<p>During industrial production, product quality issues are inevitable but rare, typically arising from incidental factors such as equipment damage or production process errors. Consequently, the limited availability of quality issue data, such as surface defects, hampers the training of high-performance detection models using deep learning techniques. To address this challenge, a novel data augmentation method tailored for surface defects on industrial products is proposed in this paper. Inspired by the Segment Anything Model (SAM), an industrial SAM adapter (ISA) is designed to enable the precise extraction of defects from industrial data. The identified defects are then seamlessly fused with classified defect-free samples using refined Poisson blending to generate new defective samples, which are used to expand the original dataset. To validate the efficacy of the proposed approach, this study establishes a new dataset named YSS, primarily focusing on scratches found on steel plate surfaces. Comprehensive ablation experiments are conducted on YSS, along with experiments on public datasets such as KSDD and NEU-SEG, to assess the generalization and robustness of the proposed method. The experimental results demonstrate that the data augmentation technique proposed in this paper significantly enhances the F1-score metrics of the detection model on the YSS, KSDD, and NEU-SEG datasets by 5.3%, 1.4%, and 4.2%, respectively. These findings effectively demonstrate the effectiveness, generalization, and robustness of the proposed method.</p>