Fabric periodic texture detection method based on deep learning

The invention relates to a fabric periodic texture detection method based on deep learning, which is used for automatically detecting periodic texture primitives of a fabric. The method specifically comprises the following steps of: calibrating an input fabric image by using dip angle transformation and a non-local variation algorithm, inputting a preprocessed image into a network in a detection process, generating an activation peak following a feature rule in each feature layer after convolution operation and a non-maximum suppression algorithm, and finally, carrying out detection on the fabric image. And further finding the positions of the periodic texture primitives through a Hough voting strategy and the centroid coordinates. According to the method, on the basis of the deep learning technology, periodic texture detection is carried out on the fabric, key point detection, feature extraction and clustering of traditional manual design are replaced, the effect of capturing higher-level pixel and regional