Vision-based Propeller Damage Inspection Using Machine Learning

Abstract

Unmanned Aerial Vehicles (UAVs) play an increasingly pivotal role in day-to-day rescue operations, offering crucial aerial support in challenging terrain and emergencies, such as drowning. Drone hangars are strategically deployed to ensure swift response in remote locations, overcoming range-limiting constraints posed by battery capacity. However, the UAV's airworthiness, typically ensured through conventional inspections by a technical individual, is paramount to guarantee mission safety. Over time, UAVs are prone to degradation through contact with the external environment, with propellers often being the cause of flight instability and potential crashes. This paper presents an innovative approach to automate UAV propeller inspection to avert incidents preemptively. Leveraging visual recordings and deep learning methodologies, we train a Convolutional Neural Network (CNN) model using both passive and active learning strategies. Our approach successfully detects physical damage on propellers with an impressive accuracy of 85.8%, promising a significant improvement in maintaining UAV flight safety and effectiveness in rescue operations.