The analysis of EEM have been improving, but its application for in-situ monitoring is hindered by the expensive cost and large size of the equipment itself. In this study, we utilize a miniaturized spectrometer to acquire discrete three-dimensional fluorescence data. By employing an adaptively trained deep generative neural network, we decode the data and restore it into continuous high-resolution data. Compared to benchtop instruments, our approach achieves a significant reduction in volume (2%) and cost (5%), while substantially preserving data quality.