双层通用自适应模式识别系统

单层n元自适应模式识别系统(WISARD)由于其大规模并行分布处理能力、通用性和自适应性而显示了巨大的应用潜力。但由于其结构的原因,系统性能受到了一定的限制。Kanerva提出的稀疏存贮器结构和WISARD的RAM阵列结构有相似之处。而且这种稀疏分布存贮的原理,可以从结构上解决WISARD的固有缺陷。根据SDM的概念,在WISARD的基础上,本文提出了一种新颖的双层自适应模式识别系统模型。它除了保持WISARD的原有特点之外,在解决大维数或非确定性模式数据的识别问题以及控制系统成本方面有着明显的优势。在该系统上进行的多体印刷汉字识别试验,初步优化了系统各参数的关系,验证了模型的可行性。

Two-Layered General-Purpose Adaptive Pattern Recognition System

Single-layered N -tuple adaptive pattern recognition system has found extensive applications in various areas because of its massive parallel distributed processing ability, versatility and self-adaptability. It has, however, the weakness in the system structure. The architecture of the sparse distributed memory-SDM, proposed by Kanerva, P. has some similarity to that of WISARD, The method of its distributed storage can overcome the limitations of WISARD. The paper presents a novel two the-layered pattern recognition system on the basis of both WISARD and SDM. It has obvious advantages in large dimensional or non-deterministic pattern data recognition problems in addition to all merits of WISARD. Experiments of the multifont Chinese character recognition have optimized the system parameters and shown the feasibity of the model.