基于关键点检测的红外弱小目标检测

An infrared small target detection method based on key point

红外弱小目标检测旨在从复杂的背景中检测出红外弱小目标,该技术在监视预警系统、精确制导等方面具有重要应用价值。针对已有的传统算法存在漏检、误检,深度学习中基于语义分割的检测方法易受“过分割”与“欠分割”影响等问题,提出了一种基于关键点检测的红外弱小目标检测算法(KeypointNet)。主要创新点为:直接优化目标中心点坐标,提高了检测效率,有效地保证了目标的检测率与虚警率;设计了一种从低层级到高层级的特征融合模块,获取了目标的多尺度信息,提升了检测效果。在相关数据集上的实验表明:KeypointNet算法的检测率能够达到97.58%,同时虚警率在2%以下,与其他算法相比取得了最好的效果。

Infrared small target detection aims to detect infrared dim small targets from complex background.With wide application of this technology in monitoring and early warning systems,precise guidance,etc.,how to realize fast and accurate detection of infrared small targets is of great importance.Existing traditional algorithms often result in missed detection or false detection,and the detection method based on semantic segmentation in deep learning is easy to be affected by“over segmentation”and“under segmentation”.To address these problems,an infrared small target detection algorithm is proposed based on key point detection(KeypointNet).The main innovations of this study are as follows.First,The coordinates of the target center point is directly optimized,improving detection efficiency and effectively ensuring the detection rate and false alarm rate of the target.Second,a low-high level feature fusion mod⁃ule is designed to obtain multi-scale information of the target and improve the detection effect.Experimental results on relevant data sets show that in comparison with other algorithms,the algorithm proposed can achieve better effect,with the detection rate reaching 97.58%and the false alarm rate being reduced to less than 2%.