基于自顶向下视觉注意的遥感影像目标检测  

Object Detection for Remote Sensing Image Based on Top-Down Visual Attention

作　　者：朱巨莲[1] 霍宏[1] 方涛[1] 

机构地区：[1]上海交通大学图像处理与模式识别研究所,上海200240

出　　处：《计算机工程》2011年第20期172-174,共3页Computer Engineering

基　　金：国家"973"计划基金资助项目(2006CB701303);国家自然科学基金资助项目(41071256)

摘　　要：针对遥感影像数据量大、背景复杂、目标自身信息不足等问题,提出一种基于自顶向下视觉注意的遥感影像目标检测方法。根据目标外观线索优先选择与其特征相符的图像区域,以提高目标检测的效率,依据上下文线索将搜索集中在最可能出现目标的环境区域,以保证目标检测的可靠性。实验结果证明,该方法能提高目标检测的效率和正确率。A novel method based on Top-Down visual attention is proposed for object detection in remote sensing image.The method considers both object appearance cue and context cue in Top-Down attention.Object appearance cue is used to choose regions with the same features as object appearance to improve the efficiency.Context cue is used to focus the search on object context environment to confirm the correctness.Experimental results validate the good performance of the method in improving both the efficiency and the accuracy of object detection.

关 键 词：视觉注意 自顶向下 物体外观 上下文 目标检测 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]