利用BP神经网络实现三维物体姿态的测定

本文利用BP(Back-Propagation)人工神经网络对三维物体的姿态测定进行了研究。姿态测定一直缺少通用而实际的方法,人工神经网络由于具有强大的自组织、自适应学习能力,迅速的并行信息处理能力,可望解决这个问题。但现有BP算法存在训练慢和易陷入局部最小两个问题.本文提出的级联形式网络结构,使BP网络的训练速度大为提高,陷入局部最小的可能性大为降低。利用这种级联结构对飞机模型姿态测定,取得了较好的实验结果。

ATTITUDE MEASUREMENT OF 3-D OBJECT USING BACK-PROPAGATION NEURAL NETWORK

For this topic we try to find a general and practical method. Due to its strong capabilities of self-or-ganizing, self-learning and fast parallel processing, neural network is expected to solve this problem. Afterconsidering the main disadvantages in back-propagation neural network: long-training and local-mini-mum problems, we propose an architecture of hierachically connected network. As a result, the trainingbecomes much faster and the local-minimum much scarcely appears. Some satisfactory results in the atti-tude measurement of aircraft have been obtained.