Macaw palm (Acrocomia aculeata) is a potential raw material for biodiesel production owing to its high oil content. In this study, a model is created that can estimate the oil percentage in Macaw fruit using spectral indices obtained via computer vision. Ninety Macaw fruits are used to perform an experiment. They are categorized into three stages of maturation and counted in terms of weeks after flowering. Images of the fruits are captured using a multispectral camera in the visible spectral region (RGB) and infra-red regions (NIR), which allows spectral indices to be obtained. A model is established using the principal component regression method and used to estimate the oil content based on the colorimetric characteristics of Macaw palm fruits. The selected model is composed of the first two principal components, which indicate an explanatory potency of 91.57% of the variance. Applying this information to the validation data, the model exhibits a determination coefficient of 0.91, mean square error of 10.61%, and standard error of 5.46% in terms of the oil content, which indicates that computer vision is a promising alternative method for estimating oil content.
Read full abstract