论文部分内容阅读
分析了ART网络在模式识别应用中的特性.在原ART1网络中,由于子集对包集的重新编码过程使得ART1网络具有学习过程不稳定性,从而影响了在任意输入环境下网络的分类性能,为此本文对ART1网络的匹配度计算提出改进方法,使匹配度计算公式更合理地反映了输入模式与模板模式的相似程度,有效地克服了学习过程不稳定性。在ART2网络中,本文根据同一类别中最大、最小模式数提出了警戒线户值的自适应调整方法,避免在固定户值下可能引起的分类过粗或过细,使网络具有一定容错能力,又有一定敏感性。
In the original ART1 network, the ART1 network has instability of the learning process due to the sub-set re-encoding of the packet set, which affects the classification performance of the network in any input environment, Therefore, this paper proposes an improved method to calculate the matching degree of ART1 network so that the matching degree calculation formula reflects the similarity degree of input mode and template mode more reasonably, and overcomes the instability of learning process effectively. In ART2 network, this paper proposes a method of adaptive adjustment of alert line value based on the maximum and minimum number of modes in the same category. This method avoids the classification that may be caused by fixed user value is too coarse or too thin, which makes the network have some fault tolerance. Have a certain sensitivity.