协同神经网络的不变性研究

Research on Spatial Invariant of Synergetic Neural Network

协同神经网络作为一种全新的神经网络，它的自上而下的构造方式，与自然界中自组织现象存在深刻的相似性．文中在介绍了模式识别协同神经网络的基础上，研究了协同神经网络用于模式识别的空间不变性．通过二维傅氏变换、复对数映射等方法，对图象进行预处理，提取图象的空间不变性，并在变换后的空间上利用协同神经网络进行识别．试验表明，协同神经网络不但能够识别空间变化的图象，并且对缺损、加噪图象也能很好识别，并且识别速度快，鲁棒性强，不会出现传统神经网络经常出现的伪状态．

The synergetic neural network is a novel kind of neural network, whose top down construction lies a great similarity to self organized phenomena in nature. Based on the concept of pattern recognition approach to synergetic neural network, the recognition of space variant images is studied. An image preprocessing procedure, using 2D FFT and Complex Log Mapping, enables synergetic recognition to be simultaneously invariant to spatial pattern translation, rotation, and scaling. The test shows that by using the synergetic neural network, the recognition speed is high, and the robustness of the synergetic system is strong. And there is no spurious state, which often occurs in traditional neural networks. Application of the concepts of synergetics in recognition is in full swing, and the synergetic neural network will certainly lead to far reaching understanding for recognition.