【摘要】 近年来，随着遥感技术的快速发展，遥感对地观测数据获取量与日俱增。在对海量遥感数据的特征提取与表征上，基于深度学习的智能遥感影像解译技术展现出了显著优势。然而，遥感影像智能处理框架和信息服务能力还相对滞后，开源的深度学习框架与模型尚不能满足遥感智能处理的需求。在分析现有深度学习框架和模型的基础上，针对遥感影像幅面大、尺度变化大、数据通道多等问题，本文设计了嵌入遥感特性的专用深度学习框架，并重点讨论了其构建方法，以及地物分类任务的初步试验结果等。本文提出的智能遥感解译框架架构将为构建具备多维时空谱遥感特性的深度学习框架与模型提供有力支撑。更多还原

【Abstract】 The rapid development of remote sensing technology has achieved massive remote sensing images, and the deep-learning-based remote sensing image interpretation has shown certain advantages in image feature extraction and representation. However, the intelligent processing framework and information service capabilities are relatively lagging. Open-source deep learning frameworks and models cannot yet meet the requirements of intelligent remote sensing processing. Based on the analysis of existing intelligent frameworks and models, we design a dedicated deep learning framework and model with remote sensing characteristics for the problems of large remote sensing image size, large-scale changes, and multiple data channels. The focus is on the construction of a dedicated framework that takes into account remote sensing data characteristics and the preliminary experimental results on remote sensing image classification. The design of this remote sensing image interpretation framework will provide strong support for the construction of a dedicated deep learning framework and models that integrate the temporal, spatial, and spectral features of remote sensing data.更多还原