多源遥感影像的分层特征信息提取与宁夏生态环境地物目标分类识别

For the requirement of information extraction and ground objects identification of desertification evolution, land utilization, forest vegetation, crop condition in Ningxia, according to the theory and method of remote sensing, statistics, information and image engineering, the project proposes to research the multi-scale remote sensing image fusion, classification, hierarchical feature representation and extraction. To deepen image analysis and understanding based on global and local features, establish multi-source remote sensing information extraction, object recognition method and model which have higher precision and more effective. Toward the lack of overall technical guidance model for a variety of different types of target, based on multi-source remote sensing data, we will enhance the high-level knowledge utilization degree, promote the potential targets search ability in large scale image and target confirmation and high-level matching calculation ability, then propose the technology framework for high resolution remote sensing information extraction and target recognition. Lay the foundation for multi-source information data processing in regional ecological environmental remote sensing dynamic monitoring , explore and solve the practical problems of remote sensing image geographic feature extraction and automatic target recognition or intelligent image classification, accurate extraction of information in ground object identification, provide technical support to extract geography and feature information quickly and accurately through remote sensing.

面向宁夏地区沙化演变、土地利用、森林植被和作物长势等信息提取和地物目标识别的需求，依据遥感、统计、信息、图像工程领域的理论和方法，研究多尺度遥感图像融合、分类，分层次特征表示、提取，深化基于全局和局部特征的图像分析与理解，建立具有更高精度和应用效果的多源遥感信息提取、目标识别方法与模型算法；在多、复杂目标提取和识别上，针对各种不同类型目标缺少总体技术模型指导，基于多源遥感数据、提升高层次知识利用程度、挖掘大尺度影像中潜在目标搜索能力以及目标的确认与高层匹配计算能力，并就此提出高分辨率遥感信息提取与目标识别技术框架。为区域生态环境遥感动态监测中多源数据的信息处理奠定基础，探究并有效解决遥感影像地理要素提取和地物目标识别中自动或智能影像分类、信息精确提取的实际问题，为使用遥感影像快速精确提取地理和地物信息提供技术支持。

利用多源遥感数据，融入智能化信息处理和数据处理的理论与算法，例如：分数阶微分差和高斯曲率滤波，CART自动决策树的遥感影像处理方法，图像/影像的边缘、变化检测算法，深度学习理论及其改进/优化等；通过特征、信息融合，边缘检测、图像/影像分割以及目标分类/识别等信息处理手段；围绕土地利用分类方法以及多源遥感影像精准分类/识别技术，绿地信息有效提取的多源遥感图像融合和分割算法改进，盐碱地、沙地和湿地等宁夏地物特征级、目标级的变化检测和识别等方面具体深入展开研究。面向宁夏地区沙化演变、土地利用、森林植被和作物长势等信息提取和地物目标识别的需求，研究多尺度遥感图像融合 、分类，分层次特征表示、提取，深化基于全局和局部特征的图像分析与理解，建立了具有更高精度和应用效果的多源遥感信息提取、目标识别方法与模型算法；为区域生态环境遥感动态监测中多源数据的信息处理奠定基础。项目研究基本达到预期目标，学术水平有待提升、研究深度和内容仍需升质、量产和系统化。