高光谱图像分类是利用高光谱数据图谱合一且光谱信息丰富的特点, 对图像中的每个像素进行分门别类, 以达到对地物目标进行高精度分类和自动化识别的目的, 是对地观测的重要组成部分。在分析高光谱图像特点的基础上, 本文从普通机器学习和深度学习这两方面对高光谱图像像素级分类的研究进展及效果进行总结、评述和比较, 通过具体实验的结果对比, 直观地展现各种算法的优劣。针对高光谱分类问题, 本文从两个方面对今后的研究方向及发展前景进行了分析和展望。一方面, 在算法研究上, 高光谱图像分类算法可在保证分类精度的前提下降低算法的复杂度, 利用多源遥感数据、多特征综合、多尺度复合, 提升小样本、少参数分类模型的分类精度, 适应智能化、快速化高光谱遥感对地观测的发展要求; 另一方面要紧密结合市场应用需求, 重视高光谱图像在实际中的应用, 研究具有市场竞争力的高效分类算法, 提升高光谱图像分类在遥感技术应用领域的竞争力。

Hyperspectral image classification comprises the classification of every pixel in an image by applying the combination of hyperspectral data atlas and rich spectral information, which can be employed for achieving high-precision classification and automatic recognition of ground objects. Hyperspectral image classification plays an important role in earth observation. Based on the analysis of the characteristics of hyperspectral images with respect to two aspects of general machine learning and deep learning, the progress in associated research and comparison of the effects of pixel-level classification of hyperspectral images are summarized and discussed in this study. The advantages and disadvantages of various algorithms were visually illustrated by comparing the corresponding results. Research objectives and development prospects of hyperspectral image classification are analyzed with respect to two aspects. Firstly, various algorithms need to be studied. A hyperspectral classification algorithm can guarantee classification accuracy required for reducing the algorithm complexity by incorporating multi-source remote sensing data with multi-feature and multi-scale composites. Such an algorithm can improve the classification accuracy of a small sample of the classification model with few parameters, and it can adapt to the intelligent and rapid development requirements of earth observation. Secondly, market applications need to be closely integrated. Practical applications of hyperspectral images should be considered and efficient classification algorithms with marketable competency should be investigated for enhancing the applicability of hyperspectral image classification in remote sensing applications.