Balanced Loss Function for Accurate Surface Defect Segmentation

The accurate image segmentation of surface defects is challenging for modern convolutional neural networks (CNN)-based models. This paper identifies that loss imbalance a critical problem in accuracy improvement. includes: label imbalance, which impairs the on less represented classes; easy–hard example misleads focus optimization valuable examples; and boundary involves an unusually large value at defect caused by confusion. In this paper, novel balanced function proposed to address problem. includes dynamical class weighting, truncated cross-entropy confusion suppression solve three types respectively. Extensive experiments are performed benchmarks various CNN models comparison with other commonly used functions. outperforms counterparts brings improvement from 5% 30%.