During thermal processing, the generation of polycyclic aromatic hydrocarbons (pahs) is strongly coupled with temperature fluctuations, lipid oxidation, smoke deposition, and volatile release, making it difficult to optimize risk mitigation and flavor retention simultaneously. In this paper, we construct a data-driven food processing control evaluation framework that incorporates multi-source sensing, feature fusion, and process optimization. The risk identification model and flavor quality evaluation model of pahs were established by synchronizing spectral signal, gas sensitivity response, temperature distribution and product color characteristics, and the roasting temperature, heating time, air flow and surface moisture were multi-objective optimized. Experimental results show that the proposed framework achieves 97.8% risk identification accuracy, 0.986 AUC and 94.3% flavor quality classification accuracy. After optimization, the total content of polycyclic aromatic hydrocarbons was reduced by 31.6%, the retention rate of key ideal volatiles was 89.4%, and the sensory acceptance was improved from 8.1 to 8.8. This framework provides a more stable and computable way to intelligently suppress the generation of harmful substances while maintaining product flavor consistency under uncertain processing conditions.
