研究目的
To propose a superpixel-based linear discriminant analysis (SP-LDA) dimension reduction method for hyperspectral image (HSI) classification that exploits spatial and spectral information.
研究成果
The proposed SP-LDA and SSDR methods were tested on two HSI data sets and showed superior classification performances compared to other widely used DR algorithms by exploiting spatial and spectral information.
研究不足
The paper does not explicitly mention limitations.
1:Experimental Design and Method Selection:
The proposed SP-LDA method uses superpixel segmentation to exploit spatial information and combines spectral and spatial dimensions for dimension reduction.
2:Sample Selection and Data Sources:
Two standard hyperspectral datasets, Indian Pines and Pavia University, are used.
3:List of Experimental Equipment and Materials:
Not explicitly mentioned.
4:Experimental Procedures and Operational Workflow:
The HSI is segmented into superpixels using SLIC segmentation, and a projection is sought to minimize superpixel neighborhood interclass distance and maximize distance among the means of samples.
5:Data Analysis Methods:
The performance is evaluated using overall accuracy (OA), average accuracy (AA), and kappa coefficient (κ) of the SVM classification results.
独家科研数据包,助您复现前沿成果,加速创新突破
获取完整内容