Quality assessment of black tea based on physical parameters using machine vision

Abstract

Tea is a valuable cash crop throughout the world. It is a major export product of India. As far as social aspect is concerned, about 1.2 million people are directly employed as labour in tea industry. This constitutes a large proportion of human resource of the country. Quality of tea plays a significant role in its marketability as international export price of tea is fixed according to its quality. At present, tea quality is validated by professional ‘Tea Tasters’ who charge exorbitantly for every sip they take. Conventionally, these experts evaluate tea quality by use of organoleptic methods during fermentation and sorting stage. In addition to this, gas chromatography and colorimetery are employed for chemical analysis of tea liquor and for colour analysis, respectively, at various stages of tea processing. These conventional methods have many shortcomings. First of all, being small in number, the Tea Tasters are difficult to hire and there is every possibility of formation of a cartel by them. Their evaluation methods are subjective and suffer from high labour costs, inconsistency and variability. The prominent physical parameters that establish tea quality include colour, texture, grain shape and size. The approach of Tea Tasters does not quantify these parameters and hence, it is difficult to correlate various parameters of tea for assessment of tea quality. Increasing competition and concerns about tea quality leading to rejection of export orders has resulted in substantial fall in tea export from India in the recent past and consequently, tea industry of India is slowly dying. If proper measures are not adopted and lessons not learnt from the past, situation may aggravate in future. There is a dire need to carry out research in this field so as to meet requirements of global standards. There has been lack of research specifically related to grading and quality assessment of tea all over the world. The above issues are aptly addressed by machine vision based techniques. This work documents the efforts carried out for objective grade assessment of tea quality at the post processing stage with the application of machine vision techniques. In addition to estimation of colour, shape, size and texture by machine vision, direct measurement of two prominent physical parameters namely, moisture and density has been carried out for quality assessment. The present work was taken up to carry out research in this field which has very high socio-economic significance not only for India but for all tea exporting South Asian countries. The main issues required to be addressed by this work include: • Determination of size, shape, texture, etc. of granules of tea non-destructively by machine vision for assessment of tea grade. • Determination of colour of brewed tea liquor for assessment of grade. • Measurement of moisture and density of various grades of tea for grade discrimination. • Development of a classifier followed by statistical validation of results. The problems addressed through machine vision technique have certain definite steps to be followed in sequence with image acquisition followed by image pre-processing, feature extraction. Finally, extracted features are classified. Image acquisition is greatly affected by factors such as selection of camera, viewing distance, orientation of illumination source etc. Due care has be taken at this stage to ensure efficient capturing of image data with a high degree of fidelity. Another critical aspect is feature extraction which involves identification and estimation of suitable features that describe the data uniquely. Towards the end, the classification stage deals with selection of appropriate classifier for classification of feature data. In the present work, as a first step, an objective discrimination amongst the various tea grades of tea was carried out on the basis of their morphological features viz. area, perimeter and aspect ratio. The results were compared with the standard samples obtained from tea industry which were duly graded by tea tasters. Finally, with the extracted features when presented to the three inputs MLP, a grading accuracy of 100% was achieved. Statistical analysis by ANOVA (Analysis of Variance) highlighted area and perimeter as key attributes for discrimination between various grades. Further, the possibility of discriminating various grades of tea granules on the basis of their texture was explored. In the present work, four diverse grades of black tea are discriminated using between using textural features on the basis of spatial location of their grey shade intensities. Certain statistical attributes like energy, entropy, contrast, correlation and homogeneity are evaluated for the image database comprising of images of diverse grades. When these features are classified using MLP, an accuracy of 87.5% was achieved. Further, upon decomposition into sub-band images by DWT, the same features were computed and an improved accuracy of 100% was observed. In the next stage, colour estimation was carried out for discriminating the different grades on the basis of colour of tea liquor. Grade assignment was done on the basis of extracted colour features using the RGB colour model and 100% accuracy was observed using MLP classifier.

Another prominent parameter that determines the shelf life and storage quality of black tea i.e. moisture has been investigated for different grades of tea. It has been observed that the moisture retention is more in the grades having larger granules than the grades having smaller granule sizes. Finally, compacted and un-compacted densities were measured for various tea grades and it has been observed that the density enjoys an inverse relation with the granule size. It is worth mentioning here that the procedures carried out in the present work for quality assessment, except colour analysis, are predominantly non-invasive in nature. If a system is developed using the proposed concept, it is expected that it can successfully assist the traditional methods in the tea industries for quality assessment and monitoring.