Application Of Rejectable Pattern Recognition Method Based On Fisher Dictionary Learning

The traditional recognition methods follow the general research rule of extracting features first and inputting classifier later,which are greatly affected by selected features,and increase the complexity of recognition.Warehouse management system is an important part of the enterprise.With the continuous development of modern enterprises,the accurate and rapid identification of warehouse goods have received more and more attention.However,more types and unobvious characteristics of warehouses goods increase the difficulty by applying traditional classification methods.On the basis of sparse representation and rejectable model,the method of rejectable pattern based on Fisher dictionary is proposed.Firstly,the Fisher criterion is applied to the discriminant dictionary learning for the problem that the classical SRC loses a lot of classification information in the process of constructing the dictionary.So the dimension of the discriminant dictionary and the number of atoms in the dictionary can be less and the identification can be simpler and faster;Then,a method of rejectable pattern recognition based on Fisher dictionary learning is proposed to effectively deny the non-training exception model;Finally,for the construction of super-ball coverage directly to the training samples,The paper proposes a modified method of rejectable pattern recognition based on Fisher dictionary learning.The training samples are decomposed by the Fisher discriminant dictionary to construct local linear blocks.The super-ball cover model is established to approximately describe the distribution of the training samples and then to reject.Experiments show that the effectiveness,recognition and rejection of this method are better in ORL face database,Caltech-101 database and MINIST handwritten digital library.Finally the proposed method is applied to the warehouse product identification.