Hyperparameter Tuning of Support Vector Machines for Wind Turbine Detection Using Drones

With the increase in demand for renewable energy and high maintenance costs caused by equipment and parts failure, the need for advanced monitoring technologies in this field is essential. Of specific concern are the frequent failures of wind turbine blades due to occasional inspections using conventional techniques and slow maintenance. One of the important steps to address these concerns and reach autonomy is to use drones to detect the wind turbine for inspection. This becomes crucial when the drone must fly a long distance to reach the wind turbine or wind farm. More specifically, as the drone is reaching the vicinity of the wind turbine, real-time GPS information may not necessarily lead the drone to reach right in front of the desired turbine. To resolve these issues, we use support vector machines (SVM) to classify the wind turbine images from those images that do not contain the wind turbine. The implemented SVM as a machine learning model is further enhanced by various hyperparameter tuning methods to ensure the highest accuracy possible. Default hyperparameters, RandomSearch, and Bayesian Optimization with Hyperband tuning are used, and the accuracies vs training times are compared. The proposed model can be ultimately used in an automated system using a drone and aerial images to detect the wind turbine for blade inspection.