关键词: 燕麦, 面絮图像, 深度学习, 识别模型, 理化特性分析

Abstract:

【Objective】This study aimed to classify the stages of oat dough mixing, analyze the mechanisms by which oat dough flocs regulate product quality at each stage, and develop a model to identify these stages, so as to provide a theoretical basis and technical support for enhancing the automation of oat flour product processing.【Method】Oat dough flocs images during oat flour mixing were used as the dataset. Morphological information was first extracted from these images to classify the stages of oat dough mixing. Subsequently, the regulatory mechanisms of dough floc properties on product quality at each stage were elucidated, focusing on gelatinization degree, amylose content, intermolecular forces, textural properties, rheological properties and moisture distribution of dough flocs. Finally, a simple and efficient prediction model for oat dough mixing stages was developed by combining a convolutional neural network (ResNet-50) with a support vector machine (SVM).【Result】Based on changes in the image shadow area combined with cluster analysis, the oat dough mixing process was divided into four distinct stages: water absorption and adhesion, agglomeration into a mass, dynamic equilibrium, and rupture followed by dispersion. Three key trends were observed during the mixing process: first, the gelatinization degree of oat dough flocs gradually increased and stabilized at the dynamic equilibrium stage, while the amylose content decreased gradually and stabilized at the same stage; second, the effects of disulfide bonds, hydrogen bonds, ionic bonds, and hydrophobic interactions progressively enhanced; third, textural analysis showed that from the water absorption and adhesion stage to the dynamic equilibrium stage, the hardness, chewiness, and elasticity of oat dough flocs all increased and reached their maximum values, whereas hardness began to decrease at the rupture and dispersion stage. Additionally, the K value of rheological properties showed an upward trend from the water absorption and adhesion stage to the dynamic equilibrium stage, indicating improved dough strength and stability of the oat dough flocs. Low-field nuclear magnetic resonance (LF-NMR) analysis further revealed that moisture migrated from the free state to the bound state (A_22-1, and A_22-2) and finally reached a stable state at the dynamic equilibrium stage. Notably, the ResNet-50-SVM-based prediction model for oat dough floc image stages achieved a recognition accuracy of 90%.【Conclusion】The oat dough mixing process could be divided into four stages based on the shadow area of dough floc images, with significant variations in dough floc quality across these stages. Specifically, at the dynamic equilibrium stage, particle uniformity and processability of oat dough flocs were optimized, making this stage the ideal processing window for oat noodle production. The established model enabled reliable identification and classification of dough mixing stages, providing the methodological and technical support for the automated processing of oat noodle products.

Key words: oat, dough flocs image analysis, deep learning, recognition model, physicochemical properties analysis