Abstract: Tea polyphenol, amino acid and moisture contents are important indicators of tea quality. Traditional detection methods have long cycles and complex processes. In this paper, machine vision was used to monitor the color and texture of tea leaves in real time during the microwave fixation process. The moisture content was detected online and the tea polyphenol and amino acid contents were also measured. The results show that the color, texture features and moisture content, tea polyphenol, amino acid contents all showed regular changes during the fixation process and had significant correlations. The principal component analysis was used to analyze the color and texture features and the first 3 principal components were taken to establish extreme learning machine (ELM), genetic neural network (GA-BP), and convolutional neural network (CNN) models to predict the quality. The results show that ELM, GA-BP and CNN models were more suitable for the prediction of moisture, tea polyphenol and amino acid contents, respectively, and their accuracies were all above 0.99. The research results show that it is feasible to predict the moisture, tea polyphenol and amino acid contents during the fixation process by monitoring the color and texture features of tea in real time.

Key words: machine vision, microwave fixation, tea polyphenols, amino acids, moisture content, prediction model