Proximal spectroscopy sensing for sugarcane quality prediction and spatial variability mapping

Sensors for predicting attributes related to the quality of agricultural products have been evaluated and implemented from production lines in industrial sectors to some initiatives in the field of agricultural production. Field applications seek to provide spatial information along the field related to the quality of the harvested product. In constant evolution, and with advanced applications in the industrial sector, the near infrared spectroscopy (NIR) presents itself as the best alternative due to the precision of the equipment, fast analysis, non-destructive, easy operation, low cost and sustainable. In addition, the trend towards miniaturization of the equipment has enabled greater flexibility of applications. The production of thematic maps of product quality, associated to yield data, represents an advance for the practice of management with precision agriculture techniques for understanding spatial variability, cause and effect relationships during the crop cycle, and rational agronomic management of production inputs. NIR sensors have been used in initiatives to understand the spatial variability of grape, grain, and forage quality. However, applications for sugarcane quality monitoring are still incipient. The first study reported in this document (Chapter 2) focused on assessing the spatial variability of sugarcane quality attributes in a commercial field, from samples manually collected in the field and processed for measurement by NIR spectroscopy in defibrated form. In addition, the maps produced were evaluated against maps produced from results obtained by conventional laboratory analysis methods. The second study (Chapter 3) sought to evaluate the potential for predicting sugarcane quality parameters with NIR spectroscopy at different levels of sample preparation: no preparation (stalks), with measurements in different sections, defibrated cane, and raw juice. In addition, we sought to achieve variability in the calibration models as a function of climatic variation with sample collections performed in different periods throughout a harvest. In this step, the experiment was performed in a quality laboratory of a mill, so that NIR and conventional analyses could be performed simultaneously. The calibration and prediction models for both studies were developed by multivariate analysis, with partial least squares regressions (PLSR), and the importance of spectral bands in the prediction of organic compounds was evaluated based on what has been reported in the literature. The third and last study (Chapter 4) was conducted with a on-board micro spectrometer in the elevator of a sugarcane harvester to collect real-time information in three areas of a commercial crop. For this, a measurement platform was built with external light sources. During the harvest, samples were collected directly from the machine, after being read by the sensor, for model calibration and validation. In addition, sub-samples of defibrated cane were taken from the samples processed for analysis by conventional methods, for bench measurement, similar to the first study. These spectra were used to build calibration transfer models to fit the spectra collected in real time at the harvester. Calibration models were then built, validated, and used to estimate sugarcane quality attributes collected in real time. At the end of the harvest, soil samples were collected to evaluate cause and effect relationships with the estimated quality data. The proposed method allowed the construction of variogram models with spatial dependence and the spatialization of sugarcane quality data obtained by measurements with the on-board sensor. Moreover, the cause and effect relationships corroborated the estimated results with previously reported by other studies, by presenting a relationship between quality parameters and soil physical attributes. The results of this research constitute a new stage in the direction of research to make it feasible to obtain spatialized data of sugarcane quality by means of on-board sensors. (AU)