| 基于Hopfield神经网络的平面拟合编码研究 |
| |
| 引用本文： | 王卫,蔡德钧,万发贯.基于Hopfield神经网络的平面拟合编码研究[J].华中科技大学学报(自然科学版),1994(10). |
| |
| 作者姓名： | 王卫 蔡德钧 万发贯 |
| |
| 作者单位： | 华中理工大学电子与信息工程系 |
| |
| 基金项目： | 国家自然科学基金,中科院自动化所国家模式识别实验室资助项目 |
| |
| 摘 要： | 提出了平面拟合编码的一种新的实现方法，即神经网络方法。为了保证Ｈｏｐｆｉｅｌｄ神经网络的收敛，对该网络模型的迭代算法进行了修改，针对Ｈｏｐｆｉｅｌｄ网络存在的局部极小问题，给出了一种扰动算法，结合初始状态的合理选择，可以有效地避免网络陷入局部极小，而接近全局最小，以求得待定系数的最优解，计算机模拟结果表明，Ｈｏｐｆｉｅｌｄ神经网络实现的平面拟合编码性能优于传统的最小二乘法，重建图像质量提高约０．６ｄＢ。
|
| 关 键 词： | Hopfield神经网络，平面拟合编码，分块编码，能量函数，最小二乘法，局部极小，扰动技术 |
A Study of Plane Fitted Encoding Based on Hopfield Neural Network |
| |
| Wang WeiDept.of Electronics & Information Engin,H.U.S.T,Wuhan ,China.,Cai Dejun ,Wan Fanguan.A Study of Plane Fitted Encoding Based on Hopfield Neural Network[J].JOURNAL OF HUAZHONG UNIVERSITY OF SCIENCE AND TECHNOLOGY.NATURE SCIENCE,1994(10). |
| |
| Authors: | Wang WeiDeptof Electronics & Information Engin HUST Wuhan China Cai Dejun Wan Fanguan |
| |
| Institution: | Wang WeiDept.of Electronics & Information Engin,H.U.S.T,Wuhan 430074,China.,Cai Dejun ,Wan Fanguan |
| |
| Abstract: | A novel approach to plane fitted encoding,or the neural network approach,is present-ed.In order to guarantee the convergence of Hopfield neural network,the updating rule hasbeen modified.A perturbation technique is developed to specially avoid the local minimumexisting in Hopfield neural network and,together with a proper selection of the initial state,to give either a global minimum or a local minimum that is very close to the global minimum.The results of computer simulation show that the performance of plane fitted encoding by theneural network approach is better than that by the least square method and the quality of thereconstructed image is improved by about 0. 6 dB. |
| |
| Keywords: | Hopfield neural network plane fitted encoding block encoding energy function least square method local minimum perturbation technique |
| 本文献已被 CNKI 等数据库收录！ |
