Joint-Conditional Mutual Information based Feature Subset Selection for Remotely Sensed Hyperspectral Image Classification

• Item type: SIG Technical Reports(1)
• 公開日: 2022-06-20
• タイトル: Joint-Conditional Mutual Information based Feature Subset Selection for Remotely Sensed Hyperspectral Image Classification
• タイトル / 言語: en
• タイトル: Joint-Conditional Mutual Information based Feature Subset Selection for Remotely Sensed Hyperspectral Image Classification
• 言語: eng
• 資源タイプ / 資源タイプ識別子: http://purl.org/coar/resource_type/c_18gh
• 資源タイプ: technical report
• 著者所属: Degree Programs in Systems and Information Engineering,Graduate School of Science and Technology, University of Tsukuba
• 著者所属: Faculty of Engineering, Information and Systems, University of Tsukuba
• 著者所属(英): en
• 著者所属(英): Degree Programs in Systems and Information Engineering,Graduate School of Science and Technology, University of Tsukuba
• 著者所属(英): en
• 著者所属(英): Faculty of Engineering, Information and Systems, University of Tsukuba
• 著者名: U., A. Md Ehsan Ali; Keisuke, Kameyama
• 著者名(英): U., A. Md Ehsan Ali; Keisuke, Kameyama
• 論文抄録 / 内容記述タイプ: Other
• 論文抄録 / 内容記述: Hundreds of contiguous bands of remotely sensed hyperspectral image (HSI) capture the spectral signatures of observed objects or materials on the earth’s surface. Although the HSI data is able to provide huge information with great details, it poses challenges to image analysis because of the high computational cost due to the large dimensionality of the feature space, redundancy in information, and curse of dimensionality. To overcome these difficulties, feature reduction techniques are used to extract informative features from hyperspectral images. This paper proposes an information-theoretic feature selection approach for selecting an informative feature subset considering the maximum of the minimum approach. The minimum of conditional mutual information based estimation is used to select a feature among the selected feature subset. This selected feature with the corresponding candidate feature is then exploited using the mutual information and joint mutual information based valuation to find the maximum relevance of the candidate features with the target classes. The effectiveness of the proposed approach, called joint-conditional mutual information for selecting informative feature (JCIF), is assessed by implementing it in some synthetic data and two remotely sensed HSI data. Several known feature selection algorithms are also used for comparison purposes. The results of the experiments with K-Nearest Neighbors and Support Vector Machine classifiers reveal that JCIF performs better in selecting informative features.
• 論文抄録(英) / 内容記述タイプ: Other
• 論文抄録(英) / 内容記述: Hundreds of contiguous bands of remotely sensed hyperspectral image (HSI) capture the spectral signatures of observed objects or materials on the earth’s surface. Although the HSI data is able to provide huge information with great details, it poses challenges to image analysis because of the high computational cost due to the large dimensionality of the feature space, redundancy in information, and curse of dimensionality. To overcome these difficulties, feature reduction techniques are used to extract informative features from hyperspectral images. This paper proposes an information-theoretic feature selection approach for selecting an informative feature subset considering the maximum of the minimum approach. The minimum of conditional mutual information based estimation is used to select a feature among the selected feature subset. This selected feature with the corresponding candidate feature is then exploited using the mutual information and joint mutual information based valuation to find the maximum relevance of the candidate features with the target classes. The effectiveness of the proposed approach, called joint-conditional mutual information for selecting informative feature (JCIF), is assessed by implementing it in some synthetic data and two remotely sensed HSI data. Several known feature selection algorithms are also used for comparison purposes. The results of the experiments with K-Nearest Neighbors and Support Vector Machine classifiers reveal that JCIF performs better in selecting informative features
• 書誌レコードID / 収録物識別子タイプ: NCID
• 書誌レコードID / 収録物識別子: AN10505667
• 書誌情報: 研究報告数理モデル化と問題解決（MPS）; 巻 2022-MPS-138, 号 22, p. 1-8, 発行日 2022-06-20
• ISSN / 収録物識別子タイプ: ISSN
• ISSN / 収録物識別子: 2188-8833
• Notice: SIG Technical Reports are nonrefereed and hence may later appear in any journals, conferences, symposia, etc.
• 出版者 / 言語: ja
• 出版者: 情報処理学会