摘要
利用神经网络的结构特征及良好的数据处理能力、在数据压缩领域拥有的先天优势,采用性能优良的BP算法构建网络模型。分析BP网络压缩的原理,搭建压缩系统模型,分别使用自适应学习率梯度下降法和BFGS拟牛顿法训练网络,从而实现交通图像的压缩与重建。实验结果表明,基于BFGS拟牛顿法的BP神经网络收敛速度快,压缩性能优良,获得低压缩率的同时重建图像视觉效果良好。
By utilizing the structural characteristics of the neural network and its good data processing ability and innate advantages in the field of data compression,the network model is constructed using BP algorithm with excellent performance.The principle of BP network compression is analyzed,and the compression system model is built.The adaptive learning rate gradient descent method and BFGS quasi-Newton method are used to train the network respectively,so as to realize the compression and reconstruction of traffic image.The experimental results show that the BP neural network based on BFGS quasi-Newton method has fast convergence speed,good compression performance,low compression rate and good visual effect.
作者
罗山
Luo Shan(School of Traffic and Automotive Engineering,Panzhihua University,Panzhihua Sichuan 617000,China)
出处
《山西电子技术》
2019年第6期31-33,共3页
Shanxi Electronic Technology
关键词
神经网络
交通图像压缩
BP算法
自适应学习率梯度下降法
BFGS拟牛顿法
neural network
traffic image compression
BP algorithm
adaptive learning rate gradient descent method
BFGS quasi-Newton method
作者简介
罗山(1979-),男,四川乐至人,讲师,硕士,主要研究方向:图像处理与模式识别在智能交通中的应用。
