论文部分内容阅读
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。
This paper presents a model for invariance identification of two-dimensional objects. We combine the neuron networks that apply planning-based learning algorithms with complex-logarithmic preprocessed transformations to identify the invariance of size, orientation and position of an object. The planning and learning algorithm solves the learning problem of the neuron network with associative memory by using the quite mature optimization technique in planning mathematics, so that the network has the characteristics of large capacity, stable training samples and optimized attracting radius. The interconnected network of associative memory can not only identify the object according to the result of the preprocessing, but also estimate the change of the object in size and direction. In this paper, some experiments are carried out to identify the workpieces on the manipulator platform. The experimental results are given and the preliminary work on how to recognize the invariance of the three-dimensional object by combining the model with the robot system in the hand is discussed.