Algorithm Research On Image Recognition About Welding Seam Of Car Body

With the rapid development of modern industry such as machinery, the nuclear industry, aerospace, energy, electronics industries and so on, the welding industry becomes more and more important. So, the demands of welding quality and welding automation are higher and higher, welding automation, intelligent have become the important directions for the development of welding technology, in 21st century. Especially in the automotive industry, in order to reduce costs and maximize profits, increase the degree of automation of production has become the inevitable choice. How to improve the level of automation about production line and production speed, while ensuring product quality is an important criterion for the establishment of production lines. The welding is the technical difficulties of the automated production line, particularly key parts of the welds are qualified for the quality of the products play a crucial role.The content of this thesis is the algorithm of image recognition, part of image recognition "the based on infrared CCD Vision Inspection System", this inspection system is developed for the inspection of the truck cab welds, aims to improve the production efficiency. Rapid and accurate are the important indicators of this weld inspection system. Truck's cab include 600 welds, of which only a small part of the welds are completed by the welding robot welding, because the reliability of welding robot so this part of the welds do not require testing; the remaining 400 welds, Because the location of welds, they are difficult to use welding robot welding, but only by hand and welded, there may be have leak welds.Previously, many researchers use infrared CCD to identification of vehicles, and have do a lot of research in infrared image recognition, while there are many researchers use CCD to weld inspection, do some research on the weld image recognition.Most of the traditional recognition systems are based on the image binarization, this method is easy to image processing, but the greatest disadvantage of this method is that bring the main features of the image information lost. In connection with accuracy and real-time issues in weld image, the paper proposed a new recognition algorithm based on Karhunen-Loeve transform and wavelet transform. This study through the use of wavelet decomposition and K-L transform means, and efficacious improve the defects of the traditional weld recognition systems, and get a greatly improve in the accuracy and computational speed about recognition systems. This new method can be effective to retain the characteristics of the original image information. Against the characteristics of infrared image brightness differences, the first, through the method of linear transformation to reduced the problem of brightness; then use wavelet transform to extract the relative stability low-frequency sub-band map of weld image, so can get the main features of the image information; and then use K-L transform to extract and identify the main characteristics of components. The results show that the recognition algorithm has basically reached the requirements of accuracy and real-time.