From fluorescence imaging to intelligent quantification: An automated and intelligent platform for Al3+ detection in food samples using a machine learning-enhanced Ratiometric probe.

This study presents an intelligent detection platform for Al3+ sensing by integrating a ratiometric fluorescence probe, smartphone imaging and a feedforward neural network (FFNN) algorithm. The probe suspension exhibits distinct fluorescence color transition from orange to green upon Al3+ binding, demonstrating a wide linear range (0.1-100 μM), high selectivity, a low detection limit (11.6 μg/kg), and a rapid response time (30 s). The fluorescence images was processed by a simple MATLAB program to extract digital RGB values, which were then input into a three-layer FFNN for Al3+ concentration prediction within minutes. Trained on an extensive dataset of 720 RGB value sets, the FFNN model demonstrated superior predictive performance with correlation coefficients of 0.9979, 0.9973, 0.9970 for training, validation, and test sets, respectively. The detection system was applied in detecting Al3+ in starch noodles, achieving recovery rates of 98.60%-102.2% and relative standard deviations below 1.190%. And the results obtained align with those generated by standard ICP-MS method. This study presents a user-friendly, rapid, and effective platform for the automated and intelligent sensing of Al3+, enabling the high-throughput analysis of large sample sets under unmanned conditions.