Research On Mechanical Part Recognition Based On Neural Network

Mechanical parts auto-test is an important segment of manufacture system. Modern manufacture technology has been widely use the auto-test and product-recognize technology to ensure the quality of product and even to make manufacture system run in more reliable .It is a key segment in the process of manufacturing industry information. So the computer vision test technology has very important use in test of product's quality and of manufacture process of manufacturing industry. ART technology is an important branch of computer vision, and is the pattern recognition technology which can recognize, location and describe the special physical target through analyzing the image's data.If we want to realize the auto-test of mechanical part, fist of all, we should get the image of part through certain testing and collecting equipment, and make the part digitalize. And then we can get the type of part by pattern recognition. At last, we test the part by image. The work of this paper is to recognize the image of part by neural network. The idea about this paper comes from observation to flexibility and stabilization vision behavior of human. In this paper, the advantages of human vision is introduced in the image recognition field and combined with general digital image processing technology to make the best of their advantages and make up their disadvantages. In this way, we can find a flexible, universal and stable algorithm of flat image recognition.To realize the pattern recognition, the main work is listed following:1) We have studied the theories about image processing, binary image, and image filter. And contrast several edge detect operators. At last, we decide to use med-filter to eliminate the image's noise and use the Canny edge detect operator to detect the edge of image.2) Study the two fixture features of image: Hu moment features and NMI feature. Through experiment, we find that the stability of 5,6,7 order moments are not as well as that of 1.2,3,4 order moments. So we use the 1,2,3,4 order moments and NMI feature as the input of neutral network.3) Discuss and contrast several methods of pattern recognition which used widely recent years, and mainly introduce the method of artificial neutral network, analyze the...