Estimating coffee crop parameters through multispectral imaging and machine learning algorithms

Abstract

Brazil plays a crucial role in the global economy due to its significant contribution to the agricultural sector, particularly in coffee production, where it stands out as the largest producer and exporter of processed coffee. Various disturbances can influence coffee plants, causing abnormalities that can hinder their successful growth. Parameters such as plant height and canopy diameter play an essential role in assessing the health and productivity of the plants, reflecting their growth, development, and ability to capture sunlight. Additionally, height is also related to the balanced distribution of nutrients and water, providing valuable information about overall performance and the capacity for healthy production. In this regard, the application of methodologies involving remote sensing and machine learning algorithms has shown promising results in the rapid and safe acquisition of information about agricultural systems. This study evaluates different machine learning algorithms, using radiometric values from multispectral images obtained by remote sensing platforms as input datasets for estimating plant height and canopy diameter in coffee cultivation. The best performance was observed for architectures that showed lower RMSE and RMSE% values. For the plant height parameter (m), the RGB sensor exhibited the best performance using the Random Tree algorithm, with an RMSE (0.27) and RMSE% (8.80). For the canopy diameter (m), the sensor showed the best performance using the Random Forest algorithm, with an RMSE (0.15) and RMSE% (8.16).