%0 Journal Article
%T 基于多尺度类注意力元学习的丝绸图案检测<br>Silk Pattern Detection Based on Meta Learning with Multi-Scale Class Attention
%A 葛亮
%A 王直杰
%J Computer Science and Application
%P 2780-2787
%@ 2161-881X
%D 2021
%I Hans Publishing
%R 10.12677/CSA.2021.1111282
%X <div style=\"text-align:justify;\">
	<span style=\"font-size:14px;\">中国是丝绸的发源地，丝绸图案作为丝绸的主要元素承载了深远的历史文化。计算机视觉技术在丝绸图案的检测并不多见，主要因为丝绸图案类别繁多，总体数量不多，一般的目标检测技术无法在丝绸这样的小样本数据集上发挥良好的效果。针对该问题，本文提出了一种适用于小样本丝绸数据集的多尺度注意力目标检测算法，通过多尺度类向量为目标检测网络提供更多的分类信息，实现丝绸图案的检测任务。本文针对古代丝绸中常见图案进行数据收集和数据标定，并且利用提出的网络实现对不同年代的狮子和花卉进行实验。实验结果表明该方法可以实现小样本丝丝绸图案目标检测任务，而且效果优于其他小样本目标检测算法。<br />
China is the birthplace of silk. Silk patterns, which are the main element of silk, carry far-reaching history and culture. Computer vision technology is rare in the detection of silk patterns, mainly because of the wide variety and small number of silk patterns. The general object detection methods can not play a good effect on a small sample data set such as silk. To solve this problem, this paper proposes a multi-scale attention target detection algorithm suitable for small sample silk data sets, which provides more classification information for the target detection network through multi-scale class vectors to realize the detection task of silk patterns. In this paper, data collection and data calibration are carried out for common patterns in ancient silk, and the proposed network is used to experiment lions and flowers in different ages. The experimental results show that this method can realize the target detection task of small sample silk pattern, and the effect is better than other small sample target detection algorithms.</span> 
</div>
%K 小样本学习，多尺度，类注意力，目标检测<br>Few-Shot Learning
%K Multiple Scale
%K Class Attention
%K Object Detection
%U http://www.hanspub.org/journal/PaperInformation.aspx?PaperID=46804