结合深度信念网络与支持向量机的地表分类算法

Surface Classification Algorithm Based on Depth Belief Network and Support Vector Machine

由于支持向量机(Support Vector Machine,SVM)在高维度、非线性等情况下仍具有极高精确性故而被广泛应用于地形识别领域研究。在复杂的地形环境以及数据的不平衡等环境下,SVM可能会因为缺少较强的鲁棒性导致分类结果并不理想。论文以提高复杂地形环境下分类算法精确度为目的,在研究了深度信念网络(Deep Belief Network,DBN)[1]与支持向量机(Support Vector Machine,SVM)的基本理论并进行有效结合后应用于地形识别领域。算法大致为将地形图片通过初始构建的深度信念网络结构对训练集进行训练进而优化重构网络结构,并通过测试集验证网络结构的有效性。在OUTEX数据集上的实验结果表明该算法对比地形分类算法中的SVM、GEPSVM等算法有更高的分类精确性。

Support vector machine(SVM)is widely used in terrain recognition because of its high accuracy under high dimen⁃sional and nonlinear conditions.In complex terrain environments and unbalanced data environments,SVM may lack strong robust⁃ness and result in unsatisfactory classification results.In order to improve the accuracy of classification algorithm in complex terrain environment,this paper studies the basic theory of deep belief network(DBN)and support vector machine(SVM)and then applies it to the terrain recognition field.The algorithm is as follows.The training set of the terrain image is trained through the depth belief network structure initially constructed,then the network structure is optimized and reconstructed,and the effectiveness of the net⁃work structure through the test set is verified.Experimental results on OUTEX data set show that this algorithm has higher classifica⁃tion accuracy than SVM and GEPSVM.