Large language models (LLMs) have exhibited remarkable capabilities in various natural language processing tasks. Notably, they have significantly enhanced the quality of machine translation. However, these advanced LLMs are designed for generic corpora translation, and their capabilities in academic corpora translation are insufficient. In this study, we propose AcademicMT, a LLM specifically designed for academic English-to-Chinese translation. AcademicMT is developed by applying supervised fine-tuning (SFT) and human preference learning to the deepseek-7B model, and the training data consists of academic papers from multiple academic domains. Additionally, we propose a novel preference optimization method, named as Dynamic Contrastive Preference Optimization (DCPO), to enable LLM to better adapt to the quality gaps in preference data, facilitating more effective learning of human preferences and generating translations of higher quality. We evaluate the proposed AcademicMT and DCPO on the Yiyan dataset, which consists of academic English-to-Chinese corpora. Experimental results demonstrate the superior translation capability of AcademicMT for academic corpora and validate the effectiveness of DCPO. The proposal of AcademicMT marks the first attempt of LLMs in academic translation, and it will serve as an effective tool for researchers.