Abstract:To address the issues of low detection accuracy and high false detection rate caused by background interference and variable feature scales in pipeline detection, this paper proposes an improved YOLOv8n-based model named FHS-YOLOv8 (Feature High Selection-YOLOv8). First, a Feature Pyramid Shared Convolution module (FPSC) is introduced, which enhances the model's multi-scale feature extraction capability by employing shared convolution layers with different dilation rates. Second, a High-level Selection Path Aggregation Network (HS-PAN) is proposed, which integrates innovative designs of high-level feature selection and path aggregation. Through dynamic feature selection and dual-path feature fusion, it effectively addresses the multi-scale challenges. Finally, the SEAM module is incorporated into the head network, enabling the model to adaptively adjust its "attention" focus to optimize detection performance. The experimental results demonstrate that the improved algorithm exhibits a 3.3% increase in accuracy and a 4.1% improvement in mean average precision (mAP@0.5) on the pipeline defect dataset compared to the YOLOv8n model, while also reducing the number of parameters by 32.3%. The FHS-YOLOv8 model significantly outperforms traditional methods in complex scenarios, showcasing enhanced robustness and precision.