【摘要】 为了更好地对没有先验知识的高光谱图像进行地物分析,利用聚类系统快速稳定的优势,通过堆栈自动编码机结合K-means算法搭建1个高光谱图像聚类系统。借用堆栈自动编码机对数据进行非线性降维,基于自编码的再表达输入原理,重构高光谱图像,验证特征数据的可靠性。利用堆栈自动编码机生成的特征数据可以显著地减少聚类算法所需求的计算强度和训练时间。实验结果表明,经过堆栈自动编码机提取特征后的聚类系统,在去掉不参与聚类地物的类别之后,聚类系统的平均精度可以达到95.26%,满足实际使用的精度要求。更多还原

【Abstract】 In order to better analyze the features of hyperspectral images without prior knowledge and take advantage of the fast and stable clustering system, a hyperspectral image clustering system is constructed through the automatic encoder combined with the k-means algorithm. The built-in automatic encoder is used to perform nonlinear dimensionality reduction on the data. Based on the principle of self-encoding re-expression input, the hyperspectral image is reconstructed, and the reliability of feature data is verified. Using the feature data generated by the automatic encoder can significantly reduce the clustering time. The experimental result shows that the clustering system that the features are extracted by the alternate automatic encoding machine, by removing the categories that do not participate in the clustering features, the average accuracy of the clustering system can reach 95.26%, which meets the accuracy requirements of actual use.更多还原