Bio-Inspired Multi-Granularity Model for Rice Pests and Diseases Named Entity Recognition in Chinese.

Area of Science:

• Agricultural Science
• Computer Science
• Bioinformatics

Background:

• Rice is a vital staple crop facing significant threats from pests and diseases.
• Effective management of agricultural data is essential for modern farming practices.
• Named entity recognition (NER) is crucial for early detection and control of crop issues.

Purpose of the Study:

• To develop an advanced deep learning model for recognizing rice pests and diseases.
• To address challenges in NER for rice pests and diseases, including structural complexity and nested entities.
• To improve the accuracy and efficiency of agricultural data utilization for crop protection.

Main Methods:

• Utilized BERT for encoding text representations.
• Employed multi-granularity convolutional neural networks (CNNs) to capture nested boundary information.
• Implemented a bidirectional long short-term memory network (BiLSTM) with a conditional random field (CRF) for sequence modeling and labeling.

Main Results:

• The proposed deep learning model effectively extracts multi-granularity features.
• The model demonstrated strong performance in identifying entities related to rice diseases and pests.
• Achieved a high F1 score of 91.74% on a custom dataset, indicating significant accuracy.

Conclusions:

• The developed model offers a robust solution for named entity recognition in the context of rice pests and diseases.
• This approach enhances the potential for precise early prevention and control strategies in agriculture.
• The findings contribute to the advancement of agricultural informatization through sophisticated data analysis techniques.