计算机科学
特征(语言学)
遮罩(插图)
图像融合
融合
人工智能
计算机视觉
干扰(通信)
红外线的
图像(数学)
模式识别(心理学)
光学
电信
物理
频道(广播)
哲学
艺术
视觉艺术
语言学
作者
Shuowei Jin,Jing-rong Chen,Shang Gao,Aiyun Yan
标识
DOI:10.1088/1361-6501/ade90b
摘要
Abstract Infrared and visible image fusion has important application value in the fields of industrial measurement, assisted driving, and satellite remote sensing. However, due to the significant differences in information distribution and feature expression between infrared and visible images, the existing fusion methods often show the problem of degradation of fusion effect under complex lighting conditions. To solve this challenge, we propose an infrared and visible image fusion network based on dynamic weight allocation to improve the clarity and target saliency of the fused image. Specifically, we propose a differential feature compensation module, which enhances the differential features of infrared and visible feature maps and supplements them to the backbone network through an improved attention mechanism, so as to improve the utilization rate of complementary features of heterogeneous images, and in addition, an environment aware module is designed to dynamically adjust the fusion ratio according to different lighting environments, so as to avoid the masking and weakening of the target features in the infrared image by the highlight area in the visible image. Experimental results show that the proposed network achieves better fusion effect than the existing methods in a variety of complex scenes, effectively suppresses strong light interference, and improves the clarity and contrast of image details.
科研通智能强力驱动
Strongly Powered by AbleSci AI