TY - GEN
T1 - CoffeeSE
T2 - 8th Annual International Conference on Information Management and Big Data, SIMBig 2021
AU - Incahuanaco-Quispe, Filomen
AU - Hinojosa-Cardenas, Edward
AU - Pilares-Figueroa, Denis A A.
AU - Beltrán-Castañón, Cesar A.
N1 - Publisher Copyright:
© 2022, The Author(s), under exclusive license to Springer Nature Switzerland AG.
PY - 2022
Y1 - 2022
N2 - Coffee is one of the most important agricultural products and consumed beverages in the world. Then, adequate control of the diseases is necessary to guarantee its production. Coffee rust is a relevant coffee disease, which is caused by the fungus hemileia vastatrix. Recently, deep learning techniques have been used to identify coffee diseases and the severity of each disease. In this paper, we propose a new interpretable transfer learning method to estimate the severity of coffee rust called CoffeeSE. The proposed method consists of four stages: Leaf segmentation, patch sampling, patch-based classification, and quantification/interpretation analysis. On the classification stage, a Brazilian dataset is used to transfer by fine-tuning new weights to a pre-trained classifier. So, this new classifier is tested in Peruvian coffee leaves infected with coffee rust. Our approach shows acceptable quantification results according to an expert agronomist. In addition, an interpretability module of the patch-classifier is proposed to provide a visual and textual explanation of the most relevant pixels used in the classification process.
AB - Coffee is one of the most important agricultural products and consumed beverages in the world. Then, adequate control of the diseases is necessary to guarantee its production. Coffee rust is a relevant coffee disease, which is caused by the fungus hemileia vastatrix. Recently, deep learning techniques have been used to identify coffee diseases and the severity of each disease. In this paper, we propose a new interpretable transfer learning method to estimate the severity of coffee rust called CoffeeSE. The proposed method consists of four stages: Leaf segmentation, patch sampling, patch-based classification, and quantification/interpretation analysis. On the classification stage, a Brazilian dataset is used to transfer by fine-tuning new weights to a pre-trained classifier. So, this new classifier is tested in Peruvian coffee leaves infected with coffee rust. Our approach shows acceptable quantification results according to an expert agronomist. In addition, an interpretability module of the patch-classifier is proposed to provide a visual and textual explanation of the most relevant pixels used in the classification process.
KW - Coffee rust
KW - Interpretability
KW - Sampling
KW - Transfer learning
UR - http://www.scopus.com/inward/record.url?scp=85128892330&partnerID=8YFLogxK
U2 - 10.1007/978-3-031-04447-2_23
DO - 10.1007/978-3-031-04447-2_23
M3 - Conference contribution
AN - SCOPUS:85128892330
SN - 9783031044465
T3 - Communications in Computer and Information Science
SP - 340
EP - 355
BT - Information Management and Big Data - 8th Annual International Conference, SIMBig 2021, Proceedings
A2 - Lossio-Ventura, Juan Antonio
A2 - Valverde-Rebaza, Jorge
A2 - Díaz, Eduardo
A2 - Muñante, Denisse
A2 - Gavidia-Calderon, Carlos
A2 - Valejo, Alan Demétrius
A2 - Alatrista-Salas, Hugo
PB - Springer Science and Business Media Deutschland GmbH
Y2 - 1 December 2021 through 3 December 2021
ER -