In this paper, a novel approach to classify the potato tubers based on their moisture content has been proposed using image processing and multilayer perceptron (MLP) neural network. Some experiments were conducted on 300 independent potato samples during three storing stages. Images of 576×768 pixel sizes from potato samples, with three different moisture contents, were captured using a color CCD camera. After preprocessing and segmentation, 84 features were extracted from the acquired images. Sensitivity analysis was used for feature selection process. Results of feature reduction showed that features in color spaces are more important than those in texture and fast Fourier transform (FFT). A process of trial-and-test procedure was carried out to find the optimum topology, the number of hidden layers and the number of neurons in hidden layers, of MLP network. Results obtained from the final ANN models showed encouraging accuracy (more than 96.22%) to apply the approach to develop an expert system for online potato quality monitoring.
