基于光谱上下文特征的多光谱舰船ROI鉴别方法  

Multispectral ship ROI identification method based on spectral context feature

作　　者：唐天翼 朱长仁[2] TANG Tianyi;ZHU Changren(Beijing Aerospace Control Center, Beijing 100089, China;Key Laboratory of ATR, School of Electronic Science and Engineering, National University of Defense Teelmology, Changsha 410073, China)

机构地区：[1]北京航天飞行控制中心,北京100089 [2]国防科技大学电子科学与工程学院ATR重点实验室,湖南长沙410073

出　　处：《现代电子技术》2018年第12期133-138,144,共7页Modern Electronics Technique

基　　金：国家自然科学基金(61301235);博士后基金(2014M550872);国防预研基金~~

摘　　要：多光谱遥感图像海洋背景下往往存在云、海浪、海岛、海岸等多种干扰情况,导致舰船检测很具有挑战性。因此提出一种基于光谱上下文特征的舰船感兴趣区域(ROI)鉴别算法。在舰船ROI切片基础上,按照主轴轴向和轴两侧子块划分,提出一种光谱"凹凸性"的特征描述方法和一种方向对称性的特征提取方法,构建并实现了组合光谱"凹凸性"和梯度方向对称性的光谱上下文特征描述方法,利用SVM分类实现舰船ROI鉴别。实验证明,在不同的图像分辨率情况下,引入光谱上下文特征的方法能够有效剔除大量云、海浪、海岛、海岸等虚警,具有较好的鲁棒性、有效性、适用性。Interferences such as clouds,waves,islands and coasts often exist in multispectral remote sensing images under the sea background,which makes ship detection more challenging. A ship region of interest（ROI）identification algorithm based on spectral context feature is proposed in this paper. On the basis of ship ROI slices which are divided according to the direction of the principle axis and the subblocks on both sides of the axis,a feature description method based on the concavity and convexity of the spectrum and a feature extraction method based on direction symmetry are proposed. A spectral context feature description method of combining the concavity and convexity of the spectrum and the symmetry of the gradient direction is constructed and realized. The SVM classification is adopted to realize ship ROI identification. The experimental results show that the spectral context feature description method in the situation of different image resolution ratios can effectively eliminate many false alarms such as clouds,waves,islands and coasts,and has good robustness,effectiveness,and applicability.

关 键 词：多光谱遥感图像 舰船ROI鉴别 光谱特征 梯度方向 SVM分类 图像分辨率 

分 类 号：TN206-34[电子电信—物理电子学] TP391.4[自动化与计算机技术—计算机应用技术]