基于Mask R-CNN的款式图衣领识别

Collar Flat Sketch Recognition Based on Mask R-CNN

为提升服装款式图领型识别精度,提出一种基于Mask R-CNN神经网络的服装款式图领型定位与识别方法。建立共1800张包含无领、立领、翻领与驳领4种领型的款式图样本库,利用迁移学习与Mask R-CNN神经网络实现领型定位与识别。结果表明,4种领型的平均识别精确度高于98%,测试集平均精确度达到99.2%,mAP值达到90%。该识别方法可以减少样板生成中的人工失误,为数字化样板生成提供参考。

In order to improve the recognition accuracy of collar flat,this paper proposed a fine-grained localization and recognition method of garment collar based on Mask R-CNN neural network.A database of 1800 images which contained collarless,lapel,turndown collar and stand collar,was created.Transfer learning and Mask R-CNN neural network were used to realize collar location and recognition.The results showed that the average recognition accuracy of four collar types were higher than 98%.The average accuracy of the test set and m AP were reached to 99.2%and 90%respectively.The recognition method can reduce the manual errors in the pattern generation,and provide a reference for the digital pattern generation.