基于遥感影像的郊外建筑物自动检测，在违章建筑监测、土地利用变更监测、反恐维稳、军事侦察等关键领域都有着迫切需求，是国内外遥感信息处理技术的热点研究方向之一。本项目针对郊外复杂大视场条件下，遥感建筑物检测中的若干关键科学问题，利用交叉学科间相互借鉴的优势，引入视觉认知机制及其中重要神经功能的相关理论研究成果，拟建立一套高效的遥感图像郊外建筑物检测模型架构。并在该顶层架构指导下开展其中的关键技术研究，包括：建立基于建筑物敏感结构特征的视觉注意计算模型，实现疑似区筛选；通过深度视觉认知学习方法建模及相应加速策略的研究，实现疑似区自主深度特征快速学习及提取；通过对所提取疑似区深度特征的多层次视觉上下文信息融合建模及双层分类网构建，实现建筑物鉴别确认。通过本项目研究，为研究郊外复杂大视场条件下高时效性遥感建筑物检测问题开辟新的切入点，提供理论方法及关键技术支撑。

Automatic building detection in suburb remote sensing images is one of the core parts of high-resolution earth observation technologies. There are many urgently demands in illegal construction monitoring, land use change monitoring, counter-terrorism, military reconnaissance and other key areas. It becomes the focus direction of remote sensing information processing technology. Under the complexity of large field in remote sensing images, this project focuses on the key scientific problems about remote sensing building detection. By introducing visual cognitive science mechanism and its several important nerve-functions, we plan to establish one unified and efficient model for remote sensing building detection. And then, under the guidance of this top-level architecture, we carry out research on the key technologies, including: the establishment computational model of visual attention with building sensitive structure, to achieve the goal of screening suspected area; through the depth of visual modeling cognitive learning methods and its acceleration strategy, to achieve autonomous feature extraction; through the multi-level integration of contextual information encoded visual modeling, to achieve the goal to identify the suspected area. By this research project, we want to establish a clear hierarchical architecture for automatic building detection in suburb remote sensing images. In particular, a new entry points is given to study this research field, and to provide some theoretical methods and key technical support.