摘要
针对传统刺绣图像分类效率低、数据集少、识别率不高等问题,研究了一种基于卷积神经网络AlexNet模型的刺绣分类算法,将深度学习技术引入到刺绣图像分类领域。为了更精准地实现多种刺绣分类,对原始AlexNet模型进行改进,引入迁移学习,并加载预训练模型参数,从而提取更深层次的刺绣特征,提高模型的收敛速度。实验选取国内10种刺绣,共计864张原始刺绣图片,使用包括基本数据增强、Mixup混合技术和GridMask技术得到9571张图片和对应标签进行训练,与原始算法模型进行对比,并进行消融实验。实验结果表明:改进后的AlexNet模型准确率达到94.74%,对多种类刺绣图像分类有更好的效果。研究结果可有效解决刺绣分类困难问题,最大程度完善和优化数字化的刺绣传承体系。
In response to the challenges posed by low efficiency,limited datasets and poor recognition rates in traditional embroidery image classification,an embroidery classification algorithm based on AlexNet model was studied,thus bringing the benefits of deep learning technology into the field of embroidery image classification.In order to achieve more accurate classification of multiple embroideries,the original AlexNet model was improved by introducing migration learning and loading pretrained model parameters,thus extracting deeper embroidery features and improving the convergence speed of the model.In the embroidery image classification experiment,a total of 864 original embroidery pictures of ten kinds of domestic embroidery were selected,and 9571 images and corresponding labels were obtained by using basic data enhancement,GridMask technology and Mixup hybrid technology for training,compared with the original algorithm model,and ablation experiment was conducted.The experimental results show that the accuracy of the improved AlexNet model reaches 94.74%,which has a better effect on the classification of multiple kinds of embroidery images.The results can effectively solve the difficult problem of embroidery classification and improve and optimize the digital embroidery inheritance system to the greatest extent.
作者
鲍亚林
唐戈
BAO Yalin;TANG Ge(School of Government,Heilongjiang University,Harbin,Heilongjiang 150080,China)
出处
《毛纺科技》
CAS
北大核心
2023年第6期81-87,共7页
Wool Textile Journal
