【作者】 王新；

【导师】 马义德；

【作者基本信息】 兰州大学 ， 电路与系统， 2010， 硕士

【摘要】 脉冲耦合神经网络(Pulse Coupled Neural Network, PCNN)是一种新型神经网络,是依据猫大脑视觉皮层神经细胞的同步脉冲发放现象提出的,是一种更接近生物视觉特性的神经网络模型,具有相似神经元同步点火特性,其应用受到广泛关注。作为非线性动力学系统,PCNN混沌特性研究的相关文献较少。目前大量文献关注PCNN模型的仿真实现研究,而对于基于硬件平台的PCNN算法实现仍是发展中的热点,是近年来文献关注的焦点。本论文主要进行PCNN混沌特性研究和PCNN最大熵分割算法的FPGA硬件实现研究。具体如下：1.通过对PCNN单个神经元模型的参数选取及非线性特性分析,分析了PCNN产生混沌特性的模型。对混沌模型如何稳定控制到期望点及对参考信号的跟踪方法进行了研究,实验结果表明,PCNN能产生混沌现象,且能按照期望点进行稳定控制,并能对参考信号进行很好地跟踪。2.结合PCNN在图像分割方面的优势,研究了PCNN对三类图像的分割,并利用最大熵原则选取最佳分割图像。相对于PC平台上MATLAB仿真实现实时性较弱的缺点,研究了基于FPGA平台PCNN二值图像分割的硬件实现,其具有速度快、实时性好等优点,并能取得和MATLAB软件仿真几乎相同的分割结果。相比于传统的PCNN硬件实现,FPGA的实现速度得到显著提高,有着CMOS无法比拟的优势,具有较高应用价值。更多还原

【Abstract】 Pulse Coupled Neural Network (PCNN) is a new kind of neural network. It was proposed according to the phenomenon of synchronization of impulses which was derived from the observation on the visual cortical neurons of cats, and its feature that similar neurons fire (generate pulses) synchronously is closer to the biological visual characteristic, so the applications of PCNN have caused extensive attention. As a kind of nonlinear dynamic system, PCNN has the chaotic characteristic, but there are few literatures about it. Most of current literatures pay close attention to the research into the simulation of PCNN model, and the PCNN algorithm implementation based on hardware platforms is still a developing hotspot and is becoming a focus of research.This dissertation focuses on the research into the chaotic characteristic of PCNN and the implementation under FPGA of the maximum entropy segmentation algorithm based on PCNN. The content can be divided into two parts concretely:1. By means of analyzing the parameter selection of a single neuron in PCNN and the nonlinear characteristic of PCNN, the PCNN model which can generate chaotic characteristic is chased down and analyzed. The stability control to the desired point of the chaotic model and tracking the reference point are studied. Experimental results show that PCNN can generate chaos, can reach the stable control according to desired control points, and can track the reference signal accurately.2. Taking the advantages of PCNN in image segmentation into account, the segmentation of three types of images based on PCNN is studied and the optimally segmented image is selected on the basis of the principle of maximum entropy. Compared to the algorithm simulation under MATLAB, the implementation of binary image segmentation based on PCNN is characteristic of high speed and strong real-time under the hardware platform of FPGA, while almost the same segmented images can be obtained. Moreover, compared to the traditional hardware implementation of PCNN, the speed under FPGA is increased significantly, so FPGA has the incomparable advantages over CMOS and is of better applied value.更多还原