基于卷积神经网络的服装领型识别与分类研究

针对服装领型特征识别困难、分类效果不理想等问题,提出了一种基于Alex Net卷积神经网络的服装领型识别与分类方法,实现了服装衣领造型的自动识别与分类。首先,从电商平台收集服装样本图并对其进行剪裁和预处理,建立一个包含圆形领、方形领、一字领、V形领等15类服装领型的样本库;其次,利用Alex Net卷积神经网络中的卷积、池化操作,提取服装领型样本中的领型特征;最后,运用Softmax回归分类器来实现服装领型的分类。结果表明,该方法可以有效地对服装领型进行识别与分类,分类准确率达到98.67%,可以有效解决服装领型识别分类困难等问题,为服装商品的可视化分类提供有效方法。

服装V领结构对视觉高度的影响

不同服装领型对着装人体的修饰效果不同,具有纵向拉伸视错觉效果的V型领比其他领型的修饰效果更好。文章就服装V领结构对视觉高度的影响展开研究,通过实验数据分析得到,深度不变的V领,宽度越大,着装人体肩宽、脖子被横向拉宽的视错觉效果越明显,人体视觉高度越差;而宽度不变的V领,领深到领宽水平线的垂直距离与领深到头顶的垂直距离之比、领宽水平线到下巴的垂直距离与领深到下巴的垂直距离之比越接近于黄金比例,领子对着装人体的脖子、上身的纵向拉伸视错觉效果越明显,人体视觉高度越好。

基于Mask R-CNN的款式图衣领识别

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。

Study on the Mathematic Model of Apparel Lapel Collar Pattern

Fabrics of different thickness were selected and made into notch lapels in the experiment. The variation between the outside and inside lapel patterns after the lapel being folded was observed and measured. Besides,many factors leading to the variation were analyzed. Then the relation between fold angle and fabric thickness,and the variation of fabric thickness were obtained. The experimental results prove that the biggest horizontal variation between the outside lapel and inside lapel is at the breast line,and when the measuring point is above (or below) the breast line level,the farther it is away from the level,the smaller the variation becomes. In addition,conclusions can be drawn that the thicker the fabric is,the bigger the curve variation will be,and there is a certain mathematic relation between them.