Evaluation of black tea appearance quality using a segmentation-based feature extraction method

• Published in: Food bioscience 2024-04, Vol.58, p.103644, Article 103644
• Main Authors: Song, Feihu, Lu, Xiaolong, Lin, Yiqing, Zhou, Qiaoyi, Li, Zhenfeng, Ling, Caijin, Song, Chunfang
• Format: Article
• Language: English
• Subjects: algorithms; Black tea; color; computer vision; gold; Gray-level run-length matrix; image analysis; Image segmentation; K-means clustering algorithm; prediction; Sensory evaluation; sensory properties; texture
• ISSN: 2212-4292; 2212-4306
• DOI: 10.1016/j.fbio.2024.103644

Compared with highly subjective manual sensory quality evaluation, the application of computer vision techniques in black tea appearance quality evaluation helps to establish an objective and efficient black tea quality evaluation system. In this study, Yinghong No. 9 black tea was taken as the research object, and the gold pekoe, color and strips were adopted as the appearance evaluation characteristics for black tea. An image segmentation method based on the improved K-means clustering algorithm was proposed to realize the segmentation of the dark background area, tea area and golden pekoe area. The CIELAB color model was used to extract color features of the tea area. The texture features extracted by GLRLM were applied to evaluate the strips. The RF, SVR and BPNN were selected to construct prediction models for evaluating tea appearance quality. The prediction accuracy and generalization ability of the RF model are superior to those of the SVR model and BP model, with Rp2, RMSEP and RPD values of 0.898, 1.548 and 3.207, respectively. The proposed feature extraction method based on regional segmentation intuitively described the key evaluation characteristics of black tea appearance, and the predicted results were highly consistent with the manual sensory evaluation. •A rapid image segmentation method based on the improved K-means clustering algorithm was proposed.•CIELAB color model and GLRLM were integrated to extract quality-related features from black tea images.•The Random Forest model showed the best performance in evaluating black tea quality.•The proposed method has high accuracy and robustness on local tea sample dataset.