摘要
针对利用 BP人工神经网络在压缩空间流体实验图象数据时 BP网络收敛速度慢和存在局部极小值的问题 ,给出了一种特殊的处理方法 ,并将 BP神经网络方法与 DCT方法的图象压缩效果进行了比较 .仿真结果表明 ,不仅网络训练时间明显减小 ,而且将 BP人工神经网络用于空间流体实验图象压缩中还取得了较高压缩比及较好的重建图象质量 ,且训练好的网络鲁棒性较高 .
This paper gives the use of BP neural network for data compression of space liquid experiment.The neural network is composed of nine sub networks to form five layer networks,and each subnetwork links in series or parallel with others.In order to reduce the training time of network,the learning rate parameter and momentum parameter variety adaptively with the average of the sum of squared errors of the neural network.Moreover,random values are added to synaptic weights and threshold to avoid premature saturation or local minimum in error surface.Simulation results declare that the training time of network is shorter markedly than that with constant learning rate parameter,the network can also avoid local minimum and premature saturation.The data compression method of BP neural network gets a high compression ratio,good performance of reconstructed image and strong robustness to any image of space liquid experiment.The results show that it is feasible to compress image of space experiment with neural network technology.The performance of reconstructed image with neural network is nearly as good as that with DCT method.The time to compress an image with 512×512 pixel is shorter than that with DCT method obviously.
出处
《中国图象图形学报（A辑）》
CSCD
北大核心
2001年第3期219-222,共4页
Journal of Image and Graphics
基金
中国科学院重点项目!( KY95 -SI-5 0 7)
关键词
空间流体实验
BP人工神经网络
图象数据压缩
Space liquid experiment, BP artifical neural network,Image data compression
