研究目的
To improve computation time and memory requirements in the classification of hyperspectral images using reduced kernel extreme learning machines (RKELM) compared to kernel extreme learning machines (KELM).
研究成果
RKELM provides significant improvements in computation time (approximately 10 times faster than KELM) with a trade-off in accuracy. It is suitable for applications where speed and memory efficiency are critical, despite the slight decrease in performance. Future work could explore optimal subset selection and application to other datasets.
研究不足
The reduced kernel method (RKELM) shows lower accuracy compared to the full kernel method (KELM), with the largest loss of 1.824% when CNi=50. This is expected due to the use of a subset for representation. The study is limited to one dataset and specific parameters (e.g., subset size fixed at 10% of training data).
1：Experimental Design and Method Selection:
The study uses a kernel-based approach with RBF kernel function for classification, specifically comparing RKELM and KELM. The methodology involves mathematical formulations and algorithms for extreme learning machines and kernel computations.
2：Sample Selection and Data Sources:
The Pavia center dataset from Italy is used, which is a hyperspectral image cube of size 1096 × 715 × 102 after preprocessing (removing noisy bands and cropping). It contains classes such as water, trees, grass, brick, soil, asphalt, tar, tile, and shadow.
3：List of Experimental Equipment and Materials:
No specific hardware mentioned; software used is Python 3.6 with scikit-learn module.
4：6 with scikit-learn module.
Experimental Procedures and Operational Workflow:
4. Experimental Procedures and Operational Workflow: Equal number of samples per class (CNi) are selected for training. RKELM uses a subset of training data (10% of total) to compute the kernel matrix, while KELM uses the full set. The classification is performed on all data (labeled and unlabeled), and performance is evaluated on labeled data only. Results include accuracy percentages and computation times.
5：Data Analysis Methods:
Performance is measured by classification accuracy (percentage), and computation time is recorded in seconds. Results are compared using tables and classification maps.
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
