Automatic pixel-level crack segmentation in images using fully convolutional neural network based on residual blocks and pixel local weights.