Predicting corn salad (Valerianella spp.) weight in a plant factory with artificial lighting (PFAL) using K-Nearest Neighbors algorithm (KNN)

The research project is aimed to use computer vision and machine learning techniques to predict the weight of corn salad in a plant factory with artificial lighting.
The goal is to monitor the growth of the plant by measuring its weight remotely.
Physical parameters such as height, number of leaves, plant width, and plant weight of various ages were collected, then we applied machine learning using the K-Nearest Neighbors (KNN) algorithm.
The data set was preprocessed and split into training sets, then the hyperparameters tuned GridSearchCV feature of KNN algorithm was used to yield testing sets.
These testing results were evaluated by their associated mean squared error (MSE) and R-squared (R²) values.
The obtained MSE of the weight data with the value 0.0595 g and below indicates a low error in predicting its weight and the R² value of 0.9695 indicates the 96.95% accuracy in predicting its weight.
After achieving the best performance, a trained KNN model was customized in this study to predict the weight of ten new corn salad plants.
The R² value of the predictions is 0.9787 indicating that this model performs the weight prediction with great accuracy.
In conclusion, our trained KNN model achieved by machine learning of corn salad’s physical parameters demonstrates high accuracy in predicting its weight and this model will become a useful tool in plant-growth monitoring in plant factory.

IV Asian Horticultural Congress – AHC2023

S. Dhammayorddee, C. Chanseetis, A. Kato, M. Takagaki, N. Lookin

machine learning, computer vision, GridSearchCV, hyperparameters

https://doi.org/10.17660/ActaHortic.2024.1404.187