A Study of Broccoli Grading System Based on Machine Vision and Neural Networks

Abstract

Broccoli grading was studied based on computer vision and neural networks. Broccoli images were captured and the five parameters of color and shape (b*, TCD, Hdeg, proportion of yellowness area, roundness) were extracted and calculated from those images by images analysis method such as background purification, color segmentation, gray transform etc. A new grading method was provided based on the results of image analysis. The five parameters used as input vector to establish BP (Back Propagation) neural network for improving prediction precision and another four artificial neural networks (Probabilistic Neural Network Self-Organizing Competition Neural Network, Learning Vector Quantization Neural Network, Self-Organizing Feature Map Neural Network) were also used as classifier in MATLAB7.0. The results showed that all five neural network could be used for broccoli grading with the forecasting accuracy at the range of 68.2~93.4%. The BP neural network was the best network with the forecasting accuracy of 93.4%. The PNN has certain application value since the difference of forecasting accuracy and the validity between PNN and BP network were small, and provided with 1/5 of running time required compares with BP neural network.