To improve the performance of hyperspectral image classification,a dictionary learning algorithm based on kernelized sparse representation was proposed.The presented method can jointly obtain dictionary and classifier by a task-driven kernel function representation.Therefore it can achieve global optimization.Stochastic gradient descent method was used for parame-ters learning in the dictionary model.The model is verified to be differentiable and can obtain optimal solution.Experimental re-sults on several data sets show that the proposed algorithm has good performance in remote sensing image classification.%为提高遥感高光谱影像分类的性能,提出基于核稀疏表示的字典学习算法。通过任务驱动的核函数表示,联合求解字典与分类器,得到遥感高光谱影像分类的最优解。采用随机梯度下降策略求解字典以及学习模型的参数,验证模型是全局可微分的,能通过随机梯度下降求得模型的最优参数。实验结果表明,该算法在多个遥感高光谱图像数据集上均具有较高的分类准确率和召回率。
展开▼