Aerospace aluminum surface defect detection method based on Multi-Scale Convolution and attention mechanism

Zhang, Rui and Cai, Shanshan and He, Zhen and Zhao, Yingjie (2026) Aerospace aluminum surface defect detection method based on Multi-Scale Convolution and attention mechanism. Scientific Reports. ISSN 2045-2322

Abstract

The detection of small target defects on aluminum surfaces is critical in modern manufacturing, particularly in sectors such as aerospace, automotive, electronics, and high-end equipment manufacturing. These defects can severely compromise the safety, stability, and durability of products. However, due to the diversity of surface defect types, their small size, and the presence of complex background interference, traditional detection methods often struggle to achieve high precision under conditions of low contrast and significant noise. To address this challenge, this paper proposes an enhanced small target detection method for aluminum surfaces, leveraging the YOLOv11n framework. Specifically, we introduce a dilation-wise residual and dilated reparameterization block module to strengthen the model's feature extraction capabilities, thereby improving the capture of fine details in small targets. In addition, the SimAM attention mechanism is integrated to optimize the model's focus on critical feature regions, further enhancing its sensitivity and recognition performance for small defects. Moreover, we incorporate the CARAFE (Content-Aware ReAssembly of Features) upsampling operator, which effectively enlarges small target details and mitigates the information loss inherent in conventional upsampling techniques, thus significantly boosting detection accuracy. Experimental results show that the proposed model achieves a mean average precision (mAP@0.5) of 79.4% and a recall of 76.6%, reflecting improvements of 2.9% and 4.4% over the baseline model, respectively. Compared to existing methods, our approach demonstrates notable advantages in both detection accuracy and recognition ability, providing a promising foundation for future practical applications in industrial scenarios.