Aiming at the weakness of visual words and the ignorance of spatial positional relationship between features in latent Dirichlet allocation（ LDA） model

a LDA fastener image detection model based on local feature coding is proposed

which is named LC-LDA. Firstly

visual dictionary is obtained by bottom layer features of images of K-means clustering. Secondly

locality-constrained linear coding（ LLC） which can reduce the reconstruction error of features is used in the coding process. And then

the spatial information in graph regularized is used to ensure the features which are similar and adjacent had the same coding. Finally

the topic distributions of the images are extracted by LDA topic model

and the fastener category is learned in the support vector machine（ SVM）. The experimental results show that the discrimination of each category is enhanced in LC-LDA model

and the rate of missed detection and false detection is significantly reduced.