基于神经网络的多光谱遥感图像无损压缩

Lossless Compression of Multispectral Remote Sensing Image Based on Neural Network

分析并改进了利用自组织特征映射(SOFM)神经网络设计码书的方法,提出了一种基于改进SOFM算法设计码书的矢量量化和分类谱间预测相结合的多光谱图像无损压缩方法。该方法对光谱信息进行矢量量化,根据分类信息生成残差图像以去除数据的空间相关性,构造分类谱间预测器去除数据的谱间结构和统计相关性。对机载64波段多光谱遥感图像的试验结果表明,该方法无论是对训练集内图像还是训练集外图像,均取得了较好的压缩效果,平均无损压缩比达到3.2以上。

The algorithm of self-organizing feature mapping neural network is analyzed and improved. A new method based on SOFM codebook design for lossless compression of multispectral image is developed. This method combines vector quantization and classified prediction technique. At first, the multispectral images are transformed to quantization form. Then, residual images are produced and predicted according to classified map. The method removes the intra-band spatial redundancy and the inter-band structural and statistic redundancy, so the better compression results can be obtained. The experimental results by using practical 64-band multispectral images have shown that the lossless compression ratio achieved by the method is not less than 3.2, better than LBG method.