Rapid Assessment of Olive Oil Adulteration Using LIF Spectroscopy and a Comparative Study of Machine Learning Models.

<title>Abstract</title> Laser-induced fluorescence (LIF) provides a rapid, non-destructive tool for detecting adulteration in olive oil. However, severe spectral overlap remains a major obstacle, hindering both qualitative identification and quantitative determination of adulteration levels. In this study, a 405 nm diode laser source was used to excite pure extra virgin olive oil (EVOO), soybean oil, peanut oil, and corn oil, as well as binary blends of these three vegetable oils in EVOO, and a total of 1,140 sets of fluorescence spectral data were obtained. Convolutional neural network (CNN), long short-term memory network (LSTM), and improved deep convolutional neural network (AlexNet) were respectively employed for the detection and quantitative analysis of adulteration in olive oil. The models achieved 100% classification accuracy, robustly differentiating pure EVOO from adulterated oil, which confirms their complete reliability for detecting olive oil adulteration. In terms of quantitative prediction, AlexNet performs better than LSTM and CNN, with the coefficient of determination (R²) of 0.9930 and root-mean-square error (RMSE) of 0.1258%. Combining LIF technology with the AlexNet deep learning model enables rapid detection of olive oil adulteration while allowing for precise quantification of adulteration levels, thereby offering an innovative approach to food quality monitoring.