Orange Sorting by Applying Pattern Recognition on Colour Image

Manual sorting/grading of oranges is done at wholesale markets/ food processing factories based upon its maturity, size, quality and breeds. With an aim to replace the manual sorting system, this paper proposes the research work for automated grading of Oranges using pattern recognition techniques applied on a single color image of the fruit. This research is carried out on 160 Orange fruits collected from varied geographical locations in Vidarbha Region of Maharashtra. System designed can automatically classify an Orange fruit from this region, given its single color image of 640 × 480 pixel resolution, taken inside a special box designed with 430 lux intensity light inside it, by a digital camera. Only 4 features are used to classify oranges into 4 different classes according to the maturity level and 3 different classes as per size of oranges. In this paper two novel techniques based on Pattern Recognition are proposed – Edited Multi Seed Nearest Neighbor Technique and Linear Regression based technique; although Nearest Neighbor Prototype technique is also deployed. Linear Regression based technique can explicitly predict the maturity of the unknown orange fruit, enabling classification into multiple classes with desired lifespan. Experimental results indicate success rate up to 90% and 98%.