Aiming at shortcomings of traditional visual bag-of-words（ BOW） model considers nothing about contextual semantic information in spatial domain

while railway fastener state is recognized using feature domain of fastener image

a fastener inspection model based on contextual semantic information is proposed. On the basis of traditional BOW model

the Gibbs random field model is introduced to combine the similarity of feature domain and contextual semantic information of spatial domain in order to more accurately define visual words. And the latent Dirichlet allocation（ LDA） is used to learn topic distribution. Support vector machine（ SVM） is applied to classify and identify fastener. Experimental results on four kinds of fastener images show that the proposed model can increase classification precision of fastener more effectively.