TY - JOUR
T1 - Detecting vineyard plants stress in situ using deep learning
AU - Cándido-Mireles, Mayra
AU - Hernández-Gama, Regina
AU - Salas, Joaquín
N1 - Publisher Copyright:
© 2023 Elsevier B.V.
PY - 2023/7
Y1 - 2023/7
N2 - Diseases and nutritional deficiencies have the potential to seriously impact the production yield and proper development of perennial species such as grapevine. The distinction between changes resulting from normal growth stages and plant alterations caused by biotic and abiotic stress is often drawn through visual inspection, where the observer's subjectivity could introduce human errors, despite the presence of experience and technical knowledge. This document presents an assessment of CNNs for detecting plant stress in grapevine RGB images captured in situ, under conditions that could include variations in light, shadows, insects, or the presence of scrubs. We evaluated five architectures for their ability to discriminate plants with stress symptoms in images captured through the annual grapevine cycle in field conditions. The best model exhibited a 97.2% accuracy, 0.996 ROC AUC, and 0.958 AP using the EfficientNetB3 architecture. Our methodology aims to support winegrowers in their decision-making by enhancing the information they collect through traditional visual inspection methods.
AB - Diseases and nutritional deficiencies have the potential to seriously impact the production yield and proper development of perennial species such as grapevine. The distinction between changes resulting from normal growth stages and plant alterations caused by biotic and abiotic stress is often drawn through visual inspection, where the observer's subjectivity could introduce human errors, despite the presence of experience and technical knowledge. This document presents an assessment of CNNs for detecting plant stress in grapevine RGB images captured in situ, under conditions that could include variations in light, shadows, insects, or the presence of scrubs. We evaluated five architectures for their ability to discriminate plants with stress symptoms in images captured through the annual grapevine cycle in field conditions. The best model exhibited a 97.2% accuracy, 0.996 ROC AUC, and 0.958 AP using the EfficientNetB3 architecture. Our methodology aims to support winegrowers in their decision-making by enhancing the information they collect through traditional visual inspection methods.
KW - Convolutional neural network
KW - Grapevine health
KW - Plant stress detection
KW - Transfer learning
UR - http://www.scopus.com/inward/record.url?scp=85158834264&partnerID=8YFLogxK
U2 - 10.1016/j.compag.2023.107837
DO - 10.1016/j.compag.2023.107837
M3 - Artículo
AN - SCOPUS:85158834264
SN - 0168-1699
VL - 210
JO - Computers and Electronics in Agriculture
JF - Computers and Electronics in Agriculture
M1 - 107837
ER -