arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2303.16636eess.IV
收藏

用于参数高效的高光谱单图像超分辨率的操作神经网络

Operational Neural Networks for Parameter-Efficient Hyperspectral Single-Image Super-Resolution

Alexander Ulrichsen, Paul Murray, Stephen Marshall, Moncef Gabbouj, Serkan Kiranyaz, Mehmet Yamac, Nour Aburaed

AI总结:

本文提出将超分辨率模型的卷积滤波器扩展为基于麦克劳林级数展开的操作滤波器,构建参数高效的操作神经网络,在高光谱图像上实现更优的超分辨率性能。

摘要:

高光谱成像是遥感中的一种重要工具,它比标准彩色图像捕获更多的光谱信息。然而,光谱信息的增加是以牺牲空间分辨率为代价的。超分辨率是一种流行的技术,其目标是生成低分辨率输入的高分辨率版本。大多数现代超分辨率方法使用卷积神经网络。然而,卷积是一种线性操作,网络依赖每层的非线性激活函数来提供学习复杂底层映射所需的非线性。这意味着卷积神经网络往往非常深才能达到其效果。最近,自组织操作神经网络被提出,旨在通过使用麦克劳林级数展开用可学习的非线性函数替代卷积滤波器来克服这一限制。本工作重点是将一个流行的超分辨率模型的卷积滤波器扩展为更强大的操作滤波器,以增强其在高光谱图像上的性能。我们还研究了残差连接和不同归一化类型对这种增强网络的影响。尽管参数比等效的卷积网络更少,我们的结果表明,操作神经网络在小型高光谱图像数据集上实现了更优的超分辨率性能。我们的代码已在Github上公开:https://github.com/aulrichsen/SRONN。

英文摘要:

Hyperspectral Imaging is a crucial tool in remote sensing which captures far more spectral information than standard color images. However, the increase in spectral information comes at the cost of spatial resolution. Super-resolution is a popular technique where the goal is to generate a high-resolution version of a given low-resolution input. The majority of modern super-resolution approaches use convolutional neural networks. However, convolution itself is a linear operation and the networks rely on the non-linear activation functions after each layer to provide the necessary non-linearity to learn the complex underlying function. This means that convolutional neural networks tend to be very deep to achieve the desired results. Recently, self-organized operational neural networks have been proposed that aim to overcome this limitation by replacing the convolutional filters with learnable non-linear functions through the use of MacLaurin series expansions. This work focuses on extending the convolutional filters of a popular super-resolution model to more powerful operational filters to enhance the model performance on hyperspectral images. We also investigate the effects that residual connections and different normalization types have on this type of enhanced network. Despite having fewer parameters than their convolutional network equivalents, our results show that operational neural networks achieve superior super-resolution performance on small hyperspectral image datasets. Our code is made available on Github: https://github.com/aulrichsen/SRONN.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑