研究目的
To propose and evaluate two difference-based target detection methods that utilize both target and background spectral information for improved detection accuracy in hyperspectral images.
研究成果
The proposed DTD-Maha and DTD-KSAM methods demonstrate superior performance in target detection across three real hyperspectral datasets, with DTD-Maha excelling in linear separability scenarios and DTD-KSAM in non-linear cases. The methods effectively utilize both target and background spectral information, offering improved detection accuracy and computational efficiency compared to existing detectors.
研究不足
The study acknowledges the challenge of limited training samples for target and background classes, especially in high-dimensional hyperspectral data. The computational complexity of calculating covariance matrices for DTD-Maha is noted as a potential limitation.
1：Experimental Design and Method Selection:
The study proposes two methods, DTD-Maha and DTD-KSAM, utilizing Mahalanobis distance and kernel-based spectral angle mapper, respectively, for target detection.
2：Sample Selection and Data Sources:
Three real hyperspectral images (San Diego, HyMap, and Indian Pines datasets) are used for experiments.
3：List of Experimental Equipment and Materials:
Hyperspectral imaging sensors (AVIRIS and HyMap) are mentioned for data collection.
4：Experimental Procedures and Operational Workflow:
The methods involve calculating distances of testing pixels to both target and background spectra, applying kernel tricks for non-linear cases, and comparing results with existing detectors.
5：Data Analysis Methods:
Performance is evaluated using ROC curves, AUC values, and McNemar's test for statistical significance.
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
