Evaluating Tomato Ripeness Using a Neural Network.

Abstract

A method for evaluating tomato ripeness, utilizing its surface color, was developed using a machine vision system with color image processing capability and a multi layered neural network-based software system. The tomato ripeness was classified into four categories; unripe, half ripe, full ripe and over ripe according to the standard commercial classification for manual sorting. Over ripe means the fruit has lost its freshness. Three color specification values, i. e., lightness L*, chroma C * and hue H were calculated from the RGB gray levels of a captured color digital image of a tomato by an on-line image processing system. Only 0.2 to 0.5 % of total surface area of a fruit is needed for color image sensing of the classification. The area size representing 0.5% of the total area was covered by 243 pixels of resolution. A three-layered neural network with four hidden layer units gave a satisfactory performance at 18000 times of BP (Back Propagation) learning. The total processing time from the image capturing to the final output for a single fruit was 0.45 seconds. The recognition rate for the ripeness classification using this method was as high as 93 %. A recognition rate of only 77 % was obtained by the multiple regression model tested. The present work provides another example to strengthen the area of neural network application research on machine vision systems including agricultural robotics, postharvest and processing systems.