The Geometry Of The Biomimetic Pattern Recognition Learning Theory

With the further development of algebra, the analytic geometry, differential geometry, differential equations and a variety of geometry in abstract space are created and has risen the fundamental shape concept in our brain to an abstract mathematical expression to accurately analyze problems by mathematical formula, however, the current data set is characteristically large amount, high dimension, nonlinear, so far there is no general mathematical analytical methods. The traditional high dimensional data analysis are based on linear methods, such as Principal Component Analysis (PCA), Independent Component Analysis (ICA), Fisher Discriminant Analysis (FDA), Multidimensional Scaling (MDS), etc. This usually adopted global linear space that is Euclidean space is used as a geometric space of dataset, which has a good effect on the linear structural data set. Using the traditional ways, we often overlook the local even global property of datebase in the observation space so that we can not really understand the internal essential structure of these nonlinear structural data information, then we need to find a new machine learning algorithms to study the geometric characteristics of the sample data set in high dimensional space. At present many scholars start from the geometric properties of the data set, has conducted a lot of analysis research on data, which impels the rise and development of geometric learning algorithm.Information from visual is the most important among all the sensory organs.80% information obtained from the outside relies on eyes. The geometric figures, which are received by human retina according to the differences of ray reflected by the geometric surface of objects in nature, are automatically combined into an object image by brain neurons. We can identify the objects from these images. In popular words, when we contact with things, firstly we will observe the appearance of things, and immediately judge whether we know such things by the geometry shapes, then we divide them into different categories through some other specific features.With the development of science and technology, the computer vision system, which simulates biological vision system using computers, becomes more and more important and has been an important branch of computer science and artificial intelligence. Most of the data we have in reality is three-dimensional or even higher, so it is complicated to deal with. The biomimetic pattern recognition has got the good application in many areas such as speech, image, face recognition, control theory, DOA estimation ect. since it is proposed. This paper firstly describes some characteristics of neurons, according to these characteristics we know neural network is the very suitable means to realize biomimetic pattern recognition, therefore this paper again gives a brief introduction on several neural network models, they are: ABF Neural Network, RBF neural network, DBF neural networks, Two-weight neural networks and Hypersausage neural networks.The main purpose of this paper is to get a more direct and effective geometry learning recognitive method based on the biomimetic pattern recognition algorithm and combined with geometric transformation of three dimension body, the projective mapping and projection transformation from three-dimensional space to two-dimensional space and the geometric characteristics of two-dimensional image.Anything can be briefly described by the unique length to width ratio of the front and side of the two planar images. The different characteristics of objects could be distingushed through simple length to width ratio. Basing on the experiments and taking the vehicle recognition for example, this paper is to distingush different types of cars or even different types of things adopting the geometric characteristics of two-dimensional images and put forward the way of biomimetic recognition.