Research on Aircraft Key Point Detection Method Based on HRNet Network

With the rapid global growth of military and civil aviation, aircraft state perception and pose estimation have become increasingly important for aviation safety assurance. Traditional methods based on manual observation and sensor monitoring suffer from response delays and insufficient coverage, making it difficult to meet the dual demands of real-time performance and high accuracy in modern aviation systems. To address these challenges, this paper proposed a lightweight improved architecture based on the High-Resolution Network (HRNet), named HRKNet, specifically designed for accurate detection of structural key points on aircraft. By integrating a ResNet backbone and a multi-scale feature fusion strategy, the proposed method significantly reduced model parameters and computational complexity while maintaining high-precision keypoint localization capabilities. To support model training and evaluation, we construct an aircraft keypoint dataset containing 1226 images, annotated with 9 critical structural points. Experimental results showed that HRKNet achieves an mAP of 96.81%, compresses the number of parameters to less than 1% of the original HRNet, and improves processing speed to 21.57 FPS, demonstrating strong potential for edge deployment. Further ablation studied validate the effectiveness of the keypoint hierarchical supervision mechanism and the high-resolution feature preservation strategy.