A segmentation algorithm for the automatic recognition of tomato at harvest

Abstract

Tomato is one of the most important products in the world, thus, tendency to grow tomato in greenhouse is increasing. On the other hand, automatic control methods in agricultural and greenhouse operations in order to increase efficiency and reduce labor costs are increasing, too. Since tomato is a plant which fruits do not ripe simultaneously, it is necessary to find a system for machines to distinguish the ripe tomato from unripe. In this study an algorithm was developed for recognizing ripe tomato and keeping unripe tomato for later. For this purpose, the 200 color images of tomato were acquired under natural light condition of greenhouse. Then the color data of ripe and unripe tomato and background were extracted and considered. Finally, observed color spectrum of unripe and ripe tomato were very close to each other and color of ripe tomato was not uniform. To overcome the problems, different color models like RGB, HIS and YIQ were used. Recognition of ripe tomato was performed in two steps, first removing background and second extracting ripe tomato. For removing background, the RGB color space was used. Extracting of ripe tomato was done in three steps: at the beginning, yellow parts of ripe and unripe tomato were extracted by using of the YIQ and HIS color spaces. In the second step, the RGB color space was used to extract red and orange parts of tomato. At the final steps, for recognizing of ripe tomato, two conditional relations in the YIQ color space were defined. Composite of above three steps gave the best results. The algorithm was able to identify the area of a ripe tomato in the image to the extent of 96%. It is very important to extract the full area of ripe tomato, because catching the tomato by robot will be easy. The harvest arm of robot can be opened as the size of tomato.