关键词: 铁路扣件检测, 词包模型, 视觉单词, 信息熵, 潜在狄利克雷分布模型

Abstract: The traditional Bag Of Words（BOW） model ignores the structure of the image while conducting the inspection of railway fasteners. To overcome this defect, this paper proposes an Entropy-Weighted BOW（EW_BOW） model for identification of the image of fasteners. On the basis of the traditional BOW model, the entropy is introduced to weight word frequency of BOW model?in the sub-image of fasteners, thus making the BOW model more distinguishable for different categories of fasteners. And then the Latent Dirichlet Allocation（LDA） is used to learn the topic distribution of the images. Finally, the Support Vector Machine（SVM） is applied to classify a new image. The experiment on four types of fasteners shows that the EW_BOW model can inspect fastener states more precisely.

Key words: railway fastener inspection, Bag Of Words（BOW） model, visual word, entropy, Latent Dirichlet Allocation（LDA） model