面向生态红线人类干扰识别的多分类对象级变化检测方法  

A Multi-Class Object-Level Change Detection Method for Identifying Human Disturbance in Ecological Red Line Areas

作　　者：官晓坤 张新胜 昝露洋 陈盼 吴朝明 向云帆 蔡明勇 Xiaokun GUAN;Xinsheng ZHANG;Luyang ZAN;Pan CHEN;Zhaoming WU;Yunfan XIANG;Mingyong CAI(Airborne Remote Sensing Center,Aerospace Information Research Institute,Chinese Academy of Sciences,Beijing 100094,China;School of Electronic,Electrical and Communication Engineering,University of Chinese Academy of Sciences,Beijing 100049,China;Satellite Application Center for Ecology and Environment,Ministry of Ecology and Environment,Beijing 100049,China;Center for Geo-Spatial Information,Shenzhen Institutes of Advanced Technology,Chinese Academy of Sciences,Shenzhen 518055,China)

机构地区：[1]中国科学院空天信息创新研究院航空遥感中心,北京100094 [2]中国科学院大学电子电气与通信工程学院,北京100049 [3]生态环境部卫星环境应用中心,北京100094 [4]中国科学院深圳先进技术研究院数字所空间信息研究中心,广东深圳518055

出　　处：《遥感技术与应用》2024年第5期1115-1127,共13页Remote Sensing Technology and Application

基　　金：国家重点研发计划项目(2021YFB390110302)。

摘　　要：生态红线界定了禁止工业化和城市化发展的区域,对生态保护具有重要意义。为保障生态红线不被破坏,需要对人类干扰进行识别。传统的方法难以精确识别由于人类干扰活动而产生的地物变化,也难以区分变化前后的地物类别。基于深度学习的像素级多分类变化检测方法存在细小误检多,样本获取难度大的问题。基于此,提出多分类的对象级变化检测网络,开展生态红线内人类干扰的精确识别研究。研究方法分为对象级二分类变化检测和场景分类两部分。对象级二分类变化检测网络以YOLO v5为基本框架,分别提取前后时相的特征,随后将特征融合,最终输出目标框形式的变化区域;场景分类网络以MobileNet v2为基础,对变化区域所对应的前后时相影像分别进行精准分类。以漓江生态保护区的两期高分影像为数据源,识别27种人类干扰活动。实验证明,多分类对象级变化检测网络的变化区域提取精度APIoU=.50达到68.8%,APIoU=.50:.05:.95达到57.2%,人类干扰活动识别的类别top-1准确率达到91.81%,top-5准确率达到99.83%。结果表明:采用对象级变化检测和场景分类两步走的方式,提升了变化区域的提取效果,解决了多分类变化检测样本不足的问题,可为生态红线人类干扰活动识别提供有效支撑。The delineation of ecological red lines,which define areas where industrialization and urbanization are prohibited,holds great significance for environmental conservation.To ensure the protection of ecological red line areas,it is essential to identify human disturbances accurately.Traditional methods face challenges in precisely detecting land cover changes caused by human interference and distinguishing the classes of objects before and after the changes.Segmatic change detection methods based on deep learning suffer from issues such as excessive false positives and difficulties in obtaining training samples.To address these challenges,this paper proposes a multi-class object-level change detection method for precise identification of human disturbances within ecological red line areas.The proposed method consists of two parts:object-level binary change detection and scene classification.The object-level binary change detection network utilizes YOLOv5 as the underlying framework to extract features from the pre-and post-change images,fuse the features,and output the changed regions in the form of bounding boxes.The scene classification network,based on MobileNet v2,accurately classifies the pre-change and post-change images corresponding to the changed regions.High-resolution satellite images from the Li River Ecological Protection Zone are used as the dataset to identify 27 types of human disturbance activities.Experimental results demonstrate that the object-level change detection network achieves APIoU=.50 of 68.8%and APIoU=.50:.05:.95 of 57.2%for change region extraction.The top-1 accuracy for human disturbance activity recognition reaches 91.81%,and the top-5 accuracy reaches 99.83%.The results indicate that the two-step approach of object-level change detection and scene classification improves the effectiveness of change region extraction and overcomes the limitation of insufficient training samples for multi-class change detection.This approach provides effective support for the identification of human

关 键 词：生态红线 人类干扰活动识别 深度学习 变化检测 多分类 对象级 场景分类 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]