A COMPARISION OF MACHINE VISION SYSTEM AND NEAR-INFRARED REFLECTANCE SPECTROSOPY IN TOMATO INNER-QULITY ANALYSIS

摘要

In order to produce the more information-added production, the grading system and automatically fruits selection system is often used. Fruits grading is carried out using non-destructive inner quality evaluation method such as relative density, imaging sensor, near-infrared spectroscopy (NIR), X-ray, terra-hertz light. However, in many cases, these methods have some problems that they can only measure some specific factor such as sugar content, etc. The inner-quality evaluation of agricultural production could not be decided on only one factor. In this paper, the sugar content, amino acid content, acid content and water content of tomatoes will be measured at the same time by two different methods. The present work compares the machine vision system technology and near-infrared reflectance spectroscopy (NIR) in tomato inner-quality analysis. The machine vision system is effective on measuring the fruit outer-quality, but the NIR is often used for the inner-quality. This paper applied machine vision technology to get the inner-quality of tomato, and make a comparison with the result from NIR. Due to the different fruit color, tomatoes of different periods were considered in the experiment, the growing period and the matured period. 4 kinds of important inner-quality of tomato, sugar content and amino-acid value were measured in this experiment. In NIR, the spectra were taken by inner quality sensor used 710-949nm spectra with 1nm resolution. The calibration model with 710-949nm spectra converted to the second derivative. As a result, sugar content, amino acid content, acid content and water content of the tomatoes were predicted through NIR spectroscopy. On the other side, a stable light box was built due to the color image is sensitive to the light condition in machine vision system. LAB color model and the gray level co-occurrence matrix (GLCM) were used to catch the color feature and texture feature from the color image. The relationship between the feature and the inner quality were analyzed. The result shows that the machine vision technology has a good prediction of acid content; the NIR method has a good prediction in sugar content, amino content and water content.