Evaluation of the utility of radar data to provide model parameters for energy system analysis

Abstract

Based on recent advances, deep learning (DL) has proven to be able to achieve an outstanding performance in remote sensing. Nevertheless, has shown to be limited to the evaluation of the optical data. Beside many previous researches to introduce DL in Synthetic Aperture Radar (SAR), analysing the energy systems by means of above-mentioned techniques offers a huge potential to be explored. From the satellite data, energy systems parameters can be derived which will lead to an improvement in quality and completeness for the existing databases. In this context, this research contributes to the generation of a new worldwide open database, by means of a uniform methodology which can be updated continuously. Hence, in this master thesis a novel methodology is developed for automatic extraction, provision of information on the type of the plant and the exact geo-location for two main energy generators respectively Wind Turbine and Coal-fired Power Plants. Simultaneously paying special heed on being competitive in accuracy and computational time. For this purpose, Faster RCNN a state-of-the-art DL framework for object detection composed with Res Net 50 + FPN as a backbone are utilized to model the architecture. The dataset annotation is designed to be automatic and utilize manually created geolocations plus open databases e.g Global Power Plant Database (GPPD) to compass this complex and time-consuming process faster. Suitable areas with extensive geographical scope are selected for implementing the developed approach. Moreover, analysis and evaluation of the achieved results graphically shows the algorithm capability to work in different geographical scales. Finally, the conducted accuracy assessment reveals the capability of the developed methodology in this master thesis to ensure an accuracy of 87.71% in large scale applications for individual Wind Turbine detection, and 92.41% in large scale applications for Coal-fired Power Plants detection.