Abstract To improve on the robustness of traditional machine learning approaches, emphasis has recently shifted to the integration of such methods with Deep Learning techniques. However, the classification problems, complexity and inconsistency in several spectral classifiers developed for hyperspectral images are some reasons warranting further research. This study investigates the application of Deep Support Vector Machine (DSVM) for hyperspectral image classification. Two hyperspectral images, Indian Pines and University of Pavia are used as tentative test beds for the experiment. The DSVM is implemented with four kernel functions: Exponential Radial Basis Function (ERBF), Gaussian Radial Basis Function (GRBF), neural and polynomial. Stand-alone Support Vector Machines form the interconnecting weights of the entire network. The network is trained with one hundred input datasets, and the interconnecting weights of the network are initialised using the regularisation parameter of the model. Numerical results show that the classification accuracies of the DSVM for Indian Pines and University of Pavia based on each DSVM kernel functions are: ERBF (98.87%, 98.16%), GRBF (98.90%, 98.47%), neural (98.41%, 97.27%), and polynomial (99.24%, 98.79%). By comparing the DSVM algorithm against well-known classifiers, Support Vector Machine (SVM), Deep Neural Network (DNN), Gaussian Mixture Model (GMM), K Nearest Neighbour (KNN), and K Means (KM) classifiers, the mean classification accuracies for Indian Pines and University of Pavia are: DSVM (98.86%, 98.17%), SVM (76.03%, 73.52%), DNN (94.45%, 93.79%), GMM (76.82%, 78.35%), KNN (76.87%, 78.80%), and KM (21.65%, 18.18%). These results indicate that the DSVM outperformed the other classification algorithms. The high accuracy obtained with the DSVM validates its efficacy as state-of-the-art algorithm for hyperspectral image classification.

中文翻译：

---

用于高光谱图像分类的深度支持向量机

摘要 为了提高传统机器学习方法的鲁棒性，最近的重点已转移到此类方法与深度学习技术的集成上。然而，为高光谱图像开发的几种光谱分类器的分类问题、复杂性和不一致是值得进一步研究的一些原因。本研究调查深度支持向量机 (DSVM) 在高光谱图像分类中的应用。两个高光谱图像，印度松树和帕维亚大学被用作实验的试验台。DSVM 使用四个核函数实现：指数径向基函数 (ERBF)、高斯径向基函数 (GRBF)、神经和多项式。独立的支持向量机构成了整个网络的互连权重。该网络使用一百个输入数据集进行训练，并使用模型的正则化参数初始化网络的互连权重。数值结果表明，基于每个DSVM核函数的印度松树和帕维亚大学DSVM的分类精度为：ERBF（98.87%，98.16%），GRBF（98.90%，98.47%），神经（98.41%，97.27%） ) 和多项式 (99.24%, 98.79%)。通过将 DSVM 算法与众所周知的分类器、支持向量机 (SVM)、深度神经网络 (DNN)、高斯混合模型 (GMM)、K 最近邻 (KNN) 和 K 均值 (KM) 分类器进行比较，平均分类印度松树和帕维亚大学的准确率是：DSVM (98.86%, 98.17%), SVM (76.03%, 73.52%), DNN (94.45%, 93.79%), GMM (76.82%, 78.35%), KNN% (76.87%) , 78.80%) 和 KM (21.65%, 18.18%)。这些结果表明 DSVM 优于其他分类算法。DSVM 获得的高精度验证了其作为高光谱图像分类的最新算法的有效性。