MFNet: A Novel Multilevel Feature Fusion Network With Multibranch Structure for Surface Defect Detection

Surface defect detection is an essential topic in the industrial inspection field. Many methods based on machine vision (MV) have been applied. However, it is still a challenging task due to the complexity of defects, including low-contrast, small objects, and irregular geometric boundaries. To deal with these problems, this article proposes a novel multilevel feature fusion network (MFNet) with a multibranch structure for surface defect detection. First, we extract low- and high-level features via the encoder based on ResNet34. Second, an improved atrous spatial pyramid pooling (ASPP) module is adapted to expand the receptive field (RF) of low-level features. Then, the decoder adopts a multibranch structure to fuse multilevel features for details, and a global attention module is introduced to strengthen the effectiveness of feature fusion and detection accuracy. Finally, the optimal result from multiple outputs can be obtained by multibranch. Extensive experiments indicate that our method enjoys a better defect detection performance compared to four excellent semantic segmentation networks. Especially the accuracy metric can be improved to 98.54%, 99.82%, and 99.79% on three representative defect datasets: CrackForest, Kolektor, and RSDDs.