Image Recognition Algorithms Of The Typical Failures Of The Freight Train Based On Machine Vision

Due to the establishment of new transport organization and the new requirements of high-speed and overloading truck, the region of truck inspection is extended and the time of truck inspection is shortened, which increase the requirement of a higher efficiency and quality of truck inspection. The traditional train inspection method based on the artificial control mode has been unable to meet the requirements, and it replaced by the running trouble of freight car detection system(TFDS) based on the human-computer cooperation mode. In recent years, many technologies, such as machine vision, image processing and pattern recognition, is more and more perfect, which lay the theoretical foundation for the transformation of TFDS system to complete mechanical control mode.Based on the in-depth analysis of the composition and functions of the TFDS system, the dedicated computer used for image automatic identification is selected as the research object, which is subjected to the machine vision fault automatic identification subsystem, which is used to realize image preprocessing, image feature extraction and fault judgment functions. The main research content of the thesis is described as follows:(1) In the stage of image preprocessing, considering the weather conditions, the environmental conditions and the influences of transmission equipment, this thesis puts forward the salt and pepper noise filtering algorithm which is used to eliminate noise, and the improved retinex algorithm is used to the lighting compensation. The image dehazing method based on fog image degradation model is presented for image defogging, and the performance of each algorithm is analyzed from two aspects of subjective visual effect and objective evaluation criteria.(2) In the stage of image feature extraction, the image features of failures are analyzed, such as the retaining bound lost, the binding bolt loose and the crossing rod bend. And they are summarized as three kinds, including regional significant change, boundary significant change and structure significant change. Image processing technologies are adopted to extract the image features of each failure, such as the method of maximum classes square error, region growing method, canny edge detection operator and hough transform.(3) In the stage of fault judgment, according to the extracted image feature, the fault recognition algorithm based on regional characteristics, the fault recognition algorithm based on boundary characteristics, and the fault recognition algorithm based on structure characteristics are designed respectively to judge the retaining bound lost, the binding bolt loose, the crossing rod bend fault.