Classification of three varieties of peach fruit using artificial neural network assisted with image processing techniques.

Peaches are rich in a variety of vitamins and minerals such as carbohydrates, organic acids, pigments, phenolics, vitamins, volatiles, antioxidants and little amounts of proteins and lipids. Iran was Seventh country of the peach producers in the world in 2010. Quality is one of the important factors in marketing of agricultural products. Grading machines have great importance in the quality inspection systems. Most of the current grading machines operate based on machine vision systems to detect blemishes and defects of products, where one image or more are taken for each individual object and the results of processing will decide the quality of the object. Grading and sorting of agricultural products using machine vision in conjunction with pattern recognition techniques, including neural networks, offers many advantages over the conventional optical or mechanical sorting devices. This paper aims to introduce a system that is using machine vision algorithms and Neural Network classifier to classify three varieties of peach fruit. Three cultivars, namely, Anjiri peach cultivar and Shalil Nectarine cultivar, varieties of Iran and Elberta peach cultivar variety of United states were randomly handpicked in two stage of growth, immature and mature on 30 July 2011 and 30 agues 2011 from an orchard located at the Miandoab, west Azerbaijan, Iran, and for each peach cultivar and stage of growth 45 fruits were randomly selected from picked peaches. Image processing technology in the agricultural research has made significant development. An image-capturing system was designed to provide an enclosed and uniform light illumination and to obtain standard images from the samples. The images were sent via a USB capture device to a computer provided with image acquisition and processing toolboxes of MATLAB software (Version R2011a, The Math Works Inc., MA, USA) to visualize, acquire and process the images directly from the computer. Some qualitative information is extracted from the objects to be analyzed in the images. This information was used as inputs to the algorithms for classifying the objects into different categories. In this study feature vector that consider as network input consist of 12 components of color spaces and three components of shape features. After network was trained, confusion matrices for mature and immature fruits were obtained. Total classification accuracy was 98.5% and 99.3% for mature and immature fruits respectively.