Research On Application Of Biomimetic Pattern Recognition

Pattern recognition is one of the most important areas of research and application on artificial intelligence. With the rapid development of artificial intelligence, pattern recognition accordingly has been widely used in many fields. However, owing to the traditional statistical pattern recognition regarding"optimal seperating"as its main principle, there exist many limitations. Hence, it is poor in response capabilities, and the actual effect is far from satisfactory.In recent years, a method called biomimetic pattern recognition has been proposed. The new method is based on"matter cognition"instead of"matter classification"which is better closer to the function of human being. In order to complete the best cover to the"matter", it must be taken the continuity between homologous similar samples as the priori probability, used the multi-weight higher order neuron network, and established the closed hypersurfaces in high dimensional feature space, thus the goal of the recognition can be achieved. This paper is mainly aimed at the research on the method and application of the biomimetic pattern recognition. And with the support of the National Natural Science Foundation Project, the recogniton of different shape of the identical algae cell is initially completed.This thesis including:1. Elaborates the pattern recognition development situation, the system constitution and the essential methods, and proposes the limitation of traditional pattern recognition. Compared with the traditional pattern recognition, the biomometic pattern recognition method and the continuity principle between homologous samples are introduced from the perspective of human cognition, and the superiority of bionic pattern recognition is theoretically highlighted.2. Study on the mathematical tools of biomimetic pattern recognition: some basic concepts and the axioms of high dimensional geometry are presented, and the relevant properties of the points'distribution and the vertex covering in high dimensional space are discussed. Combination of high-dimensional space and the geometric method provides a guideline for the development of information science especially for the exploration on pattern recognition, therefore the complex geometric structure in high dimensional space can be achieved.3. Emphasis on the application of biomimetic pattern recognition: the performace of multi-weight neuron is a closed hypersurfaces in the high dimensional space, which simultaneously contains the function both the BP algorithm neural network and the RBF neural network. And it can complete the network cover well. By selecting a hypersausages neuron, the neural network integation can be caculated, which can be characterized as the relations between one point and two points in high dimensional space. And then constructs the geometric physique cover to carry on the recogniton of the target. A lot of experimental results show that, hypersausages networks can reach better effects about image recognition, even use the fewer samples. As the number of samples increasing, the ratio of correct recognition will be higher.