Research on the Restoration of Rainy Substation Meter Images Based on Semi-Supervised Learning

With the rapid development of smart grids, the task of restoring substation meter images under rainy conditions has become a significant research direction for intelligent monitoring in power systems. Addressing this research challenge, this paper proposes a semi-supervised and unsupervised learning, enhancing the model's learning capability on unlabeled data through a collaborative optimization framework. This paper introduces a high-quality pseudo-label selection mechanism based on multiscale feature extraction and dynamic quality scoring, which accurately filters out high-quality pseudo-labels. Tmage deraining method based on the Mean-Teacher framework, named SemiRain. Semi-Rain leverages the strengths of botho mitigate the issues of overfitting and confirmation bias commonly associated with the traditional L1 loss function, a multi-dimensional contrastive regularization method is introduced. This method improves the model's robustness through cross-dimensional contrastive learning. Experimental results demonstrate that the proposed model achieves outstanding performance in the task of deraining rainy substation meter images.