本文提出了一种自学习神经网络匹配算法(SLNNMA).该算法基于图像特征点的提取,通过二维神经网络来实现图像匹配.SLNNMA有别于其它神经网络算法(如IPBNNMA)的显著特点在于:通过物理因子(Physical Factor)的引入,具有自学习的机制,其极值受初始条件的限制程度大大减弱.SLNNMA能够很好地对含有各种面目标(包含在相邻图像之间进行随机大幅度平移运动,或旋转运动,或放大变化,或包含上述三种变化的混合运动的目标)的序列图像通过特征点的匹配来进行目标跟踪. In this paper, a self-learning neural network matching algorithm (SLNNMA) is proposed, which is based on the extraction of image feature points and two-dimensional neural network to achieve image matching.SSLNNMA distinguishes itself from other neural network algorithms (such as IPBNNMA) : The mechanism of self-learning through the introduction of physical factors (Physical Factor), the extreme value of which is greatly reduced by the initial conditions.SLNNMA can well contain a variety of surface targets (including random images between adjacent A large amount of translational motion, or a rotational motion, or a magnified variation, or a target that includes a mixed motion of the above three variations) is targeted tracking by feature point matching.