Automated defect classification in the maintenance phase using a channel attention-based convolutional neural network model of natural language processing