在ART2神经网络的标准警戒测试准则中，通过引入截断双曲线函数来计算输入矢量与神经网络由顶向下权重矢量之间的相似程度，而提出了一种新的具有更严格警戒测试准则的ART2神经网络。截断双曲线函数一方面抑制输入样本中的噪声，另一方面，如果输入矢量某些分量与由顶向下权重矢量对应分量之间存在冲击变化时，则截断双曲线函数将放大这些对应分量之间的冲击变化。而且这种新的警戒测试准则具有更强的抗噪声能力。即在较低的输入信噪比水平上，具有更严格警戒测试准则的ART2神经网络比标准ART2神经网络具有更高的正确识别率。

The neural network models based on adaptive resonance theory(ART) are capable of organizing stable recognition categories for arbitrary binary or real imput patterns. However the ART neural networks are not sensitive to the distinguishing of those categories which there are only a few components in obvious difference between them. A new ART2 neural network model with more vigorous vigilance test criterion is proposed in this paper. The modified intercepted hyperbola function is adopted in the new vigilance test criterion to calculate the matching degree between the input vector and the weight vector of top to bottom. On one hand, the hyperbola function reduces the effect of noises and on the other hand, it emphasizes the effect of those components of input vector which have impulsive differences to the corresponding components of weight vector. The palm image recognition using the ART neural networks with the new vigilance test criterion has been carried out in this paper, the experiment results shows that the new vigilance test criterion is more robust to noises, and the new ART2 neural network can gain higher recognition rate under lower SNR.