Research On Pattern Recognition Methods Based On Swarm Intelligence

The method of swarm intelligence is a new technology to solve most global optimization problems effectively. This kind of method can find the optimal solutions of complex problems more quickly than traditional optimization algorithms. Now the application fields of swarm intelligence have extended to multi-goals optimization, data classification and clustering, biology system modeling, imitation and system identification, etc. and the swarm intelligence theories offer a new path to solve these application problems. As a new evolution calculating technology, the swarm intelligence method has become the attention focus for more and more researchers. So it has important academic significance and practical value to develop the research of swarm intelligence theories and applications.The methods of pattern recognition are investigated deeply by using the swarm intelligence theories in this paper, and corresponding improving algorithms are proposed. The effectiveness of the algorithms in applications is demonstrated by simulation experiments. The main contents of this paper are as follows.On how to remove noises while the image details are not affected, a de-noising method based on the fourth order cumulant aiming at the zero-mean Gaussian noise is proposed in this paper. The method is better than other methods in removing zero-mean Gaussian noise and can preserve more image details as well. And an optimization algorithm of selecting the optimal one from varied de-noising methods based on the PSO is presented so that the image quality is improved at best. Besides, this paper also applies the PSO to the soft morphology to select the best method of image edge detection. The effectiveness of the PSO in the preprocessing stage of pattern recognition is proved by experiments.Features selection plays an important role in pattern recognition field and affects the correct rate and speed rate of pattern recognition directly. To select correct and effective features has become the main procedure in pattern recognition. Though many scholars have proposed many methods in correlative research, in most cases features are selected by experience or comparative experiments. In order to select features which can detach all kinds of samples as far as, a feature selection algorithm is proposed by the improving ACO based on the entropy. Human face recognition is researched to prove the effectiveness of the theories proposed above. In order to extend and improve classification methods, a pattern recognition way based on PSO is proposed in this paper. The method is to set up the database for all samples at first. And then, by using the technology of binarization and thinning, the sample image being recognized and the images in the database set up are preprocessed. Then the points in the sample image after thinning are matched with the points in the thinning images from the database by using PSO. In the end a classifier is designed by seeking the minimum of the differences between the matching images and the thinning images from the database. The effectiveness of pattern recognition methods based on PSO is proved by the researches of automobile license plate recognition and linear robots formation recognition. General Ant Colony Optimization (GACO) is proposed in the paper and the design method of multi-sorts classifier based on GACO proposed in the paper is presented. And the robots formation transform is also researched by combining the dots coordinate transform to prove the effectiveness of GACO.