高光谱遥感图像具有丰富的光谱、纹理和空间信息，在农作物长势监测与估产以及地质勘探等领域有着广泛的应用。高光谱图像分类已成为计算机视觉以及遥感数字图像处理领域一项充满挑战的研究问题，其困难在于高维光谱信息、光谱波段间的相似性以及有限的训练数据。传统的高光谱图像分类方法，通常会单独使用其光谱信息或空间信息来进行分类，同时也需要较多的有标签的样本来进行训练，不能充分利用高光谱图像的数据特点。 为了充分利用高光谱图像光谱间的关联关系以及像素间的空间和上下文关系，本文基于亚像素、像素和超像素来提取多种级别的高光谱图像特征，并融合多级别特征实现地物的识别。为了达到这一目标，本文提出了基于像素和超像素的子空间学习，以及融合多层次特征的神经网络模型两种高光谱图像分类方法。 （1）提出了基于像素和超像素子空间学习的半监督分类方法（PSASL）。利用重建独立成分分析算法（RICA）构建子空间学习框架，在该框架中，加入光谱-空间图约束和标签空间约束，为了避免基于像素分类产生的椒盐噪声问题，同时加入了基于超像素的约束。这样，构建了联合子空间学习与像素/超像素约束的统一目标函数，利用自定义的迭代优化算法高效求解目标函数。PSASL将像素级约束、超像素级约束和分类器整合到统一的子空间学习目标函数中，分类效率高。在三个高光谱图像数据集上开展了实验，其结果表明PSASL的分类结果优于其它半监督方法。 （2）上述基于像素和超像素的子空间学习没有顾及像素内光谱和空间的相关性。为了充分利用亚像素-像素-超像素三个级别间在光谱、空间、形状和上下文信息等彼此关联和补充的信息，本文提出了用于高光谱图像分类的多级别特征融合的深度学习模型HSINet-CRF。HSINet中包含一个三层的深度神经网络模型（TDNN）和一个多级别特征卷积神经网络模型（MCNN），通过自监督的方式从亚像素级，像素级和超像素级中提取和集成多级别互补特征。为了通过概率模型增强自监督的特征学习，将条件随机场（CRF）框架嵌入到HSINet中。将条件随机场一元项、二元项和高阶项的反馈信息反馈给HSINet，以进一步增强自监督特征学习的性能。HSINet-CRF能有效地从小样本低质量的高光谱图像中获得鲁棒性特征。实验验证结果显示本文方法的分类性能要优于其他相关方法。

Hyperspectral sensors can obtain hundreds of continuous bands in a wide wavelength range, enabling hyperspectral remote sensing images to provide rich spectral and spatial information, which are widely used in agriculture, environmental science and geological exploration. The classification of hyperspectral image (HSI) has become a challenging research topic in the field of remote sensing image processing and computer vision. The difficulties lie in high-dimensional spectral information, similarity between spectral bands and limited labeled data. Traditional HSI classification approaches usually only use the spectral information or spatial information to classify the HSI and need more labeled samples for training, resulting in the data characteristics of HSI cannot be fully utilized. To obtain complementary information of multi different level in HSI, we consider subpixel, pixel, superpixel and fuse these features to get better HSI classification results. We propose two methods: pixel and superpixel-aware subspace learning (PSASL) and hyperspectral image feature learning network with CRF embedding (HSINet-CRF). To utilize spectral information and spatial correlation among pixels in HSI and avoid the “salt-and-pepper” problem generated in the pixel-based HSI classification, a novel pixel and superpixel-aware subspace learning method called PSASL is developed. The PSASL constructs the subspace learning framework based on the reconstruction ICA algorithm. The spectral-spatial graph regularization and label space regularization are developed as the pixel-level constraints. To avoid the “salt-and-pepper” problem generated in the pixel-based classification methods, superpixel-level constraints are introduced for integrating the data representations defined in the subspace and class probabilities of the pixels in the same superpixel. The subspace learning and the pixel-level regularization are combined with the superpixel-level regularization to form a unified objective function. The solution to the objective function is efficiently achieved by employing a customized iterative algorithm, and it converges very fast. The pixel-level regularization, superpixel-level regularization and a single predictive linear classifier are explicitly integrated into a unified objective function for subspace learning. The proposed method is very effective and efficient for semi-supervised HSI classification. A discriminative data representation and a universal multiclass classifier are learned simultaneously. We test the PSASL on three widely used HSI datasets. Experimental results demonstrate the superior performance of our method over many recently proposed methods in HSI classification. Considering the three different levels in HSIs, i.e. subpixel, pixel and superpixel, offer complementary information, we first develop a novel hyperspectral image feature learning network (HSINet) to learn consistent features by self-supervision for HSI classification. HSINet contains a three-layer deep neural network (TDNN) and a multifeature convolutional neural network (MCNN). It automatically extracts the features like spatial, spectral, color, and boundary as well as context information. To boost the performance of self-supervised feature learning with the likelihood maximization, the conditional random field (CRF) framework is embedded into HSINet. The potential terms of unary, pairwise and higher order in CRF are constructed by the corresponding subpixel, pixel and superpixel. Furthermore, the feedback information derived from these terms are also fused into the different-level feature learning process, which make the HSINet-CRF be a trainable end-to-end deep learning model with the back-propagation algorithm. Comprehensive evaluations are performed on three widely used HSI datasets and our method outperforms the state-of-the-art methods.