COMPUTER VISION-BASED POTATO DEFECT DETECTION USING NEURAL NETWORKS AND SUPPORT VECTOR MACHINE

Payman Moallem, Navid Razmjooy, and Mohsen Ashourian

Keywords

Potato defect detection, support vector machines, multilayer perceptrons, radial basis function, grey level co-occurrence matrix

Abstract

Detection of external defects on potatoes is one of the most important technologies in the realization of automatic potato grading stations. In this paper, a computer vision-based potato defect detection algorithm using artificial neural networks and support vector machine (SVM) is proposed. In this algorithm, to detect the potatoes pixels in background, the supervised colour segmentation based on multilayer perceptrons (MLPs), radial basis function (RBF) neural networks as well as SVM are applied to the RGB component of each pixel. Afterwards, co-occurrence texture features are extracted from the grey level component of colour-space image, and finally three different classifiers including MLP, RBF and SVM are trained and validated to apply for defect detection. Results showed that the SVM classifiers represent a higher performance than the MLP and RBF neural networks for potato defect detection. The computational cost of the proposed SVM-based algorithm shows the possibility of a real-time implementation.

Important Links:

Go Back