Authors:
Mateus Coelho Silva
1
;
2
;
Servio Pontes Ribeiro
3
;
Andrea Gomes Campos Bianchi
1
and
Ricardo Augusto Rabelo Oliveira
1
Affiliations:
1
Departamento de Computação, Instituto de Ciências Exatas e Biológicas, Universidade Federal de Ouro Preto, Brazil
;
2
Instituto Federal de Educação, Ciência e Tecnologia de Minas Gerais, Campus Avançado Itabirito, Brazil
;
3
Departamento de Biologia, Instituto de Ciências Exatas e Biológicas, Universidade Federal de Ouro Preto, Brazil
Keyword(s):
Conditional GAN, Leaf Damage Estimation, Leaf Shape Reconstruction, Deep Learning.
Abstract:
Leaf damage estimation is an important research method, metric, and topic regarding both agricultural and ecological studies. The majority of previous studies that approach shape reconstruction work with parametric curves, lacking generality when treating leaves with different shapes. Other appliances try to calculate the damage without estimating the original leaf form. In this work, we propose a procedure to predict the original leaf shape and calculate its defoliation based on a Conditional Generative Adversarial Network (Conditional GAN). We trained and validated the algorithm with a dataset with leaf images from 33 different species. Also, we tested the produced model in another dataset, containing images from leaves from 153 different species. The results indicate that this model is better than the literature, and the solution potentially works with different leaf shapes, even from untrained species.