Aiming at the problems of traditional named entity recognition methods relying on artificial dictionary and insufficient feature extraction in the process of agricultural pest information extraction, considering the complexity and fuzziness of agricultural text data, an agricultural named entity recognition method based on pre training BERT is proposed. firstly, the unlabeled pre training BERT was used to eliminate ambiguity, then BILSTM was used to capture long-distance dependence, and finally the best sequence annotation was selected through CRF. In addition, combined with the text particularity of agricultural entities, the word itself and partial radicals are selected to establish a joint multi feature PBERT-BILSTM-CRF for entity recognition. experiments show that the PBERT-BILSTM-CRF combined with the combined characteristics of the word itself and the radical has improved the precision, recall and F1 value compared with other models, and its optimal F1 value has reached 90.24%. the model has the characteristics of fast training speed and strong recognition ability. Named entity recognition is the premise of many tasks in the middle and downstream of natural language processing. The model provides a research basis for the construction of knowledge graph in agricultural field and agricultural Q&A.