Research On Object Recognition Algorithm Based On Sparse Dictionary Learning

With the rapid development of the field of artificial intelligence,the research on object recognition technology has become a hotspot for more and more people.The most critical part of object recognition technology is feature extraction.Traditional feature extraction algorithms generally use artificial design methods and cannot be accurate.How to optimize the feature extraction algorithm to obtain more accurate and complete object feature information becomes the main research direction in the field of object recognition.In order to effectively complete the feature extraction and classification of object recognition,this thesis applies a sparse dictionary learning algorithm to the feature extraction process of the object.The main work of this thesis is as follows:Firstly,several feature extraction algorithms commonly used in the field of object recognition are analyzed and introduced.The implementation principles of various algorithms are analyzed and the corresponding experimental results are given.Commonly used dictionary learning algorithms in the field of object recognition are described.The feature vectors obtained through dictionary learning are respectively tested using support vector machines,K nearest neighbors,and random forests.Secondly,for the problem that the artificially designed feature extraction algorithm in the traditional two-dimensional view-based object recognition algorithm cannot describe the feature information of the object well,this thesis uses the combination of feature learning method and preliminary feature extraction algorithm to realize the characteristics of the object image Extraction,an object recognition algorithm based on sparse dictionary learning is proposed.Compared with traditional feature extraction algorithms,this thesis adds a part that learns the features of the objects that are initially extracted,so that the feature vector of the object obtained while accurately representing the object information is more sparse,which shortens the time for later classification and recognition.Then,based on the combination of preliminary feature extraction algorithms and feature learning,this thesis combines feature fusion methods and feature learning algorithms to achieve feature extraction of object images,and proposes a feature extraction algorithm based on feature fusion and sparse dictionary learning.The feature vector after the previous feature level fusion of the improved LBP features and HOG features is used as the initial dictionary for dictionary learning,and feature learning is performed on the fusion vector to improve the recognition efficiency of the algorithm.Finally,the SVM classifier was used to classify the features extracted by the feature extraction algorithm based on the improved LBP and sparse dictionary learning and the features extracted by the feature extraction algorithm based on the improved LBP-HOG feature fusion and sparse dictionary learning.The recognition accuracy rates were respectively reached.86.37% and 89.77% verify the effectiveness and feasibility of the feature extraction algorithm in this thesis.The experiments on different feature extraction algorithms on the test set verify that the feature extraction algorithm proposed in this thesis has strong generalization ability.