本文分析基于卷积神经网络的2种目标检测方法,即两步目标检测法、单步目标检测法,并对多尺度卷积神经网络检测、遥感图像少样本学习的目标检测以及船舶遥感图像的方向目标检测3种类型的遥感图像目标检测模型进行分析。仿真实验表明,本文提出的目标检测方法能够提高船舶遥感图像检测精确度和时效性。
This paper analyzes two target detection methods based on convolutional neural network, namely two-step target detection method and single-step target detection method, and also for multi-scale convolutional neural network detection, remote sensing image few-sample learning target detection and ship remote sensing image detection. Three types of remote sensing image target detection models for direction target detection are analyzed. After simulation experiments are verified, the target detection method proposed in this paper can improve the accuracy and timeliness of ship remote sensing image detection.
2022,44(14): 151-154 收稿日期:2022-01-17
DOI:10.3404/j.issn.1672-7649.2022.14.032
分类号:U665
作者简介:胡俊梅(1988 - ),女,硕士,讲师,主要从事神经网络与图像处理的研究
参考文献:
[1] 曾禹龙. 光学遥感图像舰船目标视觉显著性检测方法[J]. 电子设计工程, 2022, 30(8): 52–56
[2] 徐芳, 刘晶红, 王宣. 光学遥感图像海面船舶目标检测技术进展[J]. 光学精密工程, 2021, 29(4): 916–931
[3] 陈跃. 基于光学遥感图像的船舶目标检测技术研究[J]. 舰船科学技术, 2016, 38(22): 121–123
[4] 王伟. 基于遥感图像的船舶目标检测方法综述[J]. 电讯技术, 2020, 60(9): 1126–1132
[5] 方梦梁, 黄刚. 一种光学遥感图像船舶目标检测技术[J]. 计算机技术与发展, 2019, 29(8): 136–141
[6] 齐亮, 陈牮华, 董梁. 基于SRM分割和分层线段特征的船舶目标检测方法[J]. 江苏科技大学学报:自然科学版, 2020, 34(3): 34–40