Abstract: Food quality serves as the cornerstone for ensuring food safety and enhancing consumer satisfaction, playing a crucial role in advancing the sustainable development of the food industry. Machine learning revolutionizes food quality assessment through efficient data processing and precise predictive modeling, enabling scientific quality management. Specifically, machine vision, employing image perception and analysis, presents distinct advantages in detecting food attributes such as color, shape, and texture, enhancing the objectivity and precision of food quality evaluations. Near-infrared spectroscopy technology utilizes the absorption characteristics of substances to infrared light, achieving non-destructive perception of composition and structure as well as in-depth analysis of spectral features. Furthermore, multi-source data fusion integrates multi-dimensional sensory data, overcoming the limitations of single-modal characterization, optimizing production processes, and improving detection accuracy. This approach provides a strong guarantee of food quality and safety. This paper systematically reviews the current applications of machine vision, near-infrared spectroscopy, and their combined use in multi-source data fusion, driven by machine learning. It further explores advancements in characterization detection and ingredient analysis, identifies current challenges, and provides insights into future research directions to advance food quality detection and evaluation.