基于机器视觉和工艺参数的针芽形绿茶外形品质评价

外形是针芽形绿茶的关键感官评价指标,通常依据色泽、条形、嫩度和匀整度等表象特征进行人工评审,难以做到精准、客观和量化评价。本文以自动化生产线机制的针芽形绿茶为研究对象,基于茶叶品质、形成工艺和视觉形态等内外因素,构建了外形品质的智能感官评价方法。首先,在线采集在制品的17个机制工艺参数和成品茶的图像,进行图像特征提取,选取9个颜色特征和6个纹理特征。进而,通过与专家感官评分进行关联分析,明确了与感官品质显著相关的特征变量。为获取高效的评价模型,采用偏最小二乘法(PLS)、极限学习机(ELM)和强预测器集成算法(ELM-Ada Boost)3种多元校正方法,分别建立了基于工艺或图像特征的针芽形绿茶外形感官的量化评价模型。建模结果表明,基于图像特征建立的ELM-Ada Boost模型(Rp=0.892,RPD大于2),其预测性能优于其他模型,且具有更小的RMSEP(0.874)、Bias(-0.148)、SEP(0.226)和CV(0.018)值。同时,非线性模型的预测性能均高于PLS线性模型,能更好地表征工艺参数、图像信息与感官评分之间的解析关系,且建模速度更快(0.014~0.281 s)。而Ada Boost法作为一种混合迭代算法,能进一步提升ELM模型的精度和泛化能力。结果表明,基于机器视觉和工艺评价针芽形绿茶外形品质是可行的,为拓展茶叶感官品质评价方法和专家工艺决策支持系统研制,提供理论依据和数据支撑。

Quality Evaluation for Appearance of Needle Green Tea Based on Machine Vision and Process Parameters

Green tea has the largest consumption in China, and needle-shaped green tea is a typical type of green tea. The appearance of green tea is the key sensory evaluation index of green tea. However, it is hard to realize an accurate, objective and quantitative evaluation of green tea through manual evaluation on the characteristics as the color, stripe, tenderness and uniformity, etc. Based on internal and external factors such as quality forming process and visual morphology of tea, an intelligent sensory evaluation method of the appearance quality of tea was established. Firstly, collecting the process parameters of tea products and image characteristics of made tea, totally 17 process parameters, nine color features and six texture features were selected, conducting correlation analysis with expert sensory evaluation, and screening out remarkably correlated characteristic variables. In order to obtain an efficient evaluation model, based on process parameters and image characteristic parameters respectively, multiple quantitative evaluation models were established for needle-shaped green tea appearance senses by using three multivariate correction methods such as partial least squares (PLS), extreme learning machine (ELM) and strong predictor integration algorithm (ELM-AdaBoost). The comparison of the results showed that the ELM-AdaBoost model based on image characteristics had the best performance (RPD was more than 2). Its predictive performance was superior to other models, with smaller RMSEP (0.874), Bias (-0.148), SEP (0.226), and CV (0.018) values of the prediction set, respectively. Meanwhile, non-linear model had better predictive performance than linear model, which can better represent the analytic relationship between process parameters, image information and sensory scores, and modeling faster (0.014~0.281s). AdaBoost method, which was a hybrid integrated algorithm, can further promote the accuracy and generalization capability of the model. The above conclusions indicated that it was feasible to evaluate the quality of appearance of needle green tea based on machine vision and process. This study provided an effective technical method and idea for developing tea sensory quality evaluation methods, and laid theoretical basis and data supports on the development of expert process strategy supporting systems of tea quality, which had a broad industry prospect in tea processing, trading and refined blend technology.