黄花梨果形的机器视觉识别方法研究

Shape Identification of Huanghua Pear Using Machine Vision

黄花梨的果形是分级的重要特征之一。利用机器视觉采集黄花梨图像，研究了不规则果品的形状描述方法，提出在黄花梨的分级过程中采用傅立叶变换与傅立叶反变换对来描述果形，开发了基于人工神经网络的果形识别软件。研究发现该傅立叶描述子的前１６个谐波的变化特性足以代表梨体的主要形状，采用傅立叶描述子与人工神经网络相结合的方法进行果形识别的精确率可达９０％。而且只要有合适的训练对。

The shape of Huanghua pear is one of the most important features in classification. Images of Huanghua pears were acquired by means of machine vision system. The method to describe the shape of irregular fruit was studied, in which the Fourier transform and Fourier inverse transform were applied. A sort of software for fruit shape identification based on artificial neural network was developed. It was concluded that the first sixteen harmonics of the Fourier descriptor were enough to represent the primary shape of pear, and the shape identification accuracy could reach 90 % by applying the Fourier descriptor in combination with artificial neural network. Furthermore, this method can also be used to identify the shape of other fruits as the suitable training set is found.