|
At present, the majority of deep learning building blocks, techniques, and architectures are based on real-valued operations and representations. However, over the past decades, neural networks in the complex domain have become a quite active field of research and have continued to open doors to various new applications, remote sensing being one of them. Complex-valued neural networks (CVNNs) deal with complex inputs, e.g. signals having phase and amplitude, which is one of the core concepts in signal processing. Despite their appealing properties and potential for opening up entirely new neural architectures, deep complex-valued neural networks have been marginalized due to limited availability of building blocks required for such model designing. This survey compactly summarizes the research efforts that demonstrates relevant applications of deep complex-valued neural networks in the realm of remote sensing. We examine the various remote sensing problems under study, the models employed and the sources of data used. An attempt is made to study the overall performance achieved by such complex-valued models, according to the evaluation metrics used by the authors and compare the competencies with their real-valued counterparts. Also, we illustrate the shortcomings, enhancements, and implementations associated and obtain a bird's-eye view of their present and future prospective. The overall findings indicate that CVNNs represent a promising technique with competitive performance in terms of classification accuracy and precision for a wide variety of remote sensing problems, outperforming conventional techniques. However, the success of each CVNN model is highly dependent on the nature of the data set used.
|
