基于时空三维卷积网络的复杂背景下红外弱小目标检测方法

打开文本图片集

关键词：红外弱小目标；深度学习；目标检测；时空三维卷积 中图分类号：TP391.4 文献标识码：A doi：10.37188/CJLCD.05-0161 CSTR：317.14.CJLCD.05-0161

Abstract： In the field of aviation early warning，infrared weak target detection technology is crucial for long-range all-weather battefield perception.Aiming at the problem of low probability of target detection and high false alarm rate caused by a small proportion of pixels and lack of features of infrared dim and small targetsunder complex background，a detection method for infrared dim and small targets under complex backgrounds via a spatio-temporal three-dimensional convolutional network was proposed. This method proposes a feature extraction backbone network that combines 2D convolution with 3D convolution，and combines spatial texture features and inter-frame motion features to achieve collaborative perception of target structure and temporal changes. According to the characteristics of infrared dim and small targets，a local contrast module is designed as a feature enhancement module to expand the receptive field for feature enhancement； In addition，introducing asymmetric attention mechanism for feature fusion increases the preservation of textureand positional information；Finally，the point regression loss function is used to calculate the detection results. In the experiment，the public data set was compared with the selfbuilt data set，labeled and trained.Experimental results show that compared with the conventional multiframe target detection network，the improved algorithm has a recallrate improvement of no less than 7.52% and an average precision improvement rate of no less than 6.46% . It can be effectively applied to infrared dim and small target detection in complex backgrounds，and embodies good robustness and adaptability.

Key words： infrared dim and smalltarget；deep learning；object detection；spatio-temporal three-dimensional convolution

1引言

随着隐身技术的发展，要求作战双方一方面要在强对抗条件下采用电磁静默来提高己方生存能力，另一方面要提高对隐身目标的探测预警能力。（剩余15420字）