Wind Turbine Classification for Autonomous Inspection: A Case Study Using Deep Learning

Recent developments in drone technology facilitate wind turbine inspection with lower downtime, cost, and health risk. The typical method of drone inspection, however, requires a certified pilot to manually fly the drone around the wind turbines being inspected. In pursuit of creating a fully autonomous drone system for inspecting wind turbines, this paper evaluates three deep learning architectures for the task of wind turbine classification’ which means determining whether an RGB image contains wind turbines. In the eventual autonomous drone system, this visual feedback will enable the drone to confirm it has reached the wind turbine rather than only relying on GPS, which has a degree of uncertainty. This research focuses on implementing and tuning the VGG19, ResNet50, and Xception convolutional neural networks to achieve high accuracy at this task. The outcomes are described in this paper along with the dataset created to train and test the networks.