CAJ | 학술논문

本文将洗牌型神经网络结构和图样间联想神经网络算法相结合,提出了一种洗牌型图样间联想神经网络(PS-IPA)模型.该模型具有极其简单、稀疏的互连权矩阵,十分适于大规模神经网络的光学实现.计算机模拟结果表明洗牌型图样间联想神经网络的稳定性和抑制噪音的能力均优于图样间联想网络IPA.本文还给出了洗牌互连的一般性原则,使网络结构得到优化,增强了洗牌型神经网络的灵活性和适应性.并采用3-洗牌和2-洗牌结合的PS-IPA对汽车牌照的字符进行识别,得到了较好的结果.
본문장세패형신경망락결구화도양간련상신경망락산법상결합,제출료일충세패형도양간련상신경망락(PS-IPA)모형.해모형구유겁기간단、희소적호련권구진,십분괄우대규모신경망락적광학실현.계산궤모의결과표명세패형도양간련상신경망락적은정성화억제조음적능력균우우도양간련상망락IPA.본문환급출료세패호련적일반성원칙,사망락결구득도우화,증강료세패형신경망락적령활성화괄응성.병채용3-세패화2-세패결합적PS-IPA대기차패조적자부진행식별,득도료교호적결과.
A perfect shuffle type of interpattern association(PS-IPA) neural network model is developed by the combination of IPA with perfect shuffle (PS) interconnected architecture.A highly sparse interconnection weight matrix (IWM) with only three gray levels can be obtained from the new model,and makes it easier to realize a large-scale optical neural system.The results of computer simulations and the optical character recognition (OCR) by PS-IPA have shown improved performances compared with the IPA model.A generalized α-shuffle principle is also given, which enhances the flexibility of the perfect shuffle type of neural networks (PSNN).The vehicle license numbers in 27×16 array were recognized by our PS-IPA neural system with a hybrid 2-shuffle and 3-shuffle strategy,and good recognizing results were gained.