Optimization of YOLOv8 for Defect Detection and Inspection in Aircraft Surface Maintenance using Enhanced Hyper Parameter Tuning

Aviation repair is governed by strict standards, because maintaining the integrity of aircraft surfaces is essential. Conventional inspection techniques, which frequently rely on manual examinations, are time-consuming and prone to human mistake. Through a SEM model picture collection of aircraft surface flaws, this study investigates the improved use of YOLOv8 (You Only Look Once version 8) for diagnosing and identifying surface concerns in aviation maintenance. In order to precisely identify, locate, and classify different defect kinds, such as aggressive pitches and thin resists, it presents a revolutionary ensemble deep learning approach. In order to particularly assess SEM pictures with defects including gaps, potential gaps, line collapses, bridges, and micro bridges, the research entails training MobileNet backbone models. To further enhance detection and classification performance, the suggested approach combines predictions from several models using a Weighted Box Fusion (WBF) ensembling strategy. According to experimental findings, the optimized YOLOv8 model outperforms conventional YOLOv8 models in terms of accuracy and speed thanks to its WBF ensembling and MobileNet backbone. This sophisticated model is excellent at spotting a variety of surface flaws, such as corrosion, fractures, and other irregularities. Aircraft surface restoration has advanced significantly with the integration of MobileNet and WBF into the improved YOLOv8 model, improving Mean Average Precision (mAP) for difficult defect categories. Metrics like the F1-score, accuracy, precision, recall, specificity, and confidence score are used to assess how well this method is at reliably detecting and localizing different types of defects in SEM pictures by utilizing deep learning.