Hyperspectral Sensing Image Classification Technology Based On Active Learning

Recently, relying on the advantages of spectral information, hyperspectral remote sensinghas become an important research field of remote sensing image processing. Therefore, it isworthy that exploring the theory and application of the technology. However, hyperspectral imagealso introduces many problems: how to solve the high-dimensional problems of the hyperspectralimage, how to take advantage of the ignore information such as the structure of information, andthe vast amounts of unlabelled sample which contains useful information, and how to select aclassifier. In the increasingly demanding of hyperspectral remote sensing image classificationperformance requirements, this paper proposes a new method based on active learning forclassifying the hyperspectral image. The active learning algorithm is an effective method toconstruct a training set, and its goal is to find samples which can enhance the classificationperformance within the limited time and resources. Until now, active learning has become ahotspot issue in the fields of pattern recognition and data mining.