论文部分内容阅读
由于高光谱遥感数据具有波段多、特征非线性、空间相关等特点,提出一种基于深度学习的空-谱联合(SSDL)特征提取算法来有效提取数据中的空-谱特征。该算法利用多层深度学习模型——堆栈自动编码机对高光谱数据进行逐层学习,挖掘图像中的深层非线性特征,然后再根据每个特征像元的空间近邻信息,对样本深度特征和空间信息进行空-谱联合,增加同类数据聚集性和非同类数据分散度,提升后续分类性能。在帕维亚大学和萨利纳斯山谷高光谱数据集上进行地物分类实验:在1%样本比例下,地物总体分类精度达到了91.05%和94.16%;在5%样本比例下,地物总体分类精度达到了97.38%和97.50%。结果表明:由于SSDL特征提取算法融合了数据中深层非线性特征和空间信息,能够提取出更具鉴别特性的特征,较其他同类算法能够获取更高分类精度。
Because hyperspectral remote sensing data has the characteristics of multi-band, non-linearity and spatial correlation, this paper proposes a space-spectrum combining (SSDL) feature extraction algorithm based on depth learning to extract the spatio-temporal features effectively. The algorithm uses multi-layer depth learning model-stack auto-coder to learn the hyperspectral data layer by layer, and then digs the deep nonlinear features in the image. Then based on the spatial neighbor information of each feature pixel, Spatial - spatial information - spectrum combination, increase the same type of data aggregation and dispersion of non-similar data to improve the follow-up classification performance. At the University of Pavia and the Salinas Valley hyperspectral dataset, the classification of terrain was performed: at the 1% sample scale, the overall classification accuracy of terrestrial objects reached 91.05% and 94.16%; at the 5% sample scale, Overall classification accuracy reached 97.38% and 97.50%. The results show that the SSDL feature extraction algorithm can extract features with more discriminative features because of the fusion of deep nonlinear features and spatial information in the data, and can obtain higher classification accuracy than other similar algorithms.