EEDN: An Efficient Edge Detection Network for Aeroengine Blade Defect Segmentation

A fast and accurate defect segmentation model is crucial for aeroengine blade inspection. However, most of the defects on the surface of aeroengine blades are micron-sized, which makes the defect features easily lost. In addition, the scales of defects on the surface of aeroengine blades vary widely. In this paper, we propose an Efficient Edge Detection Network (EEDN) for Aeroengine Blade Defect Segmentation. We design a low-complexity network backbone to learn feature information using depthwise separable convolutions efficiently. Moreover, in the decoding structure, we propose a Multiscale Feature Enhanced Attention (MFEA) to improve the multiscale expressiveness of the network and capture the long-range channel information. Experiments have proved that EEDN has achieved 0.859ODS and 97FPS on our dataset, which is superior to existing models.