智能抑制复杂空间背景下的红外弱小目标检测方法  

Intelligent suppression of infrared dim and small target detection method under complex space backgrounds

作　　者：葛诗然 刘睿恺 李娜 赵慧洁 GE Shiran;LIU Ruikai;LI Na;ZHAO Huijie(Institute of Artificial Intelligence,Beihang University,Beijing 100191,China;Beijing Institute of Remote Sensing Equipment,Beijing 100854,China)

机构地区：[1]北京航空航天大学人工智能研究院,北京100191 [2]北京遥感设备研究所,北京100854

出　　处：《红外与激光工程》2024年第11期266-278,共13页Infrared and Laser Engineering

摘　　要：在超远距离红外目标探测中,由于杂散光、探测器热传导及闪元盲元等复杂干扰,红外图像的背景常表现为非均匀性。同时,目标成像尺寸小,缺乏明显的形状和纹理特征,增加了检测与识别的难度。传统的特征提取方法易出现大量虚警,深度学习方法在特征提取方面具有优势,但在复杂背景干扰下训练难度较大。文中将计算机视觉领域中的背景重建问题与红外图像弱小目标检测任务相结合,提出了一种基于复杂背景智能抑制的红外弱小目标检测方法。该方法采用编码器-解码器架构设计了红外场景优化编解码背景抑制网络模型,引入多级融合机制和残差融合模块以实现多尺度特征提取和多层次特征融合,并提出感知一致性损失函数提高背景重建的鲁棒性。通过背景残差抵消策略有效实现背景抑制,最终结合全局阈值分割完成弱小目标检测任务。实验结果表明,与对比方法相比,文中方法在抑制背景方面背景标准差最高降幅达43.41%,目标信噪比最高提升至110.0257。在目标检测方面,四组数据中检测率均超过95%,展现出优异的检测效果,具有较强的工程实用性,为复杂背景下的红外弱小目标检测任务提供了新的解决方案。Objective Long-distance infrared target detection technology utilizes infrared detection systems to accurately capture targets,demonstrating significant potential in aerospace fields such as space target detection and battlefield reconnaissance.In practical applications of infrared imaging,numerous engineering challenges arise,such as stray light,detector thermal conduction,and structural radiation.These complex interference factors can cause non-uniform backgrounds in images,affecting image quality and target detection accuracy.Additionally,when the imaging distance is long,the target appears small,with weak signal strength and unclear characteristics,which may result in detection failure due to interference.Therefore,this paper addresses the issues of complex backgrounds interference and weak targets in space infrared images by developing a method for Infrared dim and small target detection based on intelligent suppression method for complex backgrounds.This research provides new solutions for infrared target detection applications in complex backgrounds.Methods This paper designs a method for detecting dim and small infrared targets based on intelligent suppression of complex backgrounds.The infrared scene-optimized encoder-decoder background suppression network model is designed to efficiently suppress complex backgrounds(Fig.2).A combination of multiple loss functions is employed to enhance the robustness of background reconstruction(Eq.14).The weak target detection task is accomplished using a threshold segmentation method(Eq.9).The effectiveness of the method is validated through four evaluation metrics:background standard deviation,target signal-to-noise ratio,detection rate,and false alarm rate(Fig.5,Tab.3-Tab.6).Results and Discussions The experimental results were obtained on four infrared weak target datasets.The background suppression results of different detection methods on the test images are shown in Fig.5.The results obtained by the proposed method using the infrared scene-optimized encoder-deco

关 键 词：红外弱小目标检测 背景抑制 残差网络 阈值分割 深度学习 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]