Spectral feature extraction and ensemble learning for multiclass aircraft damage identification

The research represents a robust methodology for identifying aircraft surface damage using hyperspectral imaging combined with ensemble machine learning. Surface degradation in aircraft, such as corrosion, burn marks, lightning strikes, weld defects, and paint peeling, is often difficult to detect using conventional inspection techniques. By leveraging high-resolution spectral data and domain-specific feature engineering, the proposed method enables accurate classification of ten different damage types using a structured machine-learning framework. Hyperspectral intensity data were collected from over 500 real and lab-induced samples using the Goldeneye hyperspectral camera, followed by the extraction of handcrafted features across spectral, statistical, and frequency domains. A soft voting ensemble of Random Forest (RF), Extreme Gradient Boost (XGBoost), and Support Vector Machine (SVM) models achieves a peak classification accuracy of 92.6 % with high accuracy across damage classes. This method supports real-time, non-contact, and scalable aircraft inspection workflows and demonstrates strong potential for integration with drone-based or robotic inspection systems in aerospace maintenance. Key contributions of this methodology: • A hyperspectral imaging-based pipeline for identifying different aircraft surface damage types • A robust ensemble model combining RF, XGBoost, and SVM for high-accuracy classification • Designed for integration into real-world, automated aircraft inspection systems.