研究目的
To address the misclassification of mixed pixels in hyperspectral images by developing a semisupervised classifier that leverages both global spectral and local spatial geometric structures.
研究成果
The proposed GLLR-SVM method effectively reduces misclassification of mixed pixels by integrating spatial and spectral information, achieving higher accuracy than counterparts on real datasets with limited labeled data. Future work will focus on reducing computational complexity and improving scalability.
研究不足
1. High memory requirements for large datasets. 2. Additional parameters need experimental determination. 3. High computational complexity due to iterative optimization.
1：Experimental Design and Method Selection:
The method involves constructing a GLapLRR graph to capture spectral-spatial affinities, integrating it into SVM for classification. It uses low-rank representation and spatial geometric regularizers.
2：Sample Selection and Data Sources:
Three real hyperspectral datasets are used: AVIRIS Indian Pines, ROSIS Pavia University, and AVIRIS Salinas. Labeled and unlabeled instances are selected randomly, with varying numbers per class.
3：List of Experimental Equipment and Materials:
Hyperspectral sensors (AVIRIS, ROSIS), MATLAB software for implementation, and standard computing hardware (Intel Core i5 CPU, 16GB RAM).
4：Experimental Procedures and Operational Workflow:
Steps include data preprocessing (removing water absorption bands), GLapLRR computation via optimization algorithms, graph construction, and SVM training with Laplacian regularization. Performance is evaluated using overall accuracy, average accuracy, and kappa coefficient.
5：Data Analysis Methods:
Statistical analysis of classification accuracies over multiple trials, comparison with other methods (e.g., KNNG, LLEG, L1G, SSCG, LRRG), and parameter tuning via cross-validation.
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
