Abstract—The basis of object recognition is the feature extraction from the received signals. But the real signals have the three-non characteristics of nonlinear, non-stationary and non-Gaussian properties, which make it difficult to extract features of targets accurately. Hence, Higher-Order Statistics (HOS) and generalized diagonal slices spectra are introduced. Feature vectors E1 and E2 are extracted on the basis of generalized diagonal slices spectra using simple summation and maximization schemes for underwater targets, respectively. Summation and maximization of the horizontal or vertical slices spectra, assigned feature vector names as E3 and E4, respectively, are used as comparison. In order to compare the performance and reduce the dimension of feature vectors based on the generalized diagonal slices spectra and horizontal or vertical slices spectra, feature selection scheme based on Fisher’s class separability is introduced. The classification accuracies of feature vectors E1, E2 E3, and E4 are testified by One-against-One (OAO) method of multi classification of Support Vector Machine (SVM) for different segment numbers of the training sets computing and different largest measure number M. The results show that the total performance of feature vectors based on diagonal slices spectra is better than that of horizontal or vertical slices spectra and that the performance of feature vectors based on summation scheme is better than that of maximization scheme.
Index Terms—feature extraction, feature selection, Fisher’s class separability, generalized diagonal slices
Cite: Haitao Yu, "Classification Performance Comparison of Feature Vectors Based on Summation Scheme and Maximization Scheme," International Journal of Machine Learning and Computing vol. 1, no. 1, pp. 73-78, 2011.