基于深度学习的耕地变化检测技术  

A Cultivated Land Change Detection Technology Based on Deep Learning

作　　者：高峰[1] 楚博策 帅通 王士成[1] 陈金勇[1] GAO Feng;CHU Boce;SHUAI Tong;WANG Shicheng;CHEN Jinyong(Key Laboratory of Aerospace Information Applications of CETC, Shijiazhuang 050081, China)

机构地区：[1]中国电子科技集团公司航天信息应用技术重点实验室

出　　处：《无线电工程》2019年第7期571-574,共4页Radio Engineering

基　　金：中国电子科技集团公司第五十四研究所发展基金资助项目(SXX18629X015)

摘　　要：随着人工智能技术的发展,以深度学习为主流的机器学习逐渐取代人工解译的方法,使遥感影像中地物资源的自动化判读成为现实。为解决传统人工判读引起的人力资源耗费高、解析精度差的问题,同时也为满足日益增长的遥感数据量的判读需求,基于语义分割的深度学习地物变化检测方法,实现耕地区域自动分割分类,通过对比时序影像差异得出变化区域范围,为自动化实现地物变化监测提供有效解决办法。以实际地区为例,采用deeplab语义分割网络的方法实现耕地资源的自动化提取与变化检测,实验证明该方法相比人工以及传统分类模型具有更好的检测精度。With the development of artificial intelligence technology, machine learning which takes deep learning as the mainstream has gradually replaced the manual interpretation method,making the automatic interpretation of terrestrial resources in remote sensing images become a reality. In order to solve the problems of high human resource consumption and poor resolution caused by traditional manual interpretation,and to meet the increasing demand for interpretation of remote sensing image data,the method of deep learning of landform change detection based on semantic segmentation is studied,the automatic segmentation and classification of cultivated land area are realized,the range of change area is obtained by comparing the differences of time series images,and the effective solution for automatically implementing landform change monitoring is provided. Taking Gaocheng District of Shijiazhuang as an example,the method of deeplab semantic segmentation network is used to realize automatic extraction and change detection of cultivated land resources. The experiment results show that this method has better detection accuracy compared with manual and traditional classification models.

关 键 词：遥感 耕地 语义分割 深度学习 变化检测 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]