基于师生网络协同的动态阈值半监督学习图像分类研究OA

The application of semi-supervised learning in image processing receives wide attention.Among them,the method based on consistency regularization and pseudo-label improves the performance of the model by effectively using a large amount of unlabeled data.However,most of the existing methods use high fixed thresholds to generate pseudo-labels for unlabeled data,which makes it difficult to effectively utilize unlabeled data.In addition,the teacher network may mislead the student network due to performance fluctuations.Aiming at the above problems,this paper proposes a dynamic threshold semi-supervised learning image classification method MTDT.This method only supervises student network training by teacher network when teacher network performance is better than student network.At the same time,according to the learning state of the model,the confidence threshold is dynamically adjusted in an adaptive manner.The experimental results on CIFAR-10,STL-10 and SVHN public datasets show that the proposed method has higher classification accuracy.