Papers by Author: Xue Yong Li

Abstract: Computer vision based mechanical parts recognition has been received much research attention in recent years. In this paper, we present a new framework to address this problem. The framework utilizes the computer graphic technology to model mechanical parts. Recognition is realized by comparing one query image to the instance images using improved affine transformation based on particle swarm optimization (PSO). Our experiment shows that the proposed framework outperforms the conventional invariant moments based recognition methods in recognition rates.

Abstract: Because of shortcomings of images of motion objects in deep space，including non-uniform distribution of background pixels，low SNR, general 1ow picture quality and small size，a novel image segmentation algorithm based on the combination of image gradient and its variance is adopted. An algorithm to compute the correlation coefficient based on the small region template matching is presented. The algorithm overcomes the disadvantages of correlation disturbed frequently by noise, and size difference, besides it also reduces the computational complexity and improves the speed of the system．Experiments show that the algorithm has great accuracy，high speed，and good stability，and can meet the requirements of a real-time system.

Abstract: Different laser parameters and the interactional production can affect the barcode grade profoundly. In order to mark barcodes on aluminum alloy with a laser effectively, it is important to search out the optimum parameters and establish a relationship between critical laser parameters and the quality of the symbol. In this research, a Nd:YAG laser was utilized to produce 3136 Data Matrix symbols onto aluminum substrates. The experiment focuses on all interval values of the vector step, inter-step time, laser Q frequency and laser Q release time. From the experiment results, the optimized parameters are the vector step of 0.005~0.009 mm, the inter-step time of 29~43 μs, laser Q frequency of 7~10 KHz, and laser Q release time of 13~19 μs. And then, by analyzing the higher and the lower power density processing module SEM images and EDS data, it was found that the microstructure and micro-components of formations that created by the interaction between the laser and the aluminum alloy can affect the barcodes grade level significantly.

Abstract: To improve the edge detecting effect of the pressed character, a novel method based on facet model and topographic structures is proposed. Firstly, the discrete gray scale image is approximated by a bivariate cubic function, then, the concerned features of the continuous function are computed to describe the discrete image. After every pixel in the image has been defined to Peak, Ridge, Saddle, Flat, Ravine, Pit or Hillside, the edge of the character is detected. To reduce the time in computing the coefficients of cubic function, an efficient separable algorithm is presented. In addition, an image enhancement method prior to edge detection is adopted to improve the detecting effect. The tests show that the proposed method is more suitable for complex edge detecting problems than common methods.

Abstract: In accordance with the obvious characteristics of the pressed raised character image and the shortages of the template matching method.
a new method of using the general radial-basis function neural network (GRBFN) for testing the quality of the pressed character is presented. The structures and training methods of GRBFN are fully analyzed, as well as the functionality of hidden layer, excited focus and area. The results show the checker based on GRBFN has highly checking ratio for the label pressed raised characters. It is suited to the quality testing of raised characters.