基于自适应波段聚类主成分分析和反向传播神经网络的高光谱图像压缩

高光谱遥感图像具有丰富的光谱信息，数据量大。为了能够有效地利用高光谱图像数据，促进高光谱遥感技术的发展，该文提出一种基于自适应波段聚类主成分分析(PCA)与反向传播(BP)神经网络相结合的高光谱图像压缩算法。算法利用近邻传播(AP)聚类算法对波段进行自适应聚类，对聚类后的各个分组分别进行PCA运算，最后利用BP神经网络对所有主成分进行编码压缩。该文的创新点在于BP神经网络压缩图像时，在训练步骤过程中，误差反向传播是用原图与输出作差值，再反向调整各层的权值、阈值。对高光谱图像进行波段聚类，不仅能够有效地利用谱间相关性，提高压缩性能，还可以降低PCA的运算量。实验结果表明，该文算法与其它现有算法比较，在相同压缩比下，其光谱角更小，信噪比更高。

Hyperspectral Image Compression Based on Adaptive Band Clustering Principal Component Analysis and Back Propagation Neural Network

Hyperspectral remote sensing images have a wealth of spectral information and a huge universe of data. In order to utilize effectively hyperspectral image data and promote the development of hyperspectral remote sensing technology, a hyperspectral image compression algorithm based on adaptive band clustering Principal Component Analysis (PCA) and Back Propagation (BP) neural network is proposed. Affinity Propagation (AP) clustering algorithm for adaptive band clustering is used, and PCA is performed on the each band group respectively after clustering. Finally, all principal components are encoded and compressed by BP neural network. The innovation point lies in BP neural network compressed image during the training step, the error of backpropagation is to compare difference between the original image and the output image, and then adjust the weight and threshold of each layer in the reverse direction. Band clustering of hyperspectral images can not only effectively utilize the spectral correlation and improve the compression performance, but also reduce the computational complexity of PCA. Experimental results investigate that the proposed algorithm achieve a better performance on Signal-to-Noise Ratio (SNR) and spectral angle than other algorithm under the same compression ratio.