摘要
红树林信息精细化智能提取是自然资源调查监测工作中亟待解决的关键技术。当前,红树林提取工作主要依靠人工目视解译,提取效率不高、精度较差。本文提出一种面向自然资源调查的红树林精细化智能提取方法,首先基于海量多源高分辨率遥感影像,结合多尺度优化分割技术,半自动化构建海量红树林样本集;然后结合语义分割网络,引入双注意力机制,优化训练策略,研发红树林自动化提取算法模型,经过后处理得到精细、完整的红树林提取结果。经过实验验证,本文方法提取可识别单颗红树林,有效区分易混淆的其他植被和水田要素,总精度高于90%,可有效应用于红树林专题信息的提取工作。
Intelligent extraction of refined mangrove information is a key technology to be developed urgently in the investigation and monitoring of natural resources.At present,mangrove extraction mainly relies on manual visual interpretation,with low extraction efficiency and poor accuracy.In this paper,a refined and intelligent mangrove extraction method for natural resources survey is proposed.Firstly,a massive mangrove sample set is constructed semi-automatically based on massive multi-source high-resolution remote sensing images combined with multi-scale optimization segmentation technology.And then combining the semantic network segmentation with introduction of dual attention mechanism,we optimize the training strategy,research and development of mangrove automatic extraction algorithm model,and obtain detained results after post-processing and complete mangrove extraction.After experimental verification,the method used for extraction can identify single mangrove forests,and easily distinguish other confused vegetation and paddy field,with the total accuracy higher than 90%,which can be effectively applied to the extraction of mangrove thematic information.
作者
赵彬如
张峰
邢喆
王力彦
焦红波
杨晓彤
樊妙
ZHAO Binru;ZHANG Feng;XING Zhe;WANG Liyan;JIAO Hongbo;YANG Xiaotong;FAN Miao(National Marine Data and Information Service,Tianjin 300012,China)
出处
《地理信息世界》
2022年第5期87-93,共7页
Geomatics World
基金
科技助力经济2020重点专项(SQ2020YFF0426316)、国家海洋信息中心青年科学基金(202001003)。
关键词
红树林
多尺度优化分割
语义分割网络
双注意力机制
mangrove
multi-scale optimization segmentation
semantic segmentation network
dual attention
