摘要
深度卷积网络作为一种高效的特征表示学习算法,被广泛的应用于图像分类问题中.由于遥感图像存在目标尺度与方向变化大、类内场景差异形大等问题,单一的深度网络通常不能获得准确的分类结果.为此,本文提出一种随机多选择残差网络集成的遥感图像分类算法,该算法通过多选择学习策略,集成多个残差网络共同完成分类任务,算法设置有效的集成学习目标函数,并通过随机梯度下降算法最小化多个子网络对每个样本的最优分类误差,促使各个网络之间的差异性,能够自适应于特定类别的分类任务,进而形成有效的分类,同时其泛化性通常显著优于单个学习器.在两个公开的遥感数据集上验证了本文算法的有效性,多个残差网络能够对不同类别的遥感影像形成最优分类,有效提升了分类的准确性.
As a kind of efficient feature representation learning algorithm,deep convolutional networks are widely used in image classification problems. Due to the large variation of target scale and direction,and the large difference of intra-class scenes in remote sensing images,a single deep network usually cannot obtain accurate classification results. To cope with this issue,this paper proposes a remote sensing image classification algorithm based on random multi-selective ensemble of residual sub-networks. The algorithm integrates multiple residual sub-networks to accomplish the classification task by multi-selection learning strategy. The algorithm sets an effective ensemble learning objective function and minimizes the optimal classification error of each sub-network for each sample by stochastic descent algorithm,which promotes the diversity between the sub-networks. Thus each sub-network can adapt to the classification task of a specific category,which can advance effective classification,and the generalization of our ensemble model is usually better than a single network. The effectiveness of the proposed algorithm is verified on two publicly available remote sensing datasets.Multiple residual sub-networks can form optimal classifications for different types of remote sensing images,which effectively improves the accuracy of classification.
作者
周强
徐宏伟
陈逸
孙玉宝
ZHOU Qiang;XU Hong-wei;CHEN Yi;SUN Yu-bao(Nanjing University of Information Science and Technology,School of Automation,Jiangsu Key Laboratory of Big Data Analysis Technology,Jiangsu Collaborative Innovation Center on Atmospheric Environment and Equipment Technology,Nanjing 210044,China)
出处
《小型微型计算机系统》
CSCD
北大核心
2019年第9期1946-1950,共5页
Journal of Chinese Computer Systems
基金
基金项目:国家自然科学基金项目(61672292)资助
江苏省高校重大项目(18KJA52007)资助
江苏省"六大人才高峰"项目(DZXX-037)资助
关键词
遥感图像分类
网络集成
随机多选择学习
残差网络
remote sensing image classification
network ensemble
random multi-choice learning
residual network
