As a new machine learning approach, extreme learning machine (ELM) hasreceived wide attentions due to its good performances. However, when directlyapplied to the hyperspectral image (HSI) classification, the recognition rateis too low. This is because ELM does not use the spatial information which isvery important for HSI classification. In view of this, this paper proposes anew framework for spectral-spatial classification of HSI by combining ELM withloopy belief propagation (LBP). The original ELM is linear, and the nonlinearELMs (or Kernel ELMs) are the improvement of linear ELM (LELM). However, basedon lots of experiments and analysis, we found out that the LELM is a betterchoice than nonlinear ELM for spectral-spatial classification of HSI.Furthermore, we exploit the marginal probability distribution that uses thewhole information in the HSI and learn such distribution using the LBP. Theproposed method not only maintain the fast speed of ELM, but also greatlyimproves the accuracy of classification. The experimental results in thewell-known HSI data sets, Indian Pines and Pavia University, demonstrate thegood performances of the proposed method.
展开▼
