面向多源数据地物提取的遥感知识感知与多尺度特征融合网络  

Multi-source Data Ground Object Extraction Based on Knowledge-Aware and Multi-scale Feature Fusion Network

作　　者：龚健雅[1] 张展 贾浩巍 周桓 赵元昕 熊汉江[2] GONG Jianya;ZHANG Zhan;JIA Haowei;ZHOU Huan;ZHAO Yuanxin;XIONG Hanjiang(School of Remote Sensing and Information Engineering,Wuhan University,Wuhan 430079,China;State Key Laboratory of Information Engineering in Surveying,Mapping and Remote Sensing,Wuhan 430079,China;Department of Land-Surveying and Geo-Informatics,Hong Kong Polytechnic University,Hong Kong 999077,China)

机构地区：[1]武汉大学遥感信息工程学院,湖北武汉430079 [2]武汉大学测绘遥感信息工程国家重点实验室,湖北武汉430079 [3]香港理工大学土地测量与地理资讯学系,中国香港999077

出　　处：《武汉大学学报（信息科学版）》2022年第10期1546-1554,共9页Geomatics and Information Science of Wuhan University

基　　金：国家自然科学基金(42090011,41971402)。

摘　　要：遥感地物自动提取是遥感智能解译中的关键问题,对空间信息的理解和知识发现具有重要意义。近年来,使用全卷积神经网络(fully convolutional networks, FCN)从高分影像和三维激光雷达(light detection and ranging, LiDAR)数据中提取地物信息因取得了较好效果而受到广泛关注。现有FCN网络在地物提取精度和效率等方面仍存在不足,由此提出一种基于多源数据的遥感知识感知与多尺度特征融合网络(knowledge-aware and multi-scale feature fusion network, KMFNet)。在网络编码器端融入遥感知识感知模块(knowledge-aware module, KAM),高效挖掘多源遥感数据中的遥感知识信息;在网络编码器和解码器之间添加了串并联混合空洞卷积模块(series-parallel hybrid convolution module, SPHCM),提高网络对地物多尺度特征的学习能力;在解码器端使用了渐进式多层特征融合策略,细化最终的地物分类结果。基于公开的ISPRS语义分割标准数据集,在LuoJiaNET遥感智能解译开源深度学习框架上将KMFNet与当前主流方法进行了对比。实验结果表明,所提方法提取出的地物更为完整,细节更加精确。Objectives: In recent years, the automatic ground object extraction from remote sensing images has been dramatically advanced by the fully convolutional networks(FCNs). It is an effective method to fuse high-resolution images and light detection and ranging(LiDAR) data in FCNs to improve the extraction accuracy and the robustness. However, the existing FCN-based fusion networks still face challenges in efficiency and accuracy.Methods: We propose a knowledge-aware and multi-scale fusion network(KMFNet) for robust and accurate ground object extraction. The proposed network incorporates a knowledgeaware module in the network encoder for better exploiting remote sensing knowledge between pixels. A series-parallel hybrid convolution module is developed to enhance multi-scale representative features from ground objects. Moreover, the network decoder uses a gradual bilinear interpolation strategy to obtain finegrained extraction results.Results: We evaluate KMFNet in the LuoJiaNET with four current mainstream ground object extraction methods(GRRNet, V-FuseNet, DLR and Res-U-Net) on ISPRS 2D semantic segmentation dataset. The comparative evaluation results show that KMFNet can obtain the best overall accuracy. Compared with the other four methods, KMFNet achieves a better effect by improving the overall accuracy by 3.20% and 2.82% on average in ISPRS-Vaihingen dataset and ISPRS-Potsdam dataset, respectively.Conclusions: KMFNet achieves the best extraction results by capturing the intrinsic pixel relationships and strengths the multi-scale representative and detailed features of ground objects. It shows great potential in high-precision remote sensing mapping applications.

关 键 词：高分影像 三维激光雷达 地物提取 全卷积神经网络 遥感知识 多尺度特征 

分 类 号：P237[天文地球—摄影测量与遥感]