Prediction of Cultivation Period and Canopy Area in Lettuce Using Multi-Temporal Visible RGB-Based Vegetation Indices and Computational Intelligence

  • Ronnie Concepcion II, Elmer Dadios, Argel Bandala, Edwin Sybingco

Abstract

Crop growth monitoring is a manifestation of precision cultivation that demands efficient nondestructive computational phenotyping. Vegetation index (VI) plays important role in addressing the issue of vision-based crop phenotyping as the color transformation it exhibits correspond to pattern of phytomorphological descriptors of crops by enhancing vegetation properties. This study deals with predicting the cultivation period in terms of weeks after germination (WAG) and photosynthetic canopy area in mm2 based on extracted RGB-based vegetation indices from digital imagery. In this paper, computational phenotyping was employed through combined machine learning and deep learning models. The morphotype used in this study is loose-leaf lettuce and the employed complete crop life cycle is ten weeks from germination to harvesting that all happens inside a close environment microclimatic chamber with aquaponics as the cultivation technology. Multi-temporal approach of image collection was performed by capturing 30 sample lettuces every week for ten consecutive weeks using digital camera that is oriented directly downward over 12 inches stand. 15 RGB-based VIs were extracted from each image and subjected to multidimensional reduction to avoid overfitting using hybrid neighborhood component analysis (NCA), principal component analysis (PCA) and classification tree (CT). ResNet101, InceptionV3 and MobileNetV2 deep learning models, and Naïve Bayes (NB), linear discriminant analysis (LDA) and K-nearest neighbors (KNN) machine learning models were experimented to predict cultivation period. Bayesian regression neural network, regression tree and ensemble regression machine learning models were used to predict canopy area using selected RGB-based indices, namely normalized difference index (NDI), color index of vegetation extraction  (CIVE),  excess green minus excess red index  (ExGR), vegetative index (VEG), combined indices 1 (COM1) and green minus blue index (GBI). The optimized models showed that ResNet101 with RGB color space yield the best results in cultivation period prediction with accuracy of 86.04% and coefficient of determination of 0.9211. The regression tree model with combination of NDI, CIVE, ExGR, VEG, COM1 and GBI vegetation indices in predicting canopy area resulted to a percentage difference of 0.48% and coefficient of determination of 0.8178. Thus, the developed model is highly practical and efficient for visible RGB-based imagery in crop phenotyping.

Keywords–computational intelligence, computer vision, lettuce, plant phenotype, vegetation index

Published
2020-06-11
How to Cite
Ronnie Concepcion II, Elmer Dadios, Argel Bandala, Edwin Sybingco. (2020). Prediction of Cultivation Period and Canopy Area in Lettuce Using Multi-Temporal Visible RGB-Based Vegetation Indices and Computational Intelligence. International Journal of Advanced Science and Technology, 29(06), 6899 - 6921. Retrieved from https://sersc.org/journals/index.php/IJAST/article/view/22436