LiFSO-Net: A lightweight feature screening optimization network for complex-scale flat metal defect detection

Hao Zhong, Ling Xiao, Haifeng Wang, Xin Zhang, Chenhui Wan, Youmin Hu, Bo Wu

Abstract: Highlights•Tailored for detecting defects in complex-scale flat metals, LiFSO-Net stands out for its ability to optimize feature screening process and its lightweight design, making it ideal for industrial use with less computational load.•Utilizes deformable convolution to enhance the understanding of context, thus improving the detection of local semantic details during the feature extraction phase.•Implements adaptive feature compression to retain more semantic information during downsampling, ensuring minimal loss of critical features.•Focuses on optimizing high-order semantic features to increase accuracy in defect recognition during the feature interaction stage.•Proven excellence in both recognition accuracy and efficiency on the NEU-DET, APS-DET and GC10-DET datasets, showcasing the method's effectiveness and practicality.

External IDs:dblp:journals/kbs/ZhongXWZWHW24