应用基于规划的学习算法建立对物体进行不变性识别的模型

A MODEL FOR INVARIANT OBJECT RECOGNITION USING PROGRAMMING-BASED LEARNING ALGORITHM

本文给出了一个对二维物体进行不变性识别的模型。我们把应用基于规划的学习算法的神经元网络和复数—对数预处理变换结合起来对物体进行大小、方向和位置的不变性识别。规划学习算法是用规划数学中相当成熟的优化技术求解联想记忆的神经元网络的学习问题,从而使网络具有容量大、训练样本稳定、吸引半径得到优化等特点。联想记忆的互连网络根据预处理的结果不仅可以识别物体,还可以估计出物体在尺寸和方向上的变化量。本文进行了一些实验识别机械手操作平台上的工件,给出了实验结果并讨论了把该模型与眼在手上的机器人系统相结合用来实现三维物体的不变性识别的初步工作。

A model for 2-D object recognition is described. Programming-based learning algorithm uses rather mature optimization technique to solve the learning problem of neural network with self-feeedback connections. We discuss its performance in associative memory which is combined with complex -log conformal mapping to create a system for recognizing objects independent of size,orientation and position. Recalled information from the associative memory is used to classify an object and estimate the magnitude of changes in scale and rotation. Some experiments for recognizing industrial parts on robotic platform are presented. Future work about 3-D invariant object recognition using eye-on-hand robotic system is discussed.