Machine recognition of citrus variety based on the fractal dimensions of perimeter-area

Citrus fruit variety recognition is an important issue for automated operations including diseases and insect pests prevention and cure, fertilization management and fruit picking. In order to evaluate the feasibility of automatic recognition of various fruits, samples of the citrus unshiu Marc.cv. unbergii Nakai, Skaggs Bonanza Navel orange and Luxi seedless Ponkan were studied. Images of calyx surfaces and the profile were acquired from sample fruits. Pixel numbers of fruit image contour and region were used as the perimeters and areas of fruits, and fractal dimensions of fruits were obtained by the perimeter-area method. Perimeters, areas and fractal dimensions were taken as the character values of three varieties of citrus fruits. A wavelet neural network model was presented to recognize different type of fruits based on these character values. The results showed that the correctnesses of the citrus unshiu Marc.cv. unbergii Nakai, Skaggs Bonanza Navel orange and Luxi seedless Ponkan were 95%, 95%, 97.5%, respectively. From the results we conclude that these three cultivars of citrus fruits can be automatically recognized and have a high correctness with three character values.