基于红外图像的天空检测算法研究

Research on Sky Detection Algorithm Based on Infrared Image

为解决红外目标探测虚警率高问题,提出在天空背景条件下的目标检测方法,并实践验证基于红外图像的天空检测算法。利用注意力引导的金字塔上下文网络(AGPCNet)算法,首先,注意力引导的上下文块(AGCB)将特征图划分为块来计算局部关联,并使用全局上下文注意力(GCA)来计算语义之间的全局关联。随后,上下文金字塔模型(CPM)来整合多尺度AGCB中的特征。最后,引入非对称融合模块(AFM),进一步提高特征的利用率。实验结果表明,该算法能够有效地检测红外图像中的天空部分,验证了文中算法的有效性与鲁棒性。

In order to solve the problem of high false alarm rate in infrared target detection,a target detection method under sky background is proposed and the sky detection algorithm based on infrared image is verified in practice.An attention-guided pyramid context network(AGPCNet)algorithm is used.Firstly,attention-guided context blocks(AGCB)are used to divided feature graphs into blocks to calculate local association between semantics,and global context attention(GCA)is used to calculate global associations between semantics.Subsequently,a context pyramid model(CPM)is used to integrate features in multi-scale AGCB.Finally,the asymmetric fusion module(AFM)is introduced to further improve the utilization rate of features.Experimental results show that the proposed algorithm can effectively detect the sky in infrared images and the effectiveness and robustness are verified.