【摘要】 为应对传统油漆表面缺陷检测方法效率低且易受工人经验与主观因素干扰，以及现有检测模型参数量大、精度不足的问题，提出一种轻量高效的HPSDD-YOLOv8船体油漆表面缺陷检测算法。该算法基于YOLOv8n模型进行改进，首先采用H-Feature模块，通过多分支特征提取、拼接及残差连接技术，在保证检测精度的同时降低参数量；其次将Star Block整合至C2f网络，形成计算量更小的P-C2f模块，有效保留空间与通道特征并增强特征提取能力；最后引入SDD-Detect模块，结合非对称解耦头与多级通道机制，降低模型复杂度并提升训练速度。实验结果表明，在船体油漆表面缺陷数据集上，HPSDD-YOLOv8模型的mAP较YOLOv8模型提升17.51%，检测速度提高48.1帧/s，计算复杂度（FLOPs）减少22.5 G，验证了该算法在兼顾检测精度与速度方面的有效性。更多还原

【Abstract】 To address the inefficiency and susceptibility to worker experience and subjective factors in traditional ship paint surface defect detection methods, as well as the issues of large parameter volume and insufficient accuracy in existing detection models, a lightweight and efficient HPSDD-YOLOv8 algorithm for ship paint surface defect detection is proposed.Based on the YOLOv8n model, this algorithm first employs the H-Feature module, which reduces parameter volume while maintaining detection accuracy through multi-branch feature extraction, concatenation, and residual connections. Second, the Star Block is integrated into the C2f network, forming the P-C2f module with reduced computational complexity, effectively preserving spatial and channel features while enhancing feature extraction capabilities. Finally, the SDD-Detect module is introduced, combining an asymmetric decoupled head and a multi-level channel mechanism to lower model complexity and improve training speed. Experimental results demonstrate that on the ship paint surface defect dataset, the HPSDD-YOLOv8 model achieves a 17.51% increase in mAP compared to the YOLOv8 model, a 48.1 fps improvement in detection speed, and a 22.5 G reduction in computational complexity(FLOPs), validating the algorithm’s effectiveness in balancing detection accuracy and speed.更多还原