This is the authors' abstract. We don't add key points for this paper.
Recently, surface defect detection in industrial products has received widespread attention, but it faces challenges such as significant variations in defect sizes and diverse defect categories. To address this issue, we propose a two-stage network that incorporates information fusion and integration mechanisms, dubbed FI2Net. The proposed network is enhanced with information fusion and integration through two stages, thereby enabling each layer to obtain more comprehensive information. To guarantee the effectiveness of subsequent information fusion with minimal computational burden, we designed a multi-level information alignment module (ML-IAM). In particular, an multi-level information fusion module (ML-IFM) is incorporated to capitalize on its advantages in cross-layer information utilization, facilitated by convolutional or self-attention mechanisms. Furthermore, to further target the integration and enhance the information representation of a specific layer, an information integration module (IIM) is devised, which can effectively combine the integrated information with a specific target layer. Experimental results on the NEU-DET, GC-10, and TCAL datasets demonstrate the effectiveness of the proposed method. The mean average precision achieved on the NEU-DET dataset is 82.8%; while it reaches 70.3% on GC-10 dataset; and on TCAL, it reaches 82.1%. These results surpass those of current mainstream detection methods. The introduction of FI2Net provides a new approach for surface defect detection in industrial products.
Answers are grounded in this paper and cite their sources.
No takes yet. Share an insight, caveat, or question.
Yonghong Song (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: