Feature Selection of the Spectral Signature of Cocoa Bean Based on Visible and Near-Infrared Spectroscopy for Cadmium Estimation

Cocoa production in Peru has grown significantly in the last decade. Peru is the eighth largest producer of cocoa beans and the second largest producer of organic cocoa and fine aroma cocoa in the world. However, the presence of high cadmium content restricts access to international markets. The European Union has established a limit of 0.8 ppm for the presence of heavy metals due to their harmful effects on health. To improve the cocoa bean production process and establish adequate and non-destructive cocoa quality control, this article seeks to estimate the percentage of cadmium in the cocoa bean using its spectral signature. This signature includes the wavelength of 400 nm to 900 nm, covering the visible and near-infrared spectrum. 233 samples from the department of Huánuco, located in the north center of Peru, and the partial least squares (PLS) method with feature selection algorithms is used to identify the wavelengths that most contribute to the estimation of cadmium in cocoa beans. This selection of wavelengths has allowed us to improve the precision of the cadmium estimation, reaching an R2 of 75.67% and an average error of 0.19 ppm with the test data. The selected wavelengths can be considered to design an automatic system that can be implemented in real conditions.