Research On Methods Of Feature Extraction And Recognition Of Crack Images In Tunnels

The rapid automatic inspection system and maintenance technology of high-speed railway tunnels is an indispensable technical guarantee for operation safety and it is one of the major directions for the development of tunnel engineering technology.The image processing of the inspection system has been discussed.The skeleton extraction,inflection point identification,characteristic selection and image matching of the crack images have been researched by the thesis.The main research work and results are as follows:1.The composition of image processing system and image preprocessing process were summarized by comparing different image de-noising techniques and threshold segmentation algorithms.The adaptive median filtering de-noising method and Sobel-4 direction Otsu’s improved threshold segmentation algorithm were selected to ensure that the acquired crack images were converted into binary images.2.In the skeleton extraction stage,morphological operations were used to extract the crack part separately and fill the cavity.According to the characteristics of cracks,a new method of crack skeleton extraction was proposed.Firstly,the topology refinement method was used for primary identification;then a new adaptive tracking branch culling algorithm was proposed.The endpoints and intersections were automatically determined by the number and distribution of black pixel points in the connected region of the target point.The tracking algorithm was used to record the skeleton branch coordinates and the length threshold was set for branch elimination.After several experiments,the results show that the algorithm has a good effect of branch elimination and the elimination rate reaches 100%;the calculation speed is fast and the average processing time is less than 1 second per image.3.In the inflection point identification stage,a new chain-coded inflection point recognition algorithm was proposed.Firstly,the improved Freeman chain code was used to transform the crack information into the chain code sequence and the partial crack trend was represented by the mean value of the chain code within the region length l.The locations of potential inflection points were determined after calculating absolute value of the difference of mean value between the chain codes.Secondly,the pseudo-inflection points near the actual inflection points,the redundant inflection points with big corner and close distance were removed.After experiments on crack images under different conditions,the results show that the algorithm has a good recognition effect and the curve which is consist of inflection points is coincident with the crack;the optimal range of l proposed is representative.4.In the characteristic selection stage,the ratio of distance between the inflection points and the corner were selected as the characteristic.The coefficient of variation was introduced and final formula was determined according to characteristic.Different parameter values lead to different number of characteristic,in which case the circulation was used for step-by-step matching and secondary verification was conducted by using eigenvalue data in the interval.Matching the "father crack" and "son crack" images and the results show that the accuracy of the matching algorithm is high and the matching success rate of the same inflection point reaches 75%;the characteristic can describe the characteristics of cracks accurately even if the cracks extend.5.The MATLAB software was used for algorithm implementation and the entire process of image processing could be conducted automatically.The processing speed is fast and the average time of is less than 1.5 seconds per image.