A database for automatic classification of gender in Araucaria angustifolia plants


Forests have been disorderly exploited, and many species are considered endangered. Some initiatives have been taken in order to prevent forests from being destroyed. A good alternative would be to plan a spatial distribution of plants, with higher number of females than males. Determining the gender of seedlings would provide important information for a possible strategy. Another common problem that researchers in this field very often face, in order to perform their experiments, is the lack of a representative database. To overcome this difficulty, we introduce a new database in this work, which is composed of nuclear magnetic resonance of adult Araucaria angustifolia plants. In order to gain better insight into this database, we have tested different strategies and classifiers. A first set of experiments took three classifiers trained to discriminate males from females considering the original database. A second round of experiments applied the genetic algorithm technique to select subsets of attributes based on single-objective and two-objective functions. After analyzing the achieved results, we have also proposed a new strategy based on statistical measures for selecting subsets from the attributes. A comprehensive set of experiments has shown that the proposed selecting strategy has achieved better performances, with an accuracy of 80.3% (AUC = 79.4). We believe that researchers will find this database a useful tool in their work on determining the Araucaria angustifolia gender. On the other hand, the proposed selecting strategy would be useful for reducing the complexity of databases and accelerating the process of building classification models.




MARTINS, Jefferson G. et al. A database for automatic classification of gender in Araucaria angustifolia plants. Soft Computing, Berlin, v. 25, p. 5503-5517, 2021. DOI: 10.1007/s00500-020-05551-x. Disponível em: https://link.springer.com/article/10.1007/s00500-020-05551-x. Acesso em: 29 jan. 2024.