Rice Classification Using Three-Step Neural Network Integration

Abstract

This paper proposes a three-step neural network integration. In the first step, two neural networks are trained to produce a truth output and a falsity output. The second step uses cascade generalization to improve the results of the first step. A sequence of pairs of neural networks where the output of lower-order pair and the training data are used to train the next higher-order pair. The truth and falsity output from the first step are used separately in training using this technique. In the third step, two neural networks are trained using the results of the first step, the two results from the second step, and the training data to produce the final classification results. The rice dataset from UC Irvine Machine Learning Repository are used to test the proposed technique. The accuracy of using the three-step technique is better than using other ensemble techniques.