Selection of spectral bands for estimating fishmeal quality parameters using ML models

Merino T.; Criollo A.; La Rosa G.; Valdiviezo J.; Soto J. · 2024 IEEE Colombian Conference on Applications of Computational Intelligence, ColCACI 2024 - Proceedings (2024) · DOI: 10.1109/colcaci63187.2024.10666581

Resumen

This paper presents the study of a model based on Machine Learning (ML) algorithms to estimate quality parameters in fishmeal, using hyperspectral imaging spectroscopy (HSI). It is applied feature selection algorithms to find the most important bands in the obtained spectral signatures.For this, we first calibrated the Partial Least Squares (PLS) model by defining the number of principal components to be applied to each parameter. Then, the comparison between the band selection methods using Variable Importance in Projection (VIP) and Backward Variable Elimination (BVE) was performed. Subsequently, the results obtained by each method were compared, of which BVE was the one that obtained the best performance. With the bands selected by this method, PLS models and neural network models were adjusted. Finally, the performance of both models was compared. It is concluded that the BVE-PLS method proved to be effective for the selection of the most important spectral bands for estimating fishmeal quality parameters. The Q2 values of protein, protein, fat and moisture, for which a Q2 value greater than 0.5 is obtained.

Instituciones: Universidad de Piura