Edinburgh Napier University

  Internet of Things (IoT) has been rapidly developed in recent years, being well applied in the fields of Environmental Surveillance, Smart Grid, Intelligent Transportation, and so on. As one of the typical earth-based meteorological observation methods, networked Doppler weather radars, i.e. the Internet of weather Radars (IoR) can detect the signals of large-area water particles in the atmosphere with high resolution, but suffer from beam blockage due to surrounded mountains, buildings, as well as other obstacles. In addition, how to establish a distributed platform for large-scale radar data analytics becomes critical and challenging, especially considering optimised strategies on the storage, processing and exchange of radar raw data, beam/echo signal, and final products etc. In this paper, an edge-assisted cloud framework is proposed to facilitate effective and proficient communication and progression, where echo signal from a single site radar can be analysed and pre-processed at the edge, and then trained in the cloud with elastic resources and distributed learning ability. A Residual Concatenate Fully Convolutional Network (RC-FCN) is presented for beam blockage correction, which is integrated into the framework to be compared with other deep learning models, including FCN, ResNet, VGG, etc. According to experiment results, better performance and efficiency have been achieved using the proposed framework and its fitted RC-FCN model.