UACE2017 Proceedings: An Underwater Acoustic Target Feature Selection Based on Sparse Spectral Regression

• Session:
  Towards Automatic Target Recognition. Detection, Classification and Modelling
• Paper:
  An Underwater Acoustic Target Feature Selection Based on Sparse Spectral Regression
• Author(s):
  Yue Pan, Huangfu Li, Xiaoliang Zhang, Zhang Li
• Abstract:
  In order to remove the redundant features and noise features to improve the accuracy of underwater target recognition, we propose a feature selection method based on sparse spectral regression. The proposed algorithm first constructs a graph using the original high-dimensional feature data, and then the high-dimensional feature space is transformed into a low-dimensional feature space by sparse spectral regression. The low-dimensional feature space can still preserve the attribute of the original data. All features in this low-dimensional space are then ranked and sorted by their classification ability. The top N features construct the optimal feature subset. The classification ability of the optimal feature subset is evaluated by a SVM classifier. A series of comparative recognition experiments were conducted to test the performance of the proposed method using real radiated noise of underwater acoustic targets. The experimental results show that the accuracy of SVM classifier trained by optimal feature subset is higher than original feature set, while the number of selected features decreasing to 21% of original feature number. This means that the proposed algorithm can eliminate redundant features and noise features, and improve the recognition accuracy. \n \nKeywords: underwater acoustic targets classification, sparse spectral regression, feature selection
•   Download the full paper

Contact details

• Contact person:
  Dr Yue Pan
• e-mail:
  
• Affiliation:
  National Key Laboratory of Science and Technology on Underwater Acoustic Antagonizing, Systems Engineering Research Institute
• Country:
  China