Abstract: Starch is the most important plant-derived natural polysaccharide, and precise analysis of its complex multi-scale structures and efficient regulation of its functional properties are key challenges in starch research. Conventional analytical methods are constrained by low efficiency, limited throughput, and difficulty in quantifying the intricate “structure-function” relationships. Machine learning (ML), particularly deep learning (DL), offers revolutionary tools to address these challenges by leveraging its powerful capabilities in data-driven modeling and automatic feature extraction. This review systematically summarizes the advances in ML applications across the entire spectrum of starch research. It focuses on the critical roles of ML in the fundamental characterization of starch, including the rapid and non-destructive detection of starch composition and structure (covering spectral quantification and intelligent image analysis) as well as the analysis of its gelatinization process, and in the rational design of starch-based functional materials (covering formulation optimization, performance prediction, and generation of mechanistic hypotheses). The review aims to provide theoretical support for the intelligent and precise development of starch research.

Key words: starch; composition detection; design of functional materials; machine learning; deep learning