势能记忆梯度优化的协同模式分类方法
|
摘要
传统的协同模式分类学习方法是依据原型向量再通过伪逆或M-P广义逆的方法求出满足一定关系式的伴随向量,当样本维数大时,这种方法学习过程较慢,特别当样本维数有变化时传统的方法就不太适用了;协同势能函数优化的方法是直接利用协同动力学过程,来获得原型向量和伴随向量的收敛值,相比于传统的方法具有一定的优势。将最优化理论引入到协同进化的动力学过程,以加快学习过程的收敛,并以记忆梯度法替代了传统的梯度下降的算法进行势能函数的优化,来同时进行原型向量和伴随向量的学习,新方法能显著地提高收敛速度并获得较优的原型向量。通过图像的分类识别表明,相对于传统的方法,能提高识别率且收敛更好。
The traditional training method of synergetic pattern recognition is figured out adjoint vector from prototype vector according to certain rules for example pseudo-inverse or M-P generalized-inverse.The course is slowly to the high-dimension samples,and is not fit to the change of dimension.The studying of potential energy function dynamics process can train prototype vector and adjoint vector meanwhile.The optimization approach is introduced to synergetic dynamics evolution process, using the memory gradient algorithm instead of the steepest decent algorithm to get optimization prototype vector.The prototype vector experimental result on images shows that the new algorithm can effectively search the prototype vector and ad- joint vector meanwhile,and excellent,correct and fast recognition result shows the new algorithm is more available than traditional training method.
The traditional training method of synergetic pattern recognition is figured out adjoint vector from prototype vector according to certain rules for example pseudo-inverse or M-P generalized-inverse.The course is slowly to the high-dimension samples,and is not fit to the change of dimension.The studying of potential energy function dynamics process can train prototype vector and adjoint vector meanwhile.The optimization approach is introduced to synergetic dynamics evolution process, using the memory gradient algorithm instead of the steepest decent algorithm to get optimization prototype vector.The prototype vector experimental result on images shows that the new algorithm can effectively search the prototype vector and ad- joint vector meanwhile,and excellent,correct and fast recognition result shows the new algorithm is more available than traditional training method.
引文
[1]Daffershofer A,Haken H.Synergetic computers for pattern recognition-a new approach to recognition of deformed patterns[J].Pattern Recognition,1994,27(12).
[2]Hinton G E.Connectionist learning procedures[J].Artificial Intelli-gence,1989,40(1/3):185-234.
[3]刘秉瀚,王伟智,方秀端.协同模式识别方法综述[J].系统工程与电子技术,2003,25(6):758-762.
[4]邹刚,姚伟,敖永红,等.一种粒子群优化原型模式修正力度的协同分类方法[J].信号处理,2010,26(4):558-562.
[5]韩绍卿,李夕海,宋仔标,等.基于模糊C-均值的原型模式选择及其在核爆地震识别中的应用[J].核电子学与探测技术,2007,27(5):820-821.
[6]陈丽,戚飞虎.基于梯度动力学的协同神经网络学习算法的改进[J].计算机工程与科学,2005,27(1):43-44.
[7]袁亚湘,孙文瑜.最优化理论与方法[M].北京:科学出版社,1997.
[8]栗塔山,彭维杰,周作益,等.最优化计算原理与算法程序设计[M].长沙:国防科技大学出版社,2002.
[9]Wang C Y,Du S Q,Chen Y Y.Global convergence properties of three-term conjugate gradient method with new-type line search[J].Journal of Systems Science and Complexity,2004,17(3):412-421.
[10]汤京永,时贞军.一类全局收敛的记忆梯度法及其线性收敛性[J].数学进展,2007(1):67-69.
[11]缑水平,焦李成,田小林.基于免疫克隆聚类协同神经网络的图像识别[J].电子与信息学报,2008,30(2):264-265.
[12]Ma X L,Jiao L C.An effective learning algorithm of synerget-ic neural network[C]//Lecture Notes in Computer Science,2004,3173:258-263.
[13]王伟智,刘秉瀚,朱敏琛.基于协同方法交通状态识别[J].中国体视学与图像分析,2007,12(1):38-39.
[14]王海龙.协同神经网络在图像识别中的应用研究[D].上海:上海交通大学,2000.
[15]Shi Zhenjun,Shen Jie.Convergence of PRP method with new nonmonotone line search[J].Applied Mathematics and Computa-tion,2006,181(1):423-431.
[16]Liu Binghan,Wang Weizhi,Zheng Zhiyong.Lymphocyte syner-getic classification based on competition optimum prototype[J].Chinese Journal of Stereology and Image Analysis,2006,11(2):71-75.
[17]Zou Gang,Zhou ShiLin,Yao Wei,et al.An optimal reconstruc-tion method of synergetic order parameters based on orthogo-nal polynomials[C]//The IEEE International Conference on Ad-vanced Computer Control.[S.l.]:IEEE Press,2010,5:291-295.
[2]Hinton G E.Connectionist learning procedures[J].Artificial Intelli-gence,1989,40(1/3):185-234.
[3]刘秉瀚,王伟智,方秀端.协同模式识别方法综述[J].系统工程与电子技术,2003,25(6):758-762.
[4]邹刚,姚伟,敖永红,等.一种粒子群优化原型模式修正力度的协同分类方法[J].信号处理,2010,26(4):558-562.
[5]韩绍卿,李夕海,宋仔标,等.基于模糊C-均值的原型模式选择及其在核爆地震识别中的应用[J].核电子学与探测技术,2007,27(5):820-821.
[6]陈丽,戚飞虎.基于梯度动力学的协同神经网络学习算法的改进[J].计算机工程与科学,2005,27(1):43-44.
[7]袁亚湘,孙文瑜.最优化理论与方法[M].北京:科学出版社,1997.
[8]栗塔山,彭维杰,周作益,等.最优化计算原理与算法程序设计[M].长沙:国防科技大学出版社,2002.
[9]Wang C Y,Du S Q,Chen Y Y.Global convergence properties of three-term conjugate gradient method with new-type line search[J].Journal of Systems Science and Complexity,2004,17(3):412-421.
[10]汤京永,时贞军.一类全局收敛的记忆梯度法及其线性收敛性[J].数学进展,2007(1):67-69.
[11]缑水平,焦李成,田小林.基于免疫克隆聚类协同神经网络的图像识别[J].电子与信息学报,2008,30(2):264-265.
[12]Ma X L,Jiao L C.An effective learning algorithm of synerget-ic neural network[C]//Lecture Notes in Computer Science,2004,3173:258-263.
[13]王伟智,刘秉瀚,朱敏琛.基于协同方法交通状态识别[J].中国体视学与图像分析,2007,12(1):38-39.
[14]王海龙.协同神经网络在图像识别中的应用研究[D].上海:上海交通大学,2000.
[15]Shi Zhenjun,Shen Jie.Convergence of PRP method with new nonmonotone line search[J].Applied Mathematics and Computa-tion,2006,181(1):423-431.
[16]Liu Binghan,Wang Weizhi,Zheng Zhiyong.Lymphocyte syner-getic classification based on competition optimum prototype[J].Chinese Journal of Stereology and Image Analysis,2006,11(2):71-75.
[17]Zou Gang,Zhou ShiLin,Yao Wei,et al.An optimal reconstruc-tion method of synergetic order parameters based on orthogo-nal polynomials[C]//The IEEE International Conference on Ad-vanced Computer Control.[S.l.]:IEEE Press,2010,5:291-295.
版权所有：© 2023 中国地质图书馆 中国地质调查局地学文献中心