摘要
传统的高分辨率遥感图像的语义解析大多面向中低层次的图像语义解释,不能很好地满足更高层次的图像语义需求。针对这一问题,提出DCNN特征融合的方式进行特征处理以提高特征描述力,再通过寻求优秀核函数结合的方式构建分类器。通过实验证明,特征融合结合构建的多核多类相关向量机的方式能够提高分类效果。该方法应用于构建的实际场景LSV数据集,能够准确实现遥感场景分类。
The traditional semantic analysis of high-resolution remote sensing image is mostly oriented to the middle and low-level image semantic interpretation,which cannot meet the needs of higher-level image semantics.To solve this problem,this paper proposes a DCNN-feature fusion method for feature processing to improve the ability of feature description.Then the classifier was constructed by seeking the combination of excellent kernel functions.The experiments show that the multi-kernel multi-class relevance vector machine constructed by feature fusion can improve the classification effect.The proposed method is applied to the actual scene LSV data set,which can accurately realize the remote sensing scene classification.
作者
王倩
宁芊
卢湖川
杨晓敏
陈炳才
雷印杰
Wang Qian;Ning Qian;Lu Huchuan;Yang Xiaomin;Chen Bingcai;Lei Yinjie(College of Electronics and Information Engineering,Sichuan University,Chengdu,610065,Sichuan,China;School of Physics and Electronics Engineering,Xinjiang Normal University,Urumqi 830054,Xinjiang,China;School of Information and Communication Engineering,Dalian University of Technology,Dalian 116024,Liaoning,China)
出处
《计算机应用与软件》
北大核心
2022年第9期213-218,共6页
Computer Applications and Software
基金
新疆维吾尔自治区区域协同创新专项(2019E0214)
国家自然科学基金-新疆联合基金项目(U1903215)。
关键词
深度特征
特征融合
多核多类相关向量机(MMRVM)
遥感场景分类
Depth feature
Feature fusion
Multi-kernel multi-class relevance vector machine(MMRVM)
Remote sensing classification
