Teses e Dissertações
URI permanente desta comunidadehttps://thoth.dti.ufv.br/handle/123456789/2
Navegar
1 resultados
Resultados da Pesquisa
Item Modeling of coffee ripeness and beverage quality using proximal and remote sensing(Universidade Federal de Viçosa, 2022-06-09) Martins, Rodrigo Nogueira; Pinto, Francisco de Assis de Carvalho; Valente, Domingos Sárvio Magalhães; Queiroz, Daniel Marçal deCoffee is one of the most valuable agricultural commodities, whose price setting and export potential are defined according to its beverage quality. In turn, the beverage quality results from the interaction of different factors from the fruit ripeness degree at harvest to the post-harvest practices. Traditionally, the fruit ripeness is evaluated through manual samplings in the field, whereas the beverage quality is assessed trough sensory analysis by trained tasters. These methods are time-consuming, not representative of the entire production area, and in the case of beverage quality assessment, they are subjective due to the personal influence of the taster. On the other hand, the advent of aerial remote sensing through the Unmanned Aerial Vehicles (UAV), as well as spectroscopy associated with chemometrics and statistical modeling techniques, are presented as a fast and accurate approach for monitoring the spatio-temporal variability of the fruit ripeness and predicting beverage quality. In this sense, the general objective of this thesis, structured in the form of chapters, including the introduction and general conclusions, consisted of modeling the fruit ripeness and beverage quality of Arabica coffee using proximal and remote sensing. In the second chapter, a vegetation index (VI) for monitoring the coffee ripeness using aerial images was developed. For doing that, an experiment was set up in five Arabica coffee fields in Paula Cândido, Minas Gerais state, Brazil. During the coffee ripeness stage in the 2018-2019 season, four flights were carried out to acquire spectral information on the crop canopy using two UAVs, one equipped with a five- band multispectral camera (RGB, RedEdge, and NIR) and another with an RGB camera. For validation purposes, manual counts of the percentage of unripe fruits were performed using irregular sampling grids on each data collection. After image processing, the coffee ripeness index (CRI) and five other VIs (MCARI1, NDVI, NDRE, GNDVI, and GRRI) were obtained. The CRI was developed by combining reflectance from the red band and from a ground-based red target placed on the study area. In general, the CRI showed a higher sensitivity to discriminate between coffee plants ready for harvest from not-ready for harvest regarding the fruit ripeness. Furthermore, the highest R 2 and lowest RMSE values for estimating the coffee ripeness were also presented by the CRI (R 2 : 0.70; 12.42%), whereas the other VIs showed R 2 and RMSE values ranging from 0.22 to 0.67 and from 13.28 to 16.50%, respectively. In the third chapter, two models for the prediction of fruit ripeness using spectral and textural variables were developed and the best variables for the development of spatio-temporal variability maps of fruit ripeness were determined. For that, the fruit ripeness data obtained from six coffee fields (including those described in the second chapter) in the 2018-2019 and 2020-2021 seasons and aerial images of seven flights performed in both seasons were used for data modeling. Through the images, 12 spectral and 64 textural variables composed of bands and VIs were obtained. The performance of the Random Forest algorithm using spectral and textural variables (R²: 0.71 and RMSE: 11.47%) was higher than the model based solely on spectral variables (R²: 0.67 and RMSE: 12.09%). Finally, in both scenarios, the most important variables in the prediction models were the VIs CRI and MCARI1 and the red and NIR bands. Lastly, in the fourth chapter, a method was developed for predicting the coffee beverage quality based on NIR spectroscopy of coffee samples, as well as for classifying the beverage final quality using different variables obtained from the UAV images. Initially, an experiment was set up in the 2020-2021 season in seven coffee fields in the municipalities of Paula Cândido and Araponga. During the harvesting, 13 flights were performed using a UAV equipped with an RGB camera. Then, different spectral, climatic, and terrain variables were obtained from the orthomosaics. For validation purposes, the harvested coffee was processed and subjected to sensory analysis. Next, NIR spectra (1000- 2450 nm) were obtained from 180 samples of roasted and ground coffee. The prediction of the beverage quality attributes based on the NIR spectra was performed using Partial Least Squares (PLS) regression and the combination of PLS with the variable selection algorithm (OPS – Ordered Predictors Selection). Overall, the best predictions were obtained for the aftertaste, overall perception, body, and balance quality attributes using the PLS-OPS models, whose coefficient of correlation (r P ) and the root-mean-square-error of the prediction (RMSE P ) ranged from 0.78 to 0.82 and from 0.15 to 0.13, respectively. In the second analysis, the variables extracted from the UAV images were used as input for developing classification models for the beverage final quality. The results were not satisfactory. Thus, the use of UAV images for beverage quality assessment still needs to be further explored in future studies. Keywords: Digital agriculture. UAV. Coffee fruit ripeness. Sensory analysis. NIR spectroscopy.