Terrain Recognition Based On Sparse Description Multi-plane Support Vector Machine In Unstructured Environment

Terrain recognition has always been a hot and difficult topic in image processing and pattern recognition.In order to solve the problems of feature extraction,feature selection and identification in unstructured Environments,this paper studies and puts forward own views and solutions.The main work and innovation points include:(l)for the characteristics of unstructured terrain recognition,this paper discusses the difficulties that need to be solved.Meanwhile,it introduces the algorithms and principles of sparse coding,subspace spectrum clustering,manifold learning and support vector machines in detail.(2)This paper proposes a feature sparse extraction method using low-rank and sparse decomposition and group sparse with spatial information structure.In the first place,we uses low-rank and sparse decomposition to deal with local features matrix of images,and obtain low-rank part and sparse part.After decomposition,the subspace spectral clustering is used to form two dictionaries on low-rank part and sparse part,which could solve the problem of the undiscriminating boundary in the unstructured terrain.Based on the study of the strong and weak discriminant information in the terrain image,we study the feature coding method via acquired dictionaries.The focus of this study is how to group the feature via group sparse,and introduce feature selection mechanism to select the least and optimal information in the process of unstructured terrain image sparseness.Then,we build the pyramid of extracted features,which can enhance the global spatial information of the feature.Finally,the features is put into Least Square Twin Support Vector Machine for terrain identification.The reliability of the terrain recognition algorithm is verified in the data set.(3)This paper proposes a kind of spectral clustering based on Grassmannian Manifold to group sparse grouping algorithm,which solves the dictionary learning problem from the low rank matrix and sparse matrix,due to the high dimensions of image and complex target structures affect the effect of dictionary learning.By introducing manifold learning,we redefine the distance between the sample points.In sparse subspace,a spectral clustering algorithm based on Grassmannian Manifold is used to study the dictionary.Verify the reliability of the proposed algorithm on the data set.(4)On the basis of former chapters,when using the twin support vector machine to select the features,this paper proposes a kind of cosine distance of feature pair which is achieved in the procedure of clustering in the Manifold space,to weight the feature.In the Manifold space,we weight the feature according to Manifold neighborhood of the type of the feature,which improves the feature selection ability of twin support vector machine based on external penalty.In this paper,we combine the proposed unstructured terrain recognition framework and twin support vector machine based on external penalty.Verify the reliability of the proposed algorithm on the data set.