To provide an apparatus and a method for processing a signal, which can prevent a state variable from being attenuated even though each state variable of nodes of CNN is repeatedly updated, and also realizing a high speed process.
In this apparatus each node, element 1 is provided with a means 2 for storing self state variable, a means for updating the self state variable which is stored in the means 2, by using a linear updating variation value of the self attenuation term comprising the self state variable multiplied by a coefficient of the self attenuation and a joint term comprising a coupling coefficient multiplied by the difference between the state variable for the node element coupled to oneself and the self state variable, and a means for adjusting the value for the coefficient of the self attenuation so that a ratio of an impulse input to the self state variable exceeds a threshold when the self state variable is updated for required times for obtaining a computation result by using the means for updating the state variable if the impulse input is imposed on the self state variable as an initial value.
COPYRIGHT: (C)2004,JPO&NCIPI
Takashi Morie
Akira Iwata
JP2000112913A | ||||
JP2133888A | ||||
JP793277A | ||||
JP200344840A |
梅澤 淳 他,セルラーニューラルネットワーク方式ガボール型フィルタ回路の簡略化ダイナミクスの提案,電子情報通信学会2002年総合大会講演論文集 基礎・境界,社団法人電子情報通信学会,2002年 3月 7日,p.22
Bertram E.Shi,Gabor-Type Filtering in Space and Time with Celler Neural Networks,IEEE Transactions on Circuits and Systems I:Fundamental Theory and Applications,米国,IEEE,1998年 2月,Vol.45, Issue 2,pp.121-132