MiniCrack: A simple but efficient convolutional neural network for pixel-level narrow crack detection

With the advancement of deep learning, the newly proposed neural networks are growing increasingly complicated to achieve great performance. In this context, we propose a simple but effective neural network called MiniCrack for narrow crack detection. We also propose a lightweight version, MiniCrack-Light, to adapt to scenarios with limited computing resources. MiniCrack and MiniCrack-Light outperform the current state-of-the-art neural networks on all three challenging testing data sets with fewer parameters and achieving stronger robustness. PixelShuffle and PixelUnshuffle designed for image super-resolution are successfully used to the field of image segmentation, which effectively alleviates the problems caused by pooling. • Two new networks, MiniCrack and MiniCrack-Light, are proposed for pixel-level narrow crack detection. • PixelShuffle and PixelUnshuffle are introduced to replace traditional methods for down-sampling and up-sampling in CNN. • Switchable normalization is used to solve the poor performance of batch normalization in the case of limited GPU memory. • Sigmoid linear unit is introduced to enhance the weak nonlinear representational power of a non-deep neural network.