End-to-end sequence labeling via deep learning for automatic extraction of agricultural regulations
| Autor: | Ramón Piedrafita Moreno, F. Javier Zarazaga-Soria, Borja Espejo-García, Javier Lacasta, Francisco J. Lopez-Pellicer |
|---|---|
| Rok vydání: | 2019 |
| Předmět: |
0106 biological sciences
Feature engineering Decision support system Computer science Horticulture Machine learning computer.software_genre 01 natural sciences Sequence labeling Information system media_common.cataloged_instance European union media_common Artificial neural network business.industry Deep learning Forestry 04 agricultural and veterinary sciences Computer Science Applications Softmax function 040103 agronomy & agriculture 0401 agriculture forestry and fisheries Artificial intelligence business Agronomy and Crop Science computer 010606 plant biology & botany |
| Zdroj: | Computers and Electronics in Agriculture. 162:106-111 |
| ISSN: | 0168-1699 |
| Popis: | In the European Union, production standards in the form of legal regulations play an important role in farming. Because of the increasing amount of regulations, it is desirable to transform human-oriented regulations into a set of computer-oriented rules to provide decision support through the Farm Management Information System. To obtain the logical structure of rules, automatically labeling their meaningful information is necessary. In this work, we evaluate the performance of 8 different state-of-the-art deep learning architectures to develop an end-to-end sequence labeler for phytosanitary regulations. This sequence labeler extracts different meaningful information items to determine which pesticides can be applied to a crop, the place of the treatment, when it can be applied, and the maximum number of applications. The architectures evaluated do not require feature engineering and, hence, they are applicable to the agricultural regulations of different countries. The best system is a neural network that uses character embeddings, Bidirectional Long short-term memory and Softmax. It achieves a performance of 88.3% F1 score. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|
Full Text from ScienceDirect