Tactile-Based Fabric Defect Detection Using Convolutional Neural Network With Attention Mechanism

This article proposes a fabric structure defect detection method based on the vision-based tactile sensor. The result will be robust by using the tactile sensor regardless of dyeing patterns which can influence the result if some other sensors are used, e.g., vision perception. It also reduces the influence of ambient light on defect detection. Therefore, the proposed method can be more robust and universal than conventional visual methods. A robotic arm equipped with the tactile sensors was used to automate and standardize the data collection process and construct fabric datasets. In addition, a convolutional neural network (CNN) integrated with attention mechanism in the channel domain was developed to detect fabric types. The proposed network employed frequency domain filtering to remove or weaken the influence of normal fabric texture information to improve defect detection efficiency and accuracy. Finally, several experiments were conducted to demonstrate the proposed method's superiority to a visual defect detection method for detecting structural defects. In addition, the efficiency of the proposed method is evaluated. Experimental results show that the proposed method is feasible and efficient to meet the real-world detection requirements.