天空背景红外图像序列弱小目标检测算法研究  

Research on dim and small target detection algorithm in sky background infrared image sequence

作　　者：黄夏阳 张涛[2,3] 朱秋煜 崔文楠[2,3] 李洁[2,3] Huang Xiayang;Zhang Tao;Zhu Qiuyu;Cui Wennan;Li Jie(School of Communication and Information Engineering,Shanghai University,Shanghai 200444,China;Shanghai Institute of Technical Physics,Chinese Academy of Sciences,Shanghai 200083,China;School of Electronic Electrical and Communication Engineering,University of Chinese Academy of Sciences,Beijing 100049,China)

机构地区：[1]上海大学通信与信息工程学院,上海200444 [2]中国科学院上海技术物理研究所,上海200083 [3]中国科学院大学电子电气与通信工程学院,北京100049

出　　处：《电子测量技术》2021年第13期138-144,共7页Electronic Measurement Technology

摘　　要：在天空背景的红外图像序列中检测低信噪比的弱小目标时,单纯的传统算法在不同程度上存在预处理过程复杂、特征设计困难、控制参数难以确定、检测准确率低等问题。通过引入深度学习技术,提出一种结合算法,可以显著提高算法的检测效果。在红外图像序列中,首先在起始帧中利用基于YOLOv3的时空特征提取网络高准确率地检测运动目标,再在后续帧中依据目标的速度和亮度特性,使用基于局部对比度特征的传统方法对目标进行快速检测。在搭建的天空背景的红外图像序列测试数据集中,结合方法实现了比现有方法更高的准确率和召回率,计算时间也满足实时性要求。该结果表明,两种方法互相配合,在实时性与准确度上取得了很好的平衡。When detecting dim and small targets with low signal-to-noise ratio in infrared image sequences of sky background, the simple traditional algorithm has some problems, such as complex preprocessing process, difficult feature design, difficult to determine control parameters, low detection accuracy and so on. Through the introduction of deep learning technology, a combined algorithm is proposed, which can significantly improve the detection effect of the algorithm. In the infrared image sequence, firstly, the moving target is detected with high accuracy by using the spatio-temporal feature extraction network based on YOLOv3 in the starting frame, and then the traditional method based on local contrast feature is used to detect the target quickly in the subsequent frames according to the speed and luminance characteristics of the target. In the infrared image sequence test data set of the sky background, the combined method achieves higher accuracy and recall than the existing methods, and the computing time also meets the real-time requirements. The results show that the two methods cooperate with each other and achieve a good balance in real-time and accuracy.

关 键 词：红外弱小目标 图像序列 深度学习 目标检测 

分 类 号：TP751.1[自动化与计算机技术—检测技术与自动化装置]