Akademik Veri Yönetim Sistemi
Trends in Food Science and Technology, cilt.164, 2025 (SCI-Expanded)
Background: Food processing is influenced by multiple factors (e.g., material properties and processing parameters), leading to analytical complexity. Conventional methods (experiments or mathematical/mechanistic models) have limitations, such as high costs, high extrapolation errors, and modeling constraints. Machine learning (ML), a data-driven approach, offers strong nonlinear fitting capabilities to integrate multi-factor interactions for process optimization, thereby reducing energy consumption and enhancing product quality and economic viability. Thus, ML demonstrates potential in the food processing field. Scope and approach: Firstly, the advantages and disadvantages of commonly used ML algorithms in food processing were introduced to aid researchers in selecting the most suitable algorithm. Secondly, applications of ML in food detection, drying, and fermentation were summarized and analyzed, as well as existing challenges and corresponding solutions. In addition, forward-looking strategies for ML in the food processing field were proposed. Key findings and conclusions: ML has made notable progress in food processing, covering applications in detection, drying, and fermentation, with algorithms ranging from unsupervised/supervised learning to deep learning. Current studies have shown different limitations, like the overfitting risks from small samples, interpretability limitations of models, and data acquisition difficulties due to the particularity of food processing. To construct high-quality ML models, dataset and algorithm optimization must be tailored to specific processing. The following directions were proposed to promote the future application of ML in the food processing field: developing small-sample algorithms, integrating mechanistic models with ML for physical interpretability, and establishing global big data platforms to drive efficient, intelligent, and sustainable food processing.