Research On Feature Recognition And Classification Method For Workpiece Based On Machine Vision

The technology of classification and recognition based on machine vision includes the knowledge of image process, machine vision, pattern recognition, artificial intelligence, precision instrument, optics and so on. This technology is applied in the industrial field, which raises efficiency of production and accuracy of detection, and which is helpful to informatization construction and modern management for enterprise. The main contents of this thesis are the following:(1) The scientific documentations about this technology are referred in the internationality. The important problems of the machine vision system used for detection, recognition and classification of workpiece are analyzed. These problems include acquisition of high quality image, feature selection and optimization, classifier design.(2) Disadvantageous factors of image quality are analyzed. The design principles for light source are introduced. A LED light source is made. Influence factors of flat-field for the monochrome array CCD camera are analyzed and researched. Anew method of image correction is presented.(3) The theory and method of feature selection for workpiece are researched. The application program of feature selection and optimization is designed and developed. Thirty-two functions of feature extraction and three methods of threshold segmentation are included in this application program.(4) A high efficient and general algorithm of image matching is designed through improving SSDA. Aiming at the generality and accuracy of neural network, a new three-layer neural network model based on BP feed-forward neural network is designed.(5) An experimental equipment used for transportation of product and acquisition of image is designed and developed. Seven sets of application programs are finished. The foresaid algorithms are tested through four groups of industrial products. The factors of causing error are analyzed.