基于PCA与t-SNE特征降维的城市植被SVM识别方法

Urban vegetation identification method based on pca-t-sne-svm

以高光谱图像降维为研究问题,针对主成分分析法(PCA)投影结果混叠、线性不可分和t-分布式随机邻域嵌入算法(t-SNE)内存占用大、运行时间长等不足,提出了一种基于PCA与t-SNE结合的高光谱图像降维方法。设计了基于SVM的城市植被识别模型,有效地提高了运行速率,进而更好地提取高光谱图像的本质特征,提高了高光谱图像中城市植被的分类精度。实验选取肯尼迪航天中心(KSC)数据为对象,结果表明,PCA-t-SNE-SVM算法总体分类精度可达92.06%,Kappa系数为0.91时,分类效果最优,相较于PCA-SVM和t-SNE-SVM算法,总体分类精度分别提高了13.51%和3.33%,Kappa系数分别提高了0.15和0.04,均表现出良好的性能。

Taking the dimension reduction of hyperspectral image as the research problem,the principle component analysis(PCA)projection result aliasing,linear indivisibility and t-distributed stochastic neighbor embedding(t-SNE)have large memory occupation and long running time.A hyperspectral image dimensionality reduction method based on PCA and t-SNE is proposed,and an urban vegetation recognition model based on SVM is designed to effectively improve the operating rate and better extract the essential features of hyperspectral images.The classification accuracy of urban vegetation in hyperspectral images.The experimental data of Kennedy Space Center(KSC)is selected as the object.The results show that the overall classification accuracy of PCA-t-SNE-SVM algorithm is up to 92.06%,and the Kappa coefficient is 0.91,the classification effect is optimal,compared with PCA-SVM and t-SNE-SVM algorithm,the overall classification accuracy increased by 13.51%and 3.33%,respectively,and the Kappa coefficient increased by 0.15 and 0.04,respectively,showing good performance.