多特征融合与集成学习的城市高分辨率遥感影像变化检测

项目名称： 多特征融合与集成学习的城市高分辨率遥感影像变化检测

项目编号： No.41471354

项目类型： 面上项目

立项/批准年度： 2015

项目学科： 天文学、地球科学

项目作者： 孙开敏

作者单位： 武汉大学

项目金额： 85万元

中文摘要： 随着城市化进程的加快和可持续性发展需要，城市规划、环境监测、灾害动态监测等急需能够利用高分辨率遥感影像快速获得城市变化信息的方法。由于城市地区地物种类复杂、成像特性各异，尤其建筑物在高分辨率遥感影像中普遍存在投影差和阴影，这大大增加了变化检测难度。传统变化检测方法一般用于中低分辨率遥感影像，主要利用影像光谱和纹理特征，检测模型简单、单一，无法顾及城市地物在高分辨率遥感影像上的成像特性差异，难以保证变化检测的可靠性和精度。针对上述问题，本项目提出多特征融合与集成学习的变化检测方法，其核心思想是针对城市不同地物类型设计多种变化检测器，然后利用集成学习方法进行集成检测，获得优于任一变化检测器独立工作的结果。主要内容包括多特征融合与集成学习的城市地物多概率分类（判定）、二三维同源/异源高分数据自动配准、二三维辅助数据约束下的自适应多尺度影像分割、多特征融合与集成学习的多变化检测器集成检测等方法研究

中文关键词： 变化检测；多尺度影像分割；多特征融合；集成学习；影像分类

英文摘要： With the demands of accelerated urbanization and sustainable development, the fields including urban planing, environmental monitoring and disaster dynamic monitoring are in the urgent need of change detection method which could obtain urban changes rapidly by using high-resolution remote-sensing images. Due to the complexity of urban surface features and difference of imaging properties, there are many problems in urban change detection, especially when height displacements and shadows caused by buildings appear. Traditional change detection methods generally apply to moderate or low resolution remote-sensing images, detecting changes with spectral characteristic and textural features. Also, the traditional method could not ensure accuracy and reliability because of the simple and single detection model without consideration of the surface features' characteristic and imaging difference. In order to solve the problems above, this project proposes a Multi-feature Fusion and Ensemble Learning based change detection method, with the core idea that obtain the results superior to any individual change detector through designing multi-change-detector corresponding to different surface features' types firstly and then detecting with the assistance of ensemble learning. The main research contents include urban probability classification based on multi-feature fusion and ensemble learning, 2D and 3D homologous/heterogeneous data registration, adaptive multi-scale segmentation with 2D and 3D data, and multi-feature fusion and ensemble learning based ensemble change detection with multi-change-detector.

英文关键词： Change Detection;Multi-scale Image Segmentation;Multi-feature Fusion;Ensemble Learning;Image Classification