MINet: Multiscale Interactive Network for Real-Time Salient Object Detection of Strip Steel Surface Defects

The automated surface defect detection is a fundamental task in industrial\nproduction, and the existing saliencybased works overcome the challenging\nscenes and give promising detection results. However, the cutting-edge efforts\noften suffer from large parameter size, heavy computational cost, and slow\ninference speed, which heavily limits the practical applications. To this end,\nwe devise a multi-scale interactive (MI) module, which employs depthwise\nconvolution (DWConv) and pointwise convolution (PWConv) to independently\nextract and interactively fuse features of different scales, respectively.\nParticularly, the MI module can provide satisfactory characterization for\ndefect regions with fewer parameters. Embarking on this module, we propose a\nlightweight Multi-scale Interactive Network (MINet) to conduct real-time\nsalient object detection of strip steel surface defects. Comprehensive\nexperimental results on SD-Saliency-900 dataset, which contains three kinds of\nstrip steel surface defect detection images (i.e., inclusion, patches, and\nscratches), demonstrate that the proposed MINet presents comparable detection\naccuracy with the state-of-the-art methods while running at a GPU speed of\n721FPS and a CPU speed of 6.3FPS for 368*368 images with only 0.28M parameters.\nThe code is available at https://github.com/Kunye-Shen/MINet.\n