The relation between corresponding trigonometric functions in two rotating coordinate systems is presented. The transformation formula for a vector in the two rotating spherical coordinate systems is obtained. The sca...The relation between corresponding trigonometric functions in two rotating coordinate systems is presented. The transformation formula for a vector in the two rotating spherical coordinate systems is obtained. The scattering fields for a spherical target irradiated by a plane electromagnetic wave in an arbitrary direction are derived. These fields in a particular case retrogress to those available in the literature. The obtained results have great potential in practical applications.展开更多
目的遥感图像上任意方向舰船目标的检测,是给出舰船在图像上的最小外切矩形边界框。基于双阶段深度网络的任意方向舰船检测算法速度较慢;基于单阶段深度网络的任意方向舰船检测算法速度较快,但由于舰船具有较大长宽比的形态特点,导致虚...目的遥感图像上任意方向舰船目标的检测,是给出舰船在图像上的最小外切矩形边界框。基于双阶段深度网络的任意方向舰船检测算法速度较慢;基于单阶段深度网络的任意方向舰船检测算法速度较快,但由于舰船具有较大长宽比的形态特点,导致虚警率较高。为了降低单阶段目标检测的虚警率,进一步提升检测速度,针对舰船目标的形态特点,提出了基于密集子区域切割的快速检测算法。方法沿长轴方向,将舰船整体密集切割为若干个包含在正方形标注框内的局部子区域,确保标注框内最佳的子区域面积有效占比,保证核心检测网络的泛化能力;以子区域为检测目标,训练核心网络,在训练过程对重叠子区域进行整合;基于子图分割将检测得到的子区域进行合并,进而估计方向角度等关键舰船目标参数。其中采用子区域合并后处理替代了非极大值抑制后处理,保证了检测速度。结果在HRSC2016(high resolution ship collections)实测数据集上,与最新的改进YOLOv3(you only look once)、RRCNN(rotated region convolutional neural network)、RRPN(rotation region proposal networks)、RDFPN-3(rotation dense feature pyramid network)和R-DFPN-4等5种算法进行了比较,相较于检测精度最高的R-DFPN-4对照算法,本文算法的m AP(mean average precision)(IOU(inter section over union)=0.5)值提高了1.9%,平均耗时降低了57.9%;相较于检测速度最快的改进YOLOv3对照算法,本文算法的m AP (IOU=0.5)值提高了3.6%,平均耗时降低了31.4%。结论本文所提出的任意方向舰船检测算法,结合了舰船目标的形态特点,在检测精度与检测速度均优于当前主流任意方向舰船检测算法,检测速度有明显提升。展开更多
基金Project supported by the National Natural Science Foundation of China （Grant No 60171010）, and the Education 0ffice of Shaaxi Province （Grant No 03JK070）.
文摘The relation between corresponding trigonometric functions in two rotating coordinate systems is presented. The transformation formula for a vector in the two rotating spherical coordinate systems is obtained. The scattering fields for a spherical target irradiated by a plane electromagnetic wave in an arbitrary direction are derived. These fields in a particular case retrogress to those available in the literature. The obtained results have great potential in practical applications.
文摘目的遥感图像上任意方向舰船目标的检测,是给出舰船在图像上的最小外切矩形边界框。基于双阶段深度网络的任意方向舰船检测算法速度较慢;基于单阶段深度网络的任意方向舰船检测算法速度较快,但由于舰船具有较大长宽比的形态特点,导致虚警率较高。为了降低单阶段目标检测的虚警率,进一步提升检测速度,针对舰船目标的形态特点,提出了基于密集子区域切割的快速检测算法。方法沿长轴方向,将舰船整体密集切割为若干个包含在正方形标注框内的局部子区域,确保标注框内最佳的子区域面积有效占比,保证核心检测网络的泛化能力;以子区域为检测目标,训练核心网络,在训练过程对重叠子区域进行整合;基于子图分割将检测得到的子区域进行合并,进而估计方向角度等关键舰船目标参数。其中采用子区域合并后处理替代了非极大值抑制后处理,保证了检测速度。结果在HRSC2016(high resolution ship collections)实测数据集上,与最新的改进YOLOv3(you only look once)、RRCNN(rotated region convolutional neural network)、RRPN(rotation region proposal networks)、RDFPN-3(rotation dense feature pyramid network)和R-DFPN-4等5种算法进行了比较,相较于检测精度最高的R-DFPN-4对照算法,本文算法的m AP(mean average precision)(IOU(inter section over union)=0.5)值提高了1.9%,平均耗时降低了57.9%;相较于检测速度最快的改进YOLOv3对照算法,本文算法的m AP (IOU=0.5)值提高了3.6%,平均耗时降低了31.4%。结论本文所提出的任意方向舰船检测算法,结合了舰船目标的形态特点,在检测精度与检测速度均优于当前主流任意方向舰船检测算法,检测速度有明显提升。
