DETECTING STEM AND SHAPE OF PEARS USING FOURIER TRANSFORMATION AND AN ARTIFICIAL NEURAL NETWORK

An algorithm for judging the presence of stems and a method for describing the irregular shapes of Huanghua pears found that good stems can be distinguished from broken stems.

Abstract

Huanghua pear is an important fruit in China. The shape and condition of the stems are important indices forclassifying Huanghua pears. Images of Huanghua pears were acquired with a machine vision system. Using templates withdifferent sizes, an algorithm for judging the presence of stems was developed. Meanwhile, the stem head and the joint pointbetween the stem and the pear body were labeled. After calculating slopes of the approximate tangential lines of the stem atthe head and bottom positions, the included angle of these two lines was obtained. It was found that the included angle ofa broken stem was smaller than that of a good stem. Based on this feature, good stems can be distinguished from broken stems.Results of a test on 53 pear images showed that the accuracy for judging the presence and integrity of the stems reached 100%and 93%, respectively. A method for describing the irregular shapes of Huanghua pears was also studied. Fouriertransformation and Fourier inverse transformation pairs were used as the shape descriptors. The first 16 harmoniccomponents of the Fourier descriptor were found sufficient to represent the primary shapes of Huanghua pears. Thesecomponents were used as the inputs to an artificial neural network (ANN) to classify Huanghua pears. The classificationaccuracy reached 90%.