人体着装围度松量图像的分类方法

Classification of human chest looseness based on transfer learning

针对人体着装围度松量图像分类效果不理想、收集目标数据集较为困难等问题,提出一种人体着装围度松量图像的分类方法,即基于迁移学习方法构建人体着装围度松量图像分类的模型。以胸部为例,首先,迁移VGG16模型,预训练VGG16模型,将从VGG16模型学到的知识迁移到目标小数据集上并进行微调得到人体着装时胸部松量的特征;其次,将VGG16迁移模型与inceptionV3迁移模型进行对比分析。研究结果表明:在构建的人体着装胸部松量小数据集上,迁移VGG16模型准确率最终达到了82.1%,比inceptionV3迁移模型精准度高15.9%,验证了VGG16模型的有效性。并为更好实现远程在线服装定制中人体围度的测量提供有效途径。

Aiming at problems such as unsatisfactory classification of images of loose body dress circumference and difficulty in collecting target data sets,this paper proposed a classification method for the image of loose body clothing circumference,which was based on the transfer learning method to build a classification model for the looseness of human clothing circumference image.Take the chest as an example,firstly,the VGG16 model was migrated and pre-trained,and the knowledge learned from the VGG16 model was transferred to the target small data set for fine-tuning to obtain the characteristics of chest looseness in human clothing.Secondly,the VGG16 migration model was compared with the inceptionV3 migration model.The results show that the accuracy of the model finally reaches 82.1%on the constructed human body dressing chest loose small data set,which is 15.9% higher than the migration model of inceptionV3,verifying the effectiveness of VGG16 model.It also provides an effective way for the measurement of body circumference in the remote online clothing customization.