论文部分内容阅读
稀疏轨迹网络利用输入模式时间序列形成类的轨迹，初步实现了空时联合不变性特征提取，但不同类别的轨迹可能重叠，不利于准确分类。本文首先将不同类别轨迹之间的距离等因素引入学习算法，对轨迹调整进行自适应优化，提出一种自适应轨迹神经网络模型：进而从神经元的特性出发，以自适应Ｓｉｇｍｏｉｄ函数包作为激活函数，改善不同类别在特征空间的分布情况，提出一种量子轨迹神经网络模型。模型在任意手写体数字识别上测试了特征提取能力，取得了较好的效果。
The sparse trajectory network uses the time series of input mode to form the trajectory of the trajectory, and initially implements the space-time joint invariant feature extraction, but the trajectories of different classes may overlap, which is not conducive to accurate classification. In this paper, we first introduce the distance between different types of trajectories and other factors into the learning algorithm to adaptively adjust the trajectory. An adaptive trajectory neural network model is proposed. Based on the characteristics of neurons, adaptive Sigmoid package is used as the activation Function to improve the distribution of different categories in the feature space, a quantum trajectory neural network model is proposed. The model tests the feature extraction ability on any handwritten digit recognition and achieves good results.