Nondestructive monitoring for crop fresh weight and leaf area based on a hanging scale and crop images

Crop fresh weight and leaf area are often selected because the factors are directly related with vegetative growth and carbon assimilation.
Several methods to measure them have been introduced, but the measurements are not easily applicable.
Therefore, a nondestructive measurement with a high versatility is required.
The objective of this study was to propose a nondestructive monitoring system for crop fresh weight and leaf area of a trellised crop.
The data were collected in a greenhouse growing sweet peppers (Capsicum annuum var. annuum). The target growth factors were crop fresh weight and leaf area.
The crop fresh weight was estimated with a total weight and the VWC using simple arithmetic.
The leaf area was estimated based on top-view images using convolutional neural network (ConvNet). The estimated fresh weight showed average R²=0.70, and the estimated leaf area showed R²=0.95, which showed acceptable accuracies.
Transfer learning was not effective in this study.
Simple calculation could avoid overfitting with less limitations such as preset coefficients and data collecting time than the previous study.
ConvNet could relate the raw images with the leaf area without additional sensors and features.
Since the simple calculation and ConvNet adequately estimated the target growth factors, the monitoring system could be used for the data collection in practice with its versatility.
Therefore, the monitoring system could be widely applied to the diverse data analysis with additional tests.

XXXI International Horticultural Congress (IHC2022): International Symposium on Innovative Technologies and Production Strategies for Sustainable Controlled Environment Horticulture

T. Moon, D. Kim, S. Kwon, J.E. Son

artificial intelligence, deep learning, machine learning, plant environment, precision agriculture

https://doi.org/10.17660/ActaHortic.2023.1377.17