基于混合域注意力机制的神经网络反演大气湍流强度
Inversion of Atmospheric Turbulence Strength by Neural Network Based on Mixed-domain Attention Mechanism
张宝银 1尹伟石 1孟品超 1周林华 1齐德全1
作者信息
- 1. 长春理工大学数学与统计学院,吉林长春 130022
- 折叠
摘要
本文提出了一种基于混合域注意力机制的神经网络方法反演大气湍流强度.神经网络的输入为不同大气湍流强度下的退化图像,输出为表征大气湍流强度的折射率结构常数.混合域注意力机制由空间域和通道域双重注意力机制组成,其中空间域注意力机制用于增强退化图像中受湍流影响的区域特征,通道域注意力机制用于增强由湍流引起的颜色和纹理特征.在网络训练阶段,引入的混合域注意力机制让神经网络更专注于退化图像中与大气湍流强度相关的特征,提高了模型的精度.数值实验结果表明,本文提出的方法能够较准确地实现大气湍流强度反演.
Abstract
A neural network method based on the mixed-domain attention mechanism is proposed to invert the atmospheric turbulence strength.The input of the neural network is the degraded images under different atmospheric turbulence strengths,and the output is the refractive index structure constant that characterises the atmospheric turbulence strength.The mixed-domain attention mechanism consists of dual spatial-domain and channel-domain attention mechanisms,where the spatial-domain attention mechanism is used to enhance turbulence-affected region features in the degraded image,and the channel-domain attention mechanism is used to enhance turbulence-induced colour and texture features.In the network training stage,the introduced mixed-domain attention mechanism allows the neural network to focus more on the features in the degraded image that are related to the atmospheric turbulence strength,which improves the accuracy of the model.Numerical experimental results show that the method proposed in this paper can achieve atmospheric turbulence strength inversion more accurately.
关键词
混合域注意力机制/折射率结构常数/湍流强度反演/退化图像Key words
mixed-domain attention mechanism/refractive index structure constant/turbulence strength inversion/degraded image引用本文复制引用
基金项目
吉林省自然科学基金(20220101040JC)
出版年
2024
NETL
NSTL
万方数据