A Comprehensive Survey of Deep Learning in Remote Sensing: Theories, Tools and Challenges for the Community

摘要: In recent years, deep learning (DL), a re-branding of neural networks (NNs), has risen to the top in numerous areas, namely computer vision (CV), speech recognition, natural language processing, etc. Whereas remote sensing (RS) possesses number unique challenges, primarily related sensors and applications, inevitably RS draws from many same theories as CV; e.g., statistics, fusion, machine learning, name few. This means that community should be aware of, if not at leading edge advancements like DL. Herein, we provide most comprehensive survey state-of-the-art DL research. We also review new developments field can used for RS. Namely, focus on theories, tools challenges community. Specifically, unsolved opportunities it relates (i) inadequate data sets, (ii) human-understandable solutions modelling physical phenomena, (iii) Big Data, (iv) non-traditional heterogeneous sources, (v) architectures algorithms spectral, spatial temporal data, (vi) transfer (vii) an improved theoretical understanding systems, (viii) high barriers entry, (ix) training optimizing