光学学报, 2020, 40 (23): 2312006, 网络出版: 2020-11-23   

基于关键点的快速红外目标检测方法 下载: 1043次

Method for Fast Detection of Infrared Targets Based on Key Points
苗壮 1,2张湧 1,*陈瑞敏 1,2李伟华 1,2
作者单位
1 中国科学院上海技术物理研究所红外探测与成像技术重点实验室, 上海 200083
2 中国科学院大学电子电气与通信工程学院, 北京 100049
摘要
针对红外探测系统对目标检测的实时性要求,提出了一种基于关键点的快速红外目标检测方法。以目标中心作为目标检测关键点,首先设计了轻量化的特征提取网络,之后结合红外目标较小的特点,利用不同层次特征的空间信息和语义信息设计了相应的特征融合网络,并最终实现目标类别、位置和尺寸信息的预测。在自建空中红外目标数据集上对模型进行了对比测试,与YOLOv3等经典检测模型相比,检测速度大幅提高,检测精度仅略有下降;与同类型快速检测模型Tiny-YOLOv3相比,在模型尺寸压缩至Tiny-YOLOv3尺寸的23.39%的情况下,检测精度提高了8.9%,在中央处理器(CPU)上运行的检测速度亦提高了13.9 ms/frame,检测性能显著提升,验证了方法的有效性。
Abstract
Aim

ing at the real-time request of the infrared detection system for target detection, we propose a method for fast detection of infrared targets based on key points. Taking the target center as the key point of target detection, we first design a lightweight feature extraction network. Then, we design a corresponding feature fusion network using the spatial and semantic information of features at different levels combined with the characteristics of small infrared targets. Finally, the prediction of target category, location and size is realized. The model is comparatively tested on the self-built aerial infrared target dataset. Compared with the classic detection models such as YOLOv3, the detection speed is greatly improved and the detection accuracy is only slightly reduced. Compared with the same type of fast detection model, Tiny-YOLOv3, the detection accuracy increases by 8.9% and the detection speed running on the central processing unit (CPU)increases by 13.9 ms/frame under the condition that the model size is compressed to 23.39% of Tiny-YOLOv3's size. The detection performance is significantly improved and the effectiveness of the method is confirmed.

苗壮, 张湧, 陈瑞敏, 李伟华. 基于关键点的快速红外目标检测方法[J]. 光学学报, 2020, 40(23): 2312006. Zhuang Miao, Yong Zhang, Ruimin Chen, Weihua Li. Method for Fast Detection of Infrared Targets Based on Key Points[J]. Acta Optica Sinica, 2020, 40(23): 2312006.

本文已被 6 篇论文引用
被引统计数据来源于中国光学期刊网
引用该论文: TXT   |   EndNote

相关论文

加载中...

关于本站 Cookie 的使用提示

中国光学期刊网使用基于 cookie 的技术来更好地为您提供各项服务,点击此处了解我们的隐私策略。 如您需继续使用本网站,请您授权我们使用本地 cookie 来保存部分信息。
全站搜索
您最值得信赖的光电行业旗舰网络服务平台!