Determination of aflatoxin B1 in wheat based on colourimetric sensor array technology: Optimization of sensor features and model parameters to improve the model generalization performance

Abstract

Wheat is highly susceptible to contamination by fungal toxins during storage and transportation. Among them, aflatoxin B1 (AFB1) is the most common, has the highest degree of pollution and is the most toxic and carcinogenic. Therefore, this study proposed a novel means for the detection of AFB1 content in wheat according to colourimetric sensor array technology. A highly specific colourimetric sensor array was prepared to collect volatile gas information of wheat samples. The firefly algorithm and the sparrow search algorithm (SSA) were applied to optimize the sensor features and the parameters of the back propagation neural network (BPNN), respectively. The results obtained showed that the average correlation coefficient of prediction (RP) of the optimized BPNN model improved from 0.94 to 0.97, the average root mean square error of prediction (RMSEP) decreased from 3.6 to 2.5 and the average relative percent deviation (RPD) increased from 4.3 to 6.2. The results of the study show that it is feasible to achieve AFB1 detection in wheat using a specific colourimetric sensor array. In addition, the optimization of the enter characteristics and parameters of the model is necessary in the model calibration process to enhance the predictive performance and stability of the model while reducing the model complexity to a certain extent.