With the development of the scientific technology, there has been an explosive growth in the volume of scientific literature. It is a very crucial task for archivists to achieve a fast and accurate classification of scientific literature whether online or offline. As a typical text classification task, the existing efforts mainly focus on learning the semantics containing in the scientific texts while ignoring the key information of the labels designated by human experts. To overcome this issue, in this work, we propose a kind of novel approach for Scientific Text Classification task based on a Dual Contrastive Learning (SCT-DCL), which can fuse the meaning of label candidates into the contexts by concatenating them at the tail of the texts. By the contextual learning using pre-trained language model. The proposed SCT-DCL can learning the joint semantic information including the contexts and all the designated labels. Moreover, based on the duality principle and establishing positive and negative instances in the training batch data, the introduced dual contrastive learning algorithm can help the model to differentiate the similar instances with different labels, which can facilitate the model to achieve a better performance during classifying the scientific literature. The final experimental results demonstrate that the proposed SCT-DCL method can achieve a competitive performance compared with the recent advances published in domain of the scientific text classification task.