Identification of procymidone in rapeseed oils based on olfactory visualization technology

Abstract

A new method for qualitative identification of pesticide residues in edible oil based on olfactory visualization technology is proposed. Firstly, according to the pre-experimental experience, 12 suitable chemical dyes were selected to prepare sensor arrays to form an olfactory visualization system. The system was used to gather volatile odor information in rapeseed oil samples that had different levels of procymidone residues. Subsequently, the olfactory information was translated into a visual representation. Then, the visual data, i.e. color features, were preprocessed and underwent principal component analysis to extract and reduce the dimensionality of sensor image information, resulting in the optimal number of principal components (PCs). Finally, three different models based on the sensor image features were constructed for analysis. These models included the support vector machine (SVM), the random forest (RF) and the kernel based extreme learning machine (KELM). In process of model correction, the parameters of the three models were optimized by dung beetle optimizer (DBO) algorithm, resulting in the identification of the optimal parameters. The results show that the cumulative variance contribution increases slowly when the number of PCs starts from 6, and the three models (i.e. DBO-SVM, DBO-RF, and DBO-KELM) have the best prediction results when PCs = 6. Compared to the DBO-RF and DBO-KELM models, the training and prediction performance of the SVM model based on the DBO optimization algorithm is more stable and has the higher recognition accuracy of the samples, reaching 100%. The results show that it is feasible to use olfactory visualization technology to achieve high-precision qualitative identification of edible oil pesticide residues.