摘要
遥感图像中须要识别的目标往往与已知训练的目标数据在外观、成像质量等特性方面不一样,从而导致目标正确识别效果差的问题.针对这一问题,提出一种基于迁移学习的方法用于遥感目标识别.该方法首先提取目标的Hu矩参数作为其特征向量,然后采用迁移学习方法来寻找特征空间中目标数据与不同分布的训练数据之间共同的知识实现迁移.试验结果表明:提出的方法能够有效地用于对遥感目标的识别,对比其他传统的方法,识别效果有明显的提高.
As some attributes of the test data,such as shape,imaging quality,are different from those of the training data in the remote sensing(RS)images,the different distributions of data cause the low reliability of the target recognition.Aimed at the problem,a new target recognition method based on transfer learning for remote sensing image was proposed.Hu moments were firstly extracted as the feature vectors of the targets,and then the transfer learning was used to find the common knowledge transferred in the feature spaces between the target data and the training data.The experimental results show that the proposed method can obtain approving effect in the RS target recognition,and it has greatly improved the performance compared with the other classical methods.
出处
《华中科技大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2013年第S1期140-143,共4页
Journal of Huazhong University of Science and Technology(Natural Science Edition)
基金
国家自然科学基金资助项目(61102155)
湖北省教育厅中青年自然科学研究基金资助项目(Q20111205)
第一届高分辨率对地观测学术交流基金资助项目(GFZX04060103)
关键词
目标识别
图像处理
机器学习
遥感图像
识别算法
target recognition
image processing
machine learning
remote sensing image
recognition algorithm
