DEEP LEARNING AND PHYSIOLOGICAL FEATURE MODELING FOR PLANT GROWTH ESTIMATION BASED ON THERMAL IMAGING

Abstract

Precision agriculture, climate-resilient farm management, and yield forecasting require proper and timely estimation of crop growth. Traditional growth measurements are typically either labor intensive, destructive or time-ridden. The paper suggests a non-destructive, thermal imaging-based system of quantitative estimation of plant growth using deep learning and modeling of physiological features. The thermal images of the canopies are taken at a high resolution, both under control and field conditions, to derive physiological useful variables, such as the percentage of canopy cover, average canopy temperature, and temperature fluctuations, which indicate plant health, transpiration rates and responses to stress. An overall Growth Index (GI) is obtained by combining these thermal descriptors so as to be able to standardize the crop development stages. Sequential measurements of canopy cover are modelled by logistic growth curve fitting to capture the growth behavior in the time. The growth parameters obtained are biologically interpretable including the maximum growth rate and the inflection point. Simultaneously, a shallow Convolutional Neural Network (CNN) is trained, which applies the Growth Index directly to thermal image density, without the need to employ manual feature engineering and on the basis of that, is able to run quickly and automatically to derive the growth index. Using experimental analysis, there is a good correlation and low estimation error has been shown between the predicted and reference growth measures at the various stages of growth. The suggested solution would be able to combine physiological interpretability and the efficiency of deep learning by providing a rapid, scalable, and dependable solution to real-time monitoring of plant growth. The framework is used to facilitate agronomic decisions based on data, and enhance intelligent crop phenotyping within the precision agriculture systems.