Cutting Pattern Positioning Method Based on Improved ROI Pooling of R2CNN
RIS ID
145891
Abstract
© 2020 ACM. It is of great significance for textile industry to realize automatic pattern detection and positioning. In this paper, combining with image processing technology and deep learning theory, an improved pattern location method based on R2CNN is proposed. Firstly, the multi-scale ROI pooling structure was designed on the basis of R2CNN network, the proportion of the suggestion window generated by RPN network was adjusted, and the pattern Angle prediction function was introduced. The experimental results show that the training on the self-made and labeled data sets achieves an average accuracy of 85%, which greatly improves the positioning accuracy of cut patterns.
Publication Details
L. Geng, Y. Liu, Z. Xiao, J. Tong, F. Zhang & J. Wu, "Cutting Pattern Positioning Method Based on Improved ROI Pooling of R2CNN," in ACM International Conference Proceeding Series, 2020, pp. 212-217.