Use of Fuzzy Neural Network in Industrial Sorting of Apples

Abstract

In this paper, an automated system and methodology for nondestructive sorting of apples are presented. Different from the traditional manual grading method, the automated, nondestructive sorting equipment can improve the production efficiency and the grading speed and accuracy. Most popular apple quality detection and grading methods use two-dimensional (2D) machine vision detection based on a single charge-coupled device (CCD) camera detect the external quality. Our system integrates a 3D structured laser into an existing 2D sorting system, which provides the addition third dimension to detect the defects in apples by using the curvature of the structured light strips that are acquired from the optical system of the machine. The curvature of the structured light strip will show the defects in the apple surface. Other features such as color, texture, shape, size and 3D information all play key roles in determining the grade of an apple, which can be determined using a series of feature extraction methods. After feature extraction, a method based on principal component analysis (PCA) for data dimensionality reduction is applied to the system. Furthermore, a comprehensive classification method based on fuzzy neural network (FNN), which is a combination of knowledge-based and model-based method, is used in this paper as the classifier. Preliminary experiments are conducted to verity the feasibility and accuracy of the proposed sorting system.