机器学习理论的超光谱遥感图像无损压缩

Lossless compression of hyperspectral remote sensing image based on machine learning theory

针对一般超光谱遥感图像的压缩方法无法同时实现图像信息缩减和图像完整性的问题,提出一种机器学习理论的超光谱遥感图像无损压缩方法。利用机器学习中的聚类算法进行第一次压缩,减少超光谱遥感图像中的冗余波段光谱,并降低图像维度;再利用机器学习中的人工神经网络进行第二次压缩,将不同图像子块送入不同压缩率的神经元当中,通过隐含层自主完成图像压缩编码。通过与四种一般压缩方法的对比验证,本方法图像压缩后,图像压缩率更小,图像分辨率和信息熵更高,既有效地减少了图像信息量,能够保留有效关键信息,达到了图像信息缩减和图像完整性的双重目标。

Aiming at the problem that the general compression method of hyperspectral remote sensing image can not realize image information reduction and image integrity at the same time, a lossless compression method of hyperspectral remote sensing image based on machine learning theory is proposed. The clustering algorithm in machine learning is used for the first compression to reduce the redundant band spectrum in hyperspectral remote sensing image and reduce the image dimension;Then the artificial neural network in machine learning is used for the second compression, different image sub blocks are sent to neurons with different compression rates, and the image compression coding is completed independently through the hidden layer. Compared with four general compression methods, the method in this paper has smaller image compression rate, higher image resolution and information entropy. It not only effectively reduces the amount of image information, but also retains effective key information, and achieves the dual goals of image information reduction and image integrity.