基于BP神经网络改进算法的数据压缩方案

Data Compressions Methods Based on Modified Algorithm of BP Neural Network

论述了BP神经网络进行图像数据压缩的基本原理;为了进一步提高算法的收敛速度,分析了样本数据输入顺序对权值的影响、学习参数的选择及局部极小这三个影响算法收敛速度的重要问题;提出了消除样本数据输入顺序对权值的影响、如何避免局部极小的问题、学习参数自适应调整的改进算法.

The basic principle of image compression is presented, which is realized by improved BP neural network. The three factors that influence the training speed are discussed, and how to choose the beginning weight is described, and some theoretic are discussed for the choice of the learn parameter. And the paper shows in detail how to avoid the algorithm plunge into the local minimum. These analyses can help us to understand the reason for the slow algorithm, and will do help to catch the main thought of the extended BP algorithm and the improvelop BP algorithm.