摘要
在现代化战争中,广泛利用图像等载体获取信息,但雾天环境下得到的图像不仅影响场景呈现,而且会掩盖重要特征。为提高雾天图像在现代化战争的利用价值,提出一种基于全局补偿注意力机制的战场图像去雾方法。构建全局补偿模块保证输出图像的完整性,并加入通道下采样恢复清晰图像;使用密集残差模块学习退化图像和清晰图像的非线性映射,同时加入注意力机制提高网络的灵活处理能力;通过提升输入图像的通道数量确保网络充分学习特征信息。实验结果表明,与经典和新颖图像去雾方法比较,所提方法在主观和客观评价上均取得出色成绩,说明该方法将注意力机制和全局补偿模块充分结合,有效缓解了战场图像退化问题,同时注重特征增强,使信息得以完整呈现,具有更优越的性能。
In modern war,images and other carriers are widely used to obtain information.However,the images obtained in foggy environment not only affect scene rendering,but also mask important features.In order to improve the use value of fog-degraded images in modern warfare,a battlefield image dehazing method based on the global compensation attention mechanism is proposed.A global compensation module is constructed to ensure the integrity of output image,and channel down sampling is added to restore the clear image.The dense residual module is used to learn the nonlinear mapping between degraded image and clear image.In addition,an attention mechanism is added to improve the flexible processing capability of the network.The network can fully learn the feature information by increasing the number of channels of the input image.The experimental results show that the proposed method achieves excellent results in both subjective and objective evaluation compared with classical or novel image dehazing methods.The proposed method fully combines the attention mechanism with the global compensation module to effectively alleviate the problem of battlefield image degradation.At the same time,it pays attention to feature enhancement to enable the complete presentation of information and ultimately achieve better performance.
作者
林森
王金刚
高宏伟
LIN Sen;WANG Jingang;GAO Hongwei(School of Automation and Electrical Engineering,Shenyang Ligong University,Shenyang 110159,Liaoning,China)
出处
《兵工学报》
EI
CAS
CSCD
北大核心
2024年第4期1344-1353,共10页
Acta Armamentarii
基金
辽宁省高等教育创新人才计划扶持项目(LR2019058)
辽宁省教育厅高等学校基本科研项目(LJKMZ20220615)。
关键词
战场图像去雾
全局补偿
注意力机制
密集残差模块
battlefield image dehazing
global compensation
attention mechanism
dense residual module
