Abstract Coffee fruit drying is an essential step, as it directly affects the drink’s final quality. The presence of green fruits in this process significantly reduces the quality standard. Conventional characterization techniques are performed manually, are time-consuming, and lead to increased estimation errors. Thus, the objective was to identify green, red, and black fruits in the drying patio by digital classification of multispectral and RGB images. The data were obtained by remotely piloted aircraft during daily flights over coffee fruits for eight days of drying. The analyses involved color segregation techniques, supervised classification (Random Forest), and the application of thirty vegetation indices to RGB and multispectral images, utilizing Python, RStudio, and GIS to generate statistical models and maps, in conjunction with manual fruit estimation and monitoring of temperature and humidity. The use of aerial images proved effective in monitoring colors only during the first three days, due to rapid moisture loss. In all analyses, green fruits showed better spectral separation. Image segmentation proved to be the superior method, adequate up to the third day. Random Forest showed reduced performance due to shadows and color mixing and was only effective on the first day. Among the RGB indices, ExGR and MGRVI stood out, with high potential for detecting green fruits, while TGI indicated two classes of green, highlighting the efficiency of conventional sensors. For multispectral images, NDVI was effective on the first day, and NDWI was effective up to the third day, influenced by fruit of humidity reduction. The need to develop a specific vegetative index to standardize this monitoring is emphasized, based on green separation, considering humidity and robust quality indicators.
Santana et al. (Mon,) studied this question.