Fabric Linear Defect Detection Based on Mask RCNN

Aiming at the problems of small linear defect area and soft cloth surface of fabric, a textile defect detection method based on Mask RCNN is proposed, which uses the way of image cutting to locate the existence of defects more accurately, and then uses the deformable convolution network to obtain a more accurate defect range, and then changes the anchor proportion in the RPN network to adapt to the characteristics of the detection target, The experiment is carried out through the textile defect image data set. Through experimental comparison., it is concluded that the original Mask RCNN network model is superior to other networks in solving the problem of linear textile defects., and then it is compared that the method proposed in this paper can effectively improve the accuracy of linear textile defects., reduce the false detection rate and missed detection rate of textile defects., and shorten the detection time. Compared with the original model., the accuracy is improved by 12.3%., the missed detection rate is reduced by 12.3%., and the false detection rate is reduced by 10%., Map increased by 1.6%. In order to verify the applicability of this model in other linear target detection with similar characteristics., tests on open road crack and metal surface crack data sets can achieve good results., which verifies the generalization of this algorithm.