Foreign Fiber Classification Algorithm Based on Continuous Unsupervised Domain Adaptation

Aiming at the foreign fiber classification task of the same type in different fields in the cotton spinning industry scene, a foreign fiber classification method based on continuous unsupervised domain adaptation (Cotton Adaptation-Net, CAN for short) is proposed. The algorithm applies continuous learning combined with domain adaptation method to the cotton spinning industry scene, so that the model can continuously learn the data characteristics of the source domain and improve the generalization performance in the target domain. At the same time, the classification performance of the model is enhanced by increasing the attention mechanism and optimizing the convolution method. Experimental data show that the average accuracy rate of the algorithm reaches 82.46%, and the network predicts each picture time is about 0.52ms; compared with the three classic domain adaptation methods, the average accuracy rate is increased by 12.21 %, respectively, 4.96% and 3.25%. Therefore, the algorithm meets the requirements of foreign fiber classification tasks in different fields of the same type in industrial scenarios.