关键词: 嵌入式蒸烤箱, 多模态数据, 菜谱生成, 深度学习

Abstract: To address the demand for intelligent upgrading of embedded steam ovens and solve the problems of single operation mode and insufficient matching of personalized cooking needs in traditional equipment, an accurate recipe generation system that integrates multi-modal data was constructed, providing support for technological innovation in intelligent kitchen appliances. By collecting multi-modal information such as food ingredient images, weight, user voice commands, and environmental sensor data, combined with preprocessing techniques such as wavelet transform denoising, CLAHE image enhancement, and MaskR-CNN segmentation, an early and late fusion multi-modal data algorithm architecture is adopted to construct a deep learning model based on CNN, LSTM, and MLP, and the model performance is optimized through pruning algorithm. The experiment showed that image noise (accuracy decreased from 91.45% to 68.34% with a standard deviation of 0.1) and speech recognition errors significantly affect recommendation accuracy. After 30% pruning, the computational complexity of the model decreased by 38.41%, and the accuracy loss was controlled within 3%. According to a survey of 100 users, 63.24% of young users prefer voice interaction, while 28.48% of middle-aged and elderly users prefer manual operation. The system has achieved effective application of multi-modal data in recipe generation, providing a technical path for device intelligence. In the future, it is necessary to further optimize cross-modal semantic synthesis and user personalized adaptation capabilities to promote the development of intelligent kitchen appliances.

Key words: Embedded steam oven, Multi-modal data, Recipe generation, Deep learning