UAV Vision Based Wind Turbine Blades Defects Detecion via Semantic Segmentation

The timely maintenance of wind turbine blades is considered of high importance for the safety and effectiveness of Wind power stations. To address the concerns of inefficiencies and safety issue of traditional mannual blade damage detection methods, we employed UAV vision and proposed a novel semantic segmentation model based on DeepLabV3Plus. The major enhancement involved adopting SCTNet as the backbone network and integrating the EMA mechanism. Our experiments indicate that DeepLab V3Plus-SE performs exceptionally well in segmenting wind turbine blades and identifying defects. Moreover, the model shows substantial improvements in key performance metrics, namely mIoU and mAcc.