基于视觉显著性的无监督海面舰船检测与识别
发布时间：2019-05-26 18:59
【摘要】：在航天航空光学遥感舰船目标检测中,受大气、光照、云雾和海岛等海面不确定条件的影响,传统的舰船检测方法存在检测效率低和可靠性差等问题,因此,本文提出一种无监督海面舰船目标自动检测方法。该方法以视觉显著性为依据,结合多显著性检测模型快速搜索海面目标,生成显著图后对其进行粗分割,对提取的目标切片做标记并进行精细分割,利用改进的Hough变换旋转目标主轴以保证目标对Y轴的对称性;对可能检测到的厚重云层和岛屿等伪目标使用梯度方向特征进行鉴别,通过判定目标在360°范围内8个区间的梯度幅度统计值,确认舰船目标及去除伪目标。实验结果表明,该舰船检测方法能够成功提取海面上大小不同,位置随机分布的舰船目标,准确获取舰船目标的数量和位置信息,在大量真实光学遥感图像上的测试结果显示,本文方法检测准确率高于93%,通过目标鉴别处理,剔除伪目标后,虚警率可低于4%,鲁棒性较强。
[Abstract]:In aerospace optical remote sensing ship target detection, due to the influence of uncertain sea surface conditions such as atmosphere, light, clouds and islands, the traditional ship detection methods have some problems, such as low detection efficiency and poor reliability. In this paper, an automatic detection method for unsupervised sea surface ship targets is proposed. This method is based on visual significance, combined with multi-significance detection model to quickly search the sea surface target, generates a significant map and carries on the rough segmentation to it, marks the extracted target slice and carries on the fine segmentation. The improved Hough transform is used to rotate the target spindle to ensure the symmetry of the target to the Y axis. The gradient direction characteristics of false targets such as thick clouds and islands are identified. By determining the gradient amplitude statistics of the target in the range of 360 掳, the ship target is confirmed and the pseudo target is removed. The experimental results show that the ship detection method can successfully extract the ship targets with different sizes and randomly distributed positions, and accurately obtain the number and position information of the ship targets. The test results on a large number of real optical remote sensing images show that the ship detection method can successfully extract the ship targets with different sizes and randomly distributed positions, and accurately obtain the number and position information of the ship targets. The detection accuracy of this method is higher than 93%. After eliminating false targets, the false alarm rate can be less than 4%, and the robustness is strong.
【作者单位】： 中国科学院长春光学精密机械与物理研究所航空光学成像与测量重点实验室;中国科学院大学;
【基金】：国家自然科学基金资助项目(No.60902067) 吉林省重大科技攻关项目(No.11ZDGG001)
【分类号】：TP751
本文编号：2485556
[Abstract]:In aerospace optical remote sensing ship target detection, due to the influence of uncertain sea surface conditions such as atmosphere, light, clouds and islands, the traditional ship detection methods have some problems, such as low detection efficiency and poor reliability. In this paper, an automatic detection method for unsupervised sea surface ship targets is proposed. This method is based on visual significance, combined with multi-significance detection model to quickly search the sea surface target, generates a significant map and carries on the rough segmentation to it, marks the extracted target slice and carries on the fine segmentation. The improved Hough transform is used to rotate the target spindle to ensure the symmetry of the target to the Y axis. The gradient direction characteristics of false targets such as thick clouds and islands are identified. By determining the gradient amplitude statistics of the target in the range of 360 掳, the ship target is confirmed and the pseudo target is removed. The experimental results show that the ship detection method can successfully extract the ship targets with different sizes and randomly distributed positions, and accurately obtain the number and position information of the ship targets. The test results on a large number of real optical remote sensing images show that the ship detection method can successfully extract the ship targets with different sizes and randomly distributed positions, and accurately obtain the number and position information of the ship targets. The detection accuracy of this method is higher than 93%. After eliminating false targets, the false alarm rate can be less than 4%, and the robustness is strong.
【作者单位】： 中国科学院长春光学精密机械与物理研究所航空光学成像与测量重点实验室;中国科学院大学;
【基金】：国家自然科学基金资助项目(No.60902067) 吉林省重大科技攻关项目(No.11ZDGG001)
【分类号】：TP751
【相似文献】
相关期刊论文 前1条
1 林大钧;平面场的梯度图解方法与应用[J];华东理工大学学报;1998年01期
，本文编号：2485556
本文链接：https://www.wllwen.com/guanlilunwen/gongchengguanli/2485556.html
