Homeowners are increasingly deploying rooftop solar photovoltaic (PV) arrays due to the rapid decline in solar module prices. Homeowners might need to spend up to $375 to diagnose their damaged or degraded rooftop solar PV system. Thus, recently, there is a rising interest to inspect potential damage on solar PV arrays automatically and passively with a lower cost. Current approaches that leverage machine learning (ML) and other data analytical techniques have the limitation of distinguishing solar array damages from other solar degradation, e.g., shading, dust, snow, and etc.
To address this problem, we design a new system-SolarDiagnostics that can automatically and accurately detect and localize any damage that may exist on rooftop solar PV arrays using their rooftop images with a lower cost. In essence, SolarDiagnostics first leverages an unsupervised segmentation algorithm to isolate the objects on rooftops to extract solar panel residing contours. Then, SolarDiagnostics employs a Deep Convolutional Neural Networks (CNNs) to accurately identify and characterize any damage that may exist in each image contour. We evaluated SolarDiagnostics using 60,000 damaged solar PV array images generated by Deep Convolutional Generative Adversarial Networks) (DC-GANs). In addition, we evaluated SolarDiagnostics by building a prototype (cost as $35) that integrates a drone with HD camera to accurately examine the rooftops of 10 "mock" solar deployments. We found that SolarDiagnostics is able to detect damage on solar PV arrays at a Matthews Correlation Coefficient (MCC) of 1.0. In addition, pre-trained SolarDiagnostics yields a MCC of 0.95, which is significantly better than the re-trained ML approaches and is the nearly same as the retrained SolarDiagnostics.
We make the source code and datasets that we used to build and evaluate SolarDiagnostics publicly-available.