Aiming at the task of named entity recognition in the field of aviation unsafe events, the aviation safety information weekly report was used as the data source to analyze and construct the named entity recognition dataset and domain dictionary of aviation unsafe events. In order to solve the problem of poor performance of the traditional named entity recognition models in capturing domain entity boundaries, based on the bidirectional encoder representations from transformers (BERT) pre-trained language model, a method for enhancing domain semantic information by integrating domain dictionary embedding was proposed. Several comparative experiments were carried out on self-built datasets. The results show that the proposed method can further improve the recognition rate of entity boundaries. Compared with the traditional bi-directional long short term memory-conditional random field (BiLSTM-CRF) named entity recognition model, the performance is improved by about 5%.