Abstract:

Due to traditional fingerprint localization algorithms cost too much on collecting training dataset with labels,a new semi-supervised fingerprint localization algorithm based on manifold regularization is proposed.Firstly,the graph laplacian is initialized with the labeled and unlabeled training dataset under the manifold hypothesis.Subsequently,combined with the extreme learning machine,the hidden layer is constructed by using random mapping.Finally,the output weighted matrix between the hidden layer and output layer is solved under the framework of manifold regularization,so that the location estimated model is established.The simulation results show that,compared with INN,SVR and ELM,the proposed algorithm need shorter training time and testing time.In addition,it can maintain the high accuracy and reliability while labeled training dataset is sparse.

Key words: fingerprint localization, semi-supervised learning, manifold regularization, extreme learning machine, graph Laplacian