基于结构张量的图像融合方法在海上探测的应用  

Application of Structure Tensor-Based Image Fusion Method in Marine Exploration

作　　者：马晓熠 陈奕宏[1,2,3] 王飞 谢硕[1,2] MA Xiaoyi;CHEN Yihong;WANG Fei;XIE Shuo(China Ship Scientific Research Center,Wuxi 214026,China;Taihu Laboratory of Deepsea Technological Science,Wuxi 214026,China;School of Aeronautics and Astronautics,Zhejiang University,Hangzhou 310058,China)

机构地区：[1]中国船舶科学研究中心,江苏无锡214026 [2]深海技术科学太湖实验室,江苏无锡214026 [3]浙江大学航天航空学院,浙江杭州310058

出　　处：《水下无人系统学报》2025年第1期84-91,共8页Journal of Unmanned Undersea Systems

摘　　要：单一传感器无法在海上探测中取得良好的效果。红外与可见光具有很强的互补性,将二者融合可以得到高质量的融合图像,能够更准确、全面地感知海上目标。然而现有的融合方法并未应用于海上探测领域,融合方法均缺少针对性,融合效果差,并且缺少应用于海上融合的深度学习数据集。文中对基于结构张量的深度学习图像融合方法进行研究,针对海上目标的特点进行改进与优化,加入多尺度卷积并按照通道对图像进行融合,旨在获取目标显著且信息全面的高质量彩色融合图像。使用采集的数据进行实验,综合选取多种评价指标开展对比仿真实验研究。研究结果表明,改进的图像融合方法在6个指标上的融合效果优于原始算法,综合性能优于其他常用的10种图像融合算法,改进方法的泛化性在其他公开数据集上得到了验证。改进后的基于结构张量的图像融合方法在海上感知中有优异的表现,融合结果突出目标特征,融合性能优于其他方法。A single sensor is insufficient for effective marine detection.Infrared light and visible light have strongcomplementarity,and fusing them can generate high-quality images that enable more accurate and comprehensive detection ofmarine targets.However,existing fusion methods have not been applied in marine detection and are not specifically developedfor it,leading to poor fusion results.Additionally,there is a lack of deep learning datasets tailored for marine image fusion.Toobtain high-quality color fusion images with a prominent performance in detecting targets and obtaining comprehensiveinformation,the deep learning-based image fusion method using structure tensors was optimized based on the characteristics ofmarine targets.Multi-scale convolution was incorporated,and image fusion was performed according to channels.Thecollected data were used for comparative simulation experiments,with a variety of evaluation metrics applied.The resultsindicate that the improved image fusion method outperforms the original algorithm in six metrics,and its overall performanceis better than the other ten commonly used image fusion algorithms.Furthermore,its generalization has been validated on otherpublic datasets.The improved structure tensor-based image fusion method has an excellent performance in maritime situationalawareness,with fusion results highlighting target features and surpassing the performance of other methods.

关 键 词：海上探测 图像融合 深度学习 结构张量 

分 类 号：TJ630[兵器科学与技术—武器系统与运用工程] U662.9[交通运输工程—船舶及航道工程]