来源:IEEE TGRS 2019
原文: https:// ieeexplore.ieee.org/abs tract/document/8721080
论文题目:Deep Learning Ensemble for Hyperspectral Image Classification
Introduction
高光谱图像分类是遥感领域比较火热的研究方向之一。当前高光谱分类方法主要使用基于空间——光谱信息的半监督分类方法。但是在分类过程中也存在以下两点问题:
高光分类过程中可用的标记数据少,传统的卷积神经网络容易出现过拟合的的问题,导致泛化能力差。针对这两点问题作者提出了相应的解决方案如下:
1) To the best of our knowledge, the idea of combing deep learning and ensemble methods for HSI classification is proposed.
2) Two deep learning ensemble-based classification methods [i.e., CNN ensemble and deep residual network (ResNet) ensemble] are proposed for HSI classification. The proposed method utilizes deep learning model to extract the robust and discriminate features of HSI, RFS to formulate the diversity of the ensemble system, and majority voting to obtain the final classification results.
3) A tran