Barragán-Hernández Wilson, Mahecha-Ledesma Liliana, Angulo-Arizala Joaquín, Olivera-Angel Martha
Centro de Investigación Turipaná, Corporación Colombiana de Investigación Agropecuaria (AGROSAVIA), Montería 230001, Colombia.
Facultad de Ciencias Agrarias, GRICA research group, Universidad de Antioquia, Medellin 1226, Colombia.
Foods. 2020 Jul 24;9(8):984. doi: 10.3390/foods9080984.
This study was conducted to evaluate the feasibility of using near-infrared spectroscopy (NIRS) to predict beef consumers' perceptions. Photographs of 200 raw steaks were taken, and NIRS data were collected (transmittance and reflectance). The steak photographs were used to conduct a face-to-face survey of 400 beef consumers. Consumers rated beef color, visible fat, and overall appearance, using a 5-point Likert scale (where 1 indicated "Dislike very much" and 5 indicated "Like very much"), which later was simplified in a 3-point Likert scale. Factor analysis and structural equation modeling (SEM) were used to generate a beef consumer index. A partial least square discriminant analysis (PLS-DA) was used to predict beef consumers' perceptions using NIRS data. SEM was used to validate the index, with root mean square errors of approximation ≤0.1 and comparative fit and Tucker-Lewis index values <0.9. PLS-DA results for the 5-point Likert scale showed low prediction (accuracy < 42%). A simplified 3-point Likert scale improved discrimination (accuracy between 52% and 55%). The PLS-DA model for purchasing decisions showed acceptable prediction results, particularly for transmittance NIRS (accuracy of 76%). Anticipating beef consumers' willingness to purchase could allow the beef industry to improve products so that they meet consumers' preferences.
本研究旨在评估使用近红外光谱法（NIRS）预测牛肉消费者认知的可行性。拍摄了200份生牛排的照片，并收集了NIRS数据（透过率和反射率）。利用牛排照片对400名牛肉消费者进行了面对面调查。消费者使用5点李克特量表（1表示“非常不喜欢”，5表示“非常喜欢”）对牛肉颜色、可见脂肪和整体外观进行评分，该量表后来简化为3点李克特量表。采用因子分析和结构方程模型（SEM）生成牛肉消费者指数。使用偏最小二乘判别分析（PLS-DA），利用NIRS数据预测牛肉消费者的认知。使用SEM对该指数进行验证，近似均方根误差≤0.1，比较拟合度和塔克-刘易斯指数值<0.9。5点李克特量表的PLS-DA结果显示预测能力较低（准确率<42%）。简化的3点李克特量表提高了判别能力（准确率在52%至55%之间）。购买决策的PLS-DA模型显示出可接受的预测结果，尤其是对于透过率NIRS（准确率为76%）。预测牛肉消费者的购买意愿可以使牛肉行业改进产品，以满足消费者的偏好。
