This invention relates to the field of pre-prepared food production and detection technology, specifically a method, equipment, and medium for pre-prepared food production and detection. In this invention, hyperspectral data, three-dimensional depth data, and visual data are simultaneously acquired at the detection nodes of the pre-prepared food
production line. Spectral features, depth features, and appearance features are extracted from these data, and real-time
production line status indicators, including process stage identifiers, ambient temperature values, and
conveyor belt speed feedback values, are simultaneously acquired. Through dual-model
collaborative processing, the first analysis model integrates multiple features and status indicators to output a
foreign object risk level determination result, while the second evaluation model, based on this result and combined with depth features, calculates the conformity of portion specifications. It also combines spectral and appearance features to calculate the food's thermal
processing status indicators, ultimately generating a comprehensive quality status conclusion and
quality control instructions. This invention solves the problem of difficulty in synchronously controlling multi-dimensional quality parameters in dynamic production lines and improves detection robustness through dynamic compensation of environmental parameters.