摘要
水体的自动提取对于洪水监测、水资源管理等方面有着重要意义。本文提出了SegNet_CRF语义分割方法,可从遥感影像上自动提取水体。首先,在SegNet编码器和解码器之间植入空洞卷积特征提取块,融合不同尺度的特征,然后在分类后处理中引入条件随机场,对提取结果进行精细化处理,最后与FCN、经典SegNet网络水体提取结果对比,结果表明,SegNet_CRF网络结构在Recall、Precision以及F1-score指标上都有所提高,水体提取结果更加准确完整,抑制噪声能力更强。SegNet_CRF网络可有效地实现水体提取任务。
Automatic extraction of water body is of great significance for flood monitoring and water resources management.This paper proposed SegNet_CRF semantic segmentation method to automatically extract water from remote sensing images.Firstly,the hole convolution feature extraction block is implanted between the SegNet encoder and decoder to integrate the features of different scales,and then the conditional random field is introduced into the classification post-processing to refine the extraction results.Finally,the results of water extraction are compared with those of FCN and classical SegNet networks.The results show that the SegNet_CRF network structure has improved in the indexes of recall,precision,and F1-score,the water extraction results are more accurate and complete,and the noise suppression ability is stronger.The SegNet_CRF can effectively realize the task of water extraction.
作者
林娜
王玉莹
郭江
潘鹏
李莉
LIN Na;WANG Yuying;GUO Jiang;PAN Peng;LI Li(School of Civil Engineering,Chongqing Jiaotong University,Chongqing 400074,China;School of Surveying,Mapping and Geographic Information,Chongqing Vocational Institute of Engineering,Chongqing 402260,China)
出处
《测绘与空间地理信息》
2023年第3期12-15,18,共5页
Geomatics & Spatial Information Technology
基金
重庆市教委科学技术研究项目(KJQN201800747、KJQN202103410)资助。