Tea category identification based on optimal wavelet entropy and weighted k-Nearest Neighbors algorithm

Abstract

Tea category classification is of vital importance to industrial applications. We developed a tea-category identification system based on machine learning and computer vision with the aim of classifying different tea types automatically and accurately. 75 photos of three categories of tea were obtained with 3-CCD digital camera, they are green, black, and oolong. After preprocessing, we obtained 7 coefficient subbands using 2-level wavelet transform, and extracted the entropies from the coefficient subbands as the features. Finally, a weighted k-Nearest Neighbors algorithm was trained for the classification. The experiment results over 5 × 5-fold cross validation showed that the proposed approach achieved sensitivities of 95.2 %, 90.4 %, and 98.4 %, for green, oolong, and black tea, respectively. We obtained an overall accuracy of 94.7 %. The average time to identify a new image was merely 0.0491 s. Our method is accurate and efficient in identifying tea categories.