科研之友
微信
新浪微博
Facebook
分享链接
摘要
Recently, hyperspectral image classification is attracting more and more attention. Since every pixel is represented by a high dimensional spectral vector, the ordinary machine learning algorithms usually require a large number of training samples to solve this problem. However, collecting labeled samples is time-consuming, which forces us to improve existing algorithms. Motivated by evolutional algorithms (EAs), we propose location based DNA matching algorithm for hyperspectral image classification. It aims mainly on the shortcomings such as requirement for large number of labeled samples and inseparable spectral values. It is based on EA and can be segmented into three subtasks. In the first encoding procedure, some spatial information is added to the spectral values to solve the problem of spectral mixture to some extent. In the second evolutional procedure, we introduce elite-preserving strategy and totally random operators within a specific exemplar, which can prevent deterioration and can also search for solutions in a large space. Aiming at the compared pixel-wise algorithms will end up with lots of mislabeled points in a region, we add the third procedure which utilizes some labeled samples' locations to optimize the intermediate result. Compared with three state-of-the-art algorithms, simulation results suggest the effectiveness of the proposed algorithm.
- 出版日期2019-6
- 单位西安电子科技大学
