A Novel Method Based on Multi-Molecular Infrared (MM-IR) AlexNet for Rapid Detection of Trace Harmful Substances in Flour

Due to the severe hazard of traces of harmful substances in flour and its massive production, pretreatment-free, rapid, and accurate determination of trace harmful substances was a practical and urgent demand in modern daily life. In this study, multi-molecular infrared (MM-IR), consisting of one-dimensional infrared (1D-IR), synchronous and asynchronous two-trace two-dimensional correlation infrared (2T-2DCOS-IR), as well as their RGB IR-image, was developed and employed to directly identify flour with diverse harmful substances. The present findings indicated that 2T-2DCOS analysis based on asynchronous pictures was better than others at identifying target substances and shedding new light on previous MM-IR-related studies. Systematical comparison of developed MM-IR-AlexNet flour models with conventional machine learning algorithms indicated that asynchronous MM-IR-AlexNet (2T-2DCOS-IR-AlexNet) had the best performance with 99.6% accuracy. Therefore, it was demonstrated that deep learning-based MM-IR could be a robust method for the direct detection of trace harmful substances in power foods.