摘要
目前学者常用的线性遥感图像融合方法大部分会有光谱失真较大,图像质量退化和计算量大等问题,深度学习是一种非线性的融合方法,利用非线性运算能提取到更具体的图像特征,利用卷积层进行特征融合和图像重构得到遥感图像融合图像。在卷积层使用小卷积核多卷积层和小池化核增强网络的拟合能力,同时为了防止过拟合在全连接层进入dropout层,增加神经网络的鲁棒性。
Traditional linear remote sensing image fusion methods usually have problems such as large computation amount of spectral distortion, etc. Deep learning based non-linear fusion method can make full use of the information of each convolution layer, extract more image features, and use the convolution layer for feature fusion and image reconstruction to obtain remote sensing image fusion image. At the convolution layer, small convolution kernels and small pooling kernels are used to enhance the fitting ability of the network, and to prevent overfitting from falling into the dropout layer at the fully connected layer, the robustness of the neural network is increased.
作者
赵学军
闫雪
杨威
梁轩宇
ZHAO Xuejun;YAN Xue;YANG Wei;LIANG xuanyu(School of Mechatronics and Information Engineering,China University of Mining and Technology(Beijing),Beijing 100083,China)
出处
《长江信息通信》
2022年第5期1-4,共4页
Changjiang Information & Communications