高光谱影像的DAE分类

针对高光谱影像非线性分类问题,根据高光谱影像光谱分辨率高且光谱具有非线性的特点,结合深度学习理论,提出了一种采用降噪自动编码器(DAE)的高光谱影像分类方法。该方法结合降噪自动编码器与SOFTMAX分类器,构造深层网络分类模型;然后,利用加噪后的光谱数据,采用Dropout方法对分类模型进行预训练和微调;最后,利用训练得到的网络模型学习高光谱影像光谱的隐含特征,实现高光谱影像的分类。采用该方法对AVIRIS和PHI的高光谱影像分别进行分类对比实验,结果表明该方法能有效提高高光谱影像分类精度。

Classification of Hyperspectral Imagery Based on Denoising Autoencoders

Hyperspectral imagery has high spectral resolution, and its spectrum is always been non-linear. Com-bined with the theory of deep learning, a new hyperspectral classification method based on denoising autoencoders ( DAE) is proposed. Firstly, combining denoising autoencoders with SOFTMAX classifier, a deep network model is constructed. Then, the spectrum with noise is used as input signal, and Dropout method is utilized to train the model in pre-training and fine-tuning processes. Finally, using the trained network model, the implicit feature of spectrum of hyperspectral imagery can be learned and the classification of hyperspectral imagery is achieved. Ac-cording to two comparative experiments of AVIRIS and PHI hyperspectral imagery, the results indicate that the proposed method is effective to improve the hyperspectral image classification accuracy.