摘要
红外和可见光图像融合的优势在于不同图像之间进行图像信息互补,以增强图像的细节信息。目前,深度学习融合方法存在融合性能和计算资源消耗之间的不平衡的问题。此外,部分融合规则不能有效结合不同类型图像的特征信息。因此,提出了一种基于改进边缘卷积的红外和可见光图像融合算法。首先,在训练阶段,利用嵌入式边缘算子的结构重参数化边缘卷积,增强了源图像中纹理特征的提取能力;其次,利用注意力融合模块从不同类型图像特征中获得独特信息和共有信息。实验结果表明,该算法以低计算成本实现了优越性能,不仅具有更好的视觉效果,还能提供更丰富的场景细节信息。
The advantage of infrared and visible light image fusion is that the image information is complementary between different images to enhance the detailed information of the image.Currently,deep learning fusion methods have the imbalance between fusion performance and computational resource consumption.Moreover,some fusion rules do not effectively combine the feature information of different types of images.Therefore,a fusion algorithm for infrared and visible images based on improved edge convolution.First,the structure reparametric edge convolution of the embedded edge operator is used in the training stage to enhance the extraction ability of texture features in the source image.Secondly,an attention fusion module was used to obtain unique and common information from different types of image features.The experimental results show that the algorithm in the paper achieves superior performance with low computing cost.Therefore,the algorithm has better visual results and provides richer scene detail information.
作者
郝昱权
郭相茹
HAO Yuquan;GUO Xiangru(Applied Technology College,Zhumadian Teachers College of Early Childhood Education,Zhumadian 463000,China;School of Art,Zhumadian College of Preschool Education,Zhumadian 463000,China)
出处
《传感器世界》
2025年第5期27-33,40,共8页
Sensor World
基金
2024年度驻马店幼儿师范高等专科学校校级课题(No.24L010)。
