一种多尺度特征关联网络的遥感影像场景分类方法  

A Remote Sensing Image Scene Classification Method Based on Multi-scale Feature Correlation Network

作　　者：段烨 陈国坤 李佳田[1] 金维 胡浩 DUAN Ye;CHEN Guokun;LI Jiatian;JIN Wei;HU Hao(School of Land Resources Engineering,Kunming University of Science and Technology,Kunming 650093,China)

机构地区：[1]昆明理工大学国土资源工程学院,昆明650093

出　　处：《遥感信息》2023年第5期157-164,共8页Remote Sensing Information

基　　金：云南省教育厅重点项目(202101AU070161);中国科学院专项基金项目(XDA26050301-01)。

摘　　要：针对遥感影像场景分类中提取特征信息不准确以及融合冗余信息的问题,提出一种多尺度特征关联网络的遥感影像场景分类方法。首先,利用ResNet18提取多尺度影像特征,并在特征金字塔结构中引入空洞卷积和多头自注意力模块关注各层有效信息,以增强语义信息和抑制周围噪声;其次,使用多特征关联模块来增强关键特征信息,对各层使用全局平均池化并加和,得到特征向量;最后,采用全连接层进行分类。该方法在NWPU数据集上的总体分类精度为90.51%,相比VGG_VD16-MSCP、VGG_VD16-SAFF以及DTDCNN等网络分别提升1.58%、2.65%和6.39%。结果表明,文章所用的多尺度融合方式能够更好地提取特征地物并抑制周围背景,从而取得更高的遥感影像场景分类精度。Aiming at the problems of incorrect extraction of feature information and merging redundant information in remote sensing image scene classification,we propose a multi-scale feature correlation network on remote sensing image scene classification in this paper.Firstly,ResNet18 is used to extract multi-scale image features,and the atrous convolution and multi-head self-attention modules are introduced in the feature pyramid structure,which focuse on the effective information of each layer and enhances semantic information while suppressing surrounding noise.Then,a multi-feature correlation module is used to enhance the critical feature information,and global average pooling and summing are applied to each layer to obtain the feature vector.Finally,a fully connected layer is used for classification.The proposed method achieves an overall classification accuracy of 90.51%on the NWPU dataset,demonstrating superior performance compared to existing networks such as VGG_VD16-MSCP,VGG_VD16-SAFF,and DTDCNN,with the improvement of 1.58%,2.65%,and 6.39%,respectively.The results show that the multi-scale fusion method used in this paper can effectively extract feature information of land cover and suppress surrounding background,thereby improving the classification accuracy of remote sensing image scenes.

关 键 词：高分辨率遥感影像 场景分类 多头自注意力 多尺度融合 特征关联 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]