Near Infrared feature waveband selection for fishmeal quality assessment by frequency adaptive binary differential evolution

Fishmeal is a high-energy feed. Fishmeal quality has an important influence on animal growth. The detection of ash content is essential for evaluating fishmeal quality. Near infrared (NIR) spectroscopy is a rapid detection technique that can be used to determine the ash content in fishmeal. The wavelength related to the target component must be selected in NIR quantitative analysis considering that the NIR spectrum has a strong correlation. In this paper, the adaptive binary differential evolution (ABDE) algorithm was proposed for feature selection. The algorithm was an improvement of the mutation operator of binary differential evolution (BDE) and improved binary differential evolution (IBDE) algorithms. The performance of the BDE, IBDE, and ABDE algorithms was compared in secondary wavelength selection on 11 combined intervals selected by the move window partial least squares (MWPLS) algorithm. The performance of the model established by the ABDE algorithm in feature selection was superior to that constructed by the BDE and IBDE algorithms. Frequency ABDE (FABDE) was designed by introducing frequency into the ABDE algorithm to minimize the influence of random factors. The root mean square error (RMSET) and the correlation coefficient (R-T) of the FABDE model for the test set were 1.160% and 0.941, respectively. By comparison, the RMSET and the RT of the ABDE model were 1.212% and 0.935, respectively. The results of model prediction indicated that the FABDE algorithm proposed herein can be feasibly used to improve the model effects for the rapid NIR analysis of ash content in fishmeal.

Automated summary

This study utilizes NIR spectroscopy for rapid detection of ash content in fishmeal, proposing the adaptive binary differential evolution (ABDE) algorithm for feature selection and improving model performance by introducing frequency into the ABDE algorithm.