.The research contents are as follows:(1)this paper presents the overall framework.The automatic detection module includes image preprocessing,rapid identification of crack image,crack segmentation and parameter calculation.The crack information query consists of the crack image and the crack parameters.(2)Because the image of the track beam is included in the non-target area,the first step is to reduce the image according to the gray value and remove the non-target area.At the same time,for better later processing,the reduced image is further divided into 200*200 image sub blocks.(3)Due to the important effects of noise on track beam image,crack edge blurs,which increases the difficulty in later processing.For that reason,the technology of sparse representation and dictionary learning is first brought to bear for image pre-processing.After the image data analysis of the track beam,aiming to the crack area,non-crack area respectively,regarding the traditional discrete cosine transform(DCT)as learning dictionary,and combine the two a joint dictionary.Track beam image is decomposited sparsely under the joint dictionary.Sparse component contains useful information,and residual has noise information,and the removal of residual runs up to good results.The results present that the algorithm has a good effect on pre-treatment of track beam image,providing a good basis for the subsequent treatment.(4)In the light of the features of the crack image on the track beam,an image recognition algorithm based on dense SIFT(dense SIFT)and support vector machine(SVM)is put forward in the paper.Detailed introduction of the principle and extraction steps of dense SIFT is written.After the formation characteristic descriptor,cluster a large number of feature vector through K-means algorithm in images.In virtue of the characteristic of bag of words(BOW),put the statistics of each category after clustering as the final characteristics,and carry out the track beam identification with the SVM.The experiment indicates that the algorithm brings high detection accuracy.(5)Further crack extraction and analysis would be done for the crack image detected.The paper uses the crack image segmentation algorithm,blend of threshold segmentation algorithm crack and SVM.And extract the image sub block,including the gray feature and histograms of oriented gradients(HOG)from the track beam image to form descriptor.For high training efficiency,the pixel segmentation is first used.Serving the image sub block in super pixel center point coordinate block as training data,extract feature extraction before sent into the SVM training model.Thus,the training sample is more representative.The image to be measured is segmented according to the threshold,and the low pixel area is performed as the potential crack region,using the trained SVM model for segmentation.Finally,use the morphological algorithm to remove the pseudo cracks and impurities in segment image.(6)The experimental on the real track beam image set shows that the correct rate of crack image reaches 92.33%.The detection rate of crack segmentation is 3.91%,which proves the validity of the proposed system.