摘要
针对海洋弱目标监测存在因背景、海雾影响而"认不清"、"看不远"的问题,提出非下采样轮廓波变换和神经网络结合的多源图像融合算法。首先使用非下采样塔式结构滤波器组分解其预处理得到的偏振长波红外图像和可见光图像,进而采用神经网络得到初次融合图像与短波红外图像的图像特征,并从这些特征中提取权重,然后将特征图像取相对应的权重,融合得到最后的图像。该算法充分利用了红外图像亮度、强度信息和偏振光穿云透雾的特性,突出了目标轮廓细节,提高图像对比度,从而达到清晰识别海面目标的目的。
Aiming at the problems of weak target monitoring in the ocean with background and sea fog,which leads to the problems of"unrecognizable"and"not seeing far".A multi-source image fusion algorithm combining with non-sub-sampled contourlet transform and convolutional neural network is proposed.Firstly,the non-sub-sampled pyramid tower structure filter bank is used to decompose the polarized long-wave infrared image and visible light image obtained by preprocessing.Then,the neural network is used to obtain the image features of the first fusion image and the short-wave infrared image,and the weights are extracted from these features.Taking the corresponding weights of the feature images and being fused to get the final image.The algorithm makes full use of the characteristics of infrared image brightness,intensity information,and polarized light passing through clouds and fog to highlight the details of the target outline and improve the contrast of the image.Thereby the purpose of clearly identifying targets on the sea surface is achieved.
作者
孙雪峰
韩文波
黄丹飞
钟艾琦
田博文
SUN Xuefeng;HAN Wenbo;HUANG Danfei;ZHONG Aiqi;TIAN Bowen(College of Optoelectronic Engineering,Chahgchun University of Science and Technology,Changchun 130022,China)
出处
《激光杂志》
北大核心
2020年第7期80-84,共5页
Laser Journal
基金
国家自然科学基金重大项目(No.61893096010)。
