Multidimensional characterization of vegetable breads using multisource intelligent sensory technologies and feature-level data fusion.

Abstract

Background: Food sensory quality is multidimensional, and a single intelligent sensory technique cannot simultaneously characterize color, taste, and odor across multiple dimensions. Multisource intelligent sensory analysis combined with data fusion provides an approach for the objective characterization of multidimensional food sensory characteristics.

Methods: Control bread (CK) and four vegetable breads prepared with pumpkin (PU), Pueraria thomsonii (PG), red pepper (RP), and celery (CE) were analyzed using an electronic eye, an electronic tongue, and an ultrafast gas chromatography electronic nose. These instruments provided color, taste-response, and volatile-compound information, respectively. Principal component analysis (PCA), differential-peak screening, and tentative assignment based on dual-column retention-index (RI) matching were used to compare characteristics among formulations. Principal components from each single-modality dataset were then retained for feature-level data fusion followed by PCA.

Results: The vegetable juices produced distinct color, taste-response, and volatile-compound characteristics in the breads. The first three electronic-eye principal components explained 91.95% of the variance. PU was dominated by strong yellow and vivid yellow, RP showed a clear orange-yellow appearance, and CE retained a green hue. The first two electronic-tongue principal components explained 90.03% of the variance. PG showed relatively high umami and bitterness response indices, whereas CE showed a relatively high saltiness response index. The first five principal components of the electronic-nose data explained 82.89% of the variance, and 17 volatile compounds were tentatively assigned. PG showed high responses for several fermentation-related volatiles. A peak tentatively assigned to acetone was detected only in RP under the current analytical conditions, whereas peaks tentatively assigned to myrcene and β-phellandrene showed pronounced responses in CE. After feature-level data fusion, the first three principal components explained 80.78% of the variance. Loading analysis showed that multiple latent variables from the electronic eye, electronic tongue, and electronic nose jointly contributed to the multidimensional characterization of the formulations.

Conclusion: Breads prepared with different vegetable juices exhibited distinct characteristics in terms of color, taste-response profiles, and volatile-compound profiles. Feature-level fusion of multisource intelligent sensory data enables the joint representation of information from these three instrumental dimensions within a unified analytical framework, providing a methodological framework for the multidimensional characterization of vegetable breads.