SEA-YOLOv8: An Enhanced Method for Detecting Small Targets in Aeroengine Components Based on YOLOv8n

Abstract

As object detection technology continues to gain widespread application in industrial and commercial domains, deep learning-based object detection methods are showing considerable potential in the maintenance of aeroengine. The complex structure of the aeroengine, along with the diversity in component shapes and the high precision required during detection, significantly increases the technical challenges and complexity involved in this process.To address these challenges, this article presents an improved YOLOv8n-based detection model, specifically designed to enhance small target detection performance in complex backgrounds. First, the Neck network was redesigned using the selective boundary aggregation (SBA) module for multiscale feature fusion, improving the model's ability to capture features at different resolutions and enhancing small target detection. Second, the proposed C2f-ELA module, which incorporates an efficient local attention (ELA) mechanism, further strengthens the model's ability to capture long-range dependencies and increases its robustness in complex backgrounds. Finally, an attention-based intrascale feature interaction module was introduced during the feature integration process to enhance the model's feature representation capabilities. Experimental results show that, compared to the YOLOv8n model, the proposed model achieves a 15.3% reduction in parameters, a 9.4% increase in precision, a 6% improvement in recall, and gains of 8.1% and 5.6% in mAP50 and mAP50:95, respectively. Furthermore, this approach allows for accurate identification of aeroengine components under constrained hardware resources, making it more suitable for engineering deployment and practical use in aeroengine inspection systems.