Investigating the viability of detecting rooftop solar photovoltaics via convolutional neural networks using remote sensing images

One of the finest solutions to producing electricity from non-renewable sources is rooftop solar photovoltaics. It is crucial to diversify energy sources from renewable sources as much as possible from a consumption point of view in order to reach global net zero rapidly. In order to reduce loss and use the sun’s energy, we must install rooftop solar photovoltaics for captive consumption to a significant level. Understanding current scale installations of solar photovoltaics on building roofs under various built-up conditions is crucial for achieving this. This study uses high resolution remote sensing images to train a lightweight convolutional neural network architecture referred to Faster R-CNN to identify solar panels installed on building roofs. Pre-trained models are trained using publicly accessible datasets to achieve generalization. The model is also evaluated using manually prepared labels in several cities around the world. The limits of such models in generalizing to different geographies are demonstrated through statistical model evaluation and a visual examination.