The size of the primitive unit of the fabrics is estimated using distance matching function

and the fabrics are split into samples of one unit. Then

nonnegative matrix factorization（ NMF） method is used to obtain the coefficient matrix of image blocks and the coefficient matrix is converted to coefficient feature. According to normal samples

the standard coefficient feature and similarity criterion are determined. The coefficient feature of the test sample and the standard coefficient are compared. According to the similarity criterion

whether the image block contains defects can be judged. In the process of nonnegative matrix factorization

nonnegative double singular value decomposition initialization method（ NNDSVD） is adopted to reduce the number of iterations of matrix decomposition. The results show that the proposed method can accurately detect whether the fabric contains defects

and the accuracy rate is over 96 %.