Pattern recognition of fruit shape based on the concept of chaos and neural networks

Abstract

Shape evaluation of fruit is quite empirical and uncertain. In this study, a new technique is proposed to evaluate the fruit shape quantitatively using attractor, fractal dimension and neural networks. The shape of a fruit is usually described from its own profile information. A one-dimensional profile data consisting of radii between the centroid of the fruit and sampling points on the fruit profile was made to characterize the fruit shape. It includes six profiles of the fruit in different directions. The irregularities of one-dimensional profile data in several types of fruits were quantitatively measured by introducing the concepts of attractor and fractal dimension. A three-layer neural network was also used for identifying and tracing the one-dimensional profile data and then evaluating their irregularities. The relationships among identification errors, the shapes of attractors and fractal dimensions for several types of fruits were investigated. Significant correlations were observed in their relationships. The results showed that the uses of the attractor, the fractal dimension and the neural network allowed the complexity of the fruit shape to be quantitatively evaluated.