| Carrying out road traffic information survey is the basic method to master highway condition, and the identification of disease is a core component of the highway investigation. In current situation, artificial road-based identification is still the main way to recognize road disease, but problems are increasingly appear, including high cost of recognition, slow speed, long period, tedious work, high funding. With the ongoing construction of the highway, the total number of road network operations increase, which presents more and more arduous task in road survey. The traditional way of working has been unable to adapt to the needs of the present stage of conservation, scientific research, and intelligence, so that systematic pavement typical disease identification technology is imperative. This paper, with the basis of wavelet moment and SVM (Support Vector Machine), and relying on the highway in Hunan, will conduct survey on Asphalt Pavement image recognition technology, to build recognition system of highway asphalt pavement typical disease image, and take the example of Xiangtan Expressway Asphalt Pavement image in 2011 and 2012, so as to conduct case study on typical disease of Expressway Asphalt Pavement image recognition system Based on the in-depth analysis on domestic and foreign related research, this paper has specific research results as follows:(1) Asphalt pavement typical feature disease extraction algorithm. Of today’s mainstream moment invariants algorithm research, considering the various noise immunity invariant moments and image detail extraction capacity, a study will be conducted on principles of Hu moment, Zernike moment, legendre moment, and wavelet moment. And among those, to choose the most suitable for the extraction of typical diseases of asphalt pavement image feature points for the basis of image feature classification referring in the next point.(2) Characteristics Classification Algorithm. A study is conducted on common characteristios of the classification algorithm, BP network, SVM algorithm mechanism And among these, choose the most suitable classification algorithm for image features of asphalt pavement typical disease. Based on feature points extracting from the same matrix referring in(1), analyze and compare of learning time, recognition time, and other aspects of recognition accuracy through experiments, which demonstrates the superiority of SVM and wavelet moments.(3)Establish a recognition system of typical asphalt pavement disease image. Typical road diseases are divided into four steps to be identified:first, capture the image information, and to acquire needed image for recognition; second, pre-process image to remove noise and other interference and highlight the main diseases image, image segmentation etc.; third, select the appropriate wavelet function, wavelet moments algorithm to extract feature points from disease images; fourth, train classification model in the setup of SVM. After parameters has been set by a trained classification model, features can be identified by hyperplane and final recognition results can be drawa The paper describes theoretical models and establishment of the pre-processing model, feature extraction model and classification model. It determines selection principle of noise processing method, image segmentation, selection of wavelet methods, wavelet bases, and SVM initial parameter, in the context of high-speed asphalt pavement image(4) Take Hunan Xiangtan speed Asphalt Pavement image in 2011 and 2012 as samples to establish experimental disease image sample library so as to train SVM. Then extract feature points by wavelet moments, and input the trained SVM to classify features. Finally experimentally demonstrate the feasibility of the establishment of a high-speed asphalt pavement typical image recognition system diseases, correctness and superiority referring in (3). In the case of small sample study, it obtains 89.86% of correct identification rate and 93.86% of final recognition rate. |
PDF Full Text Request