###
:2019,28(2):41-48
本文二维码信息
码上扫一扫!
基于改进Faster R-CNN的Logo目标检测方法
(浙江理工大学 信息学院, 杭州 310018)
Logo Object Detection Method Based on Improved Faster R-CNN
(School of Information Science and Technology, Zhejiang Sci-Tech University, Hangzhou 310018, China)
摘要
图/表
参考文献
相似文献
本文已被:浏览 298次   下载 154
投稿时间:2018-08-10    修订日期:2018-09-06
中文摘要: 社交网络的发展和经济全球化赋予了Logo巨大的商业价值,让Logo检测有很大的应用前景.而现实情况中Logo目标在图片中往往占据很小的一部分,Logo的低分辨率导致检测性能难以进一步提升.因此本文提出了一种基于改进Faster R-CNN的Logo检测方法.该方法在Faster R-CNN框架中结合了生成对抗模型,利用网络先将分辨率较低的Logo特征映射成高分辨率的表达能力更强的特征,再送入完全连接层进行分类和回归,从而提高检测的性能.本文在公开的Logo数据集上进行了实验结果评估,结果表明了本文提出的方法能有效地提高Logo物体检测的准确率的同时也没有因为结构的变化影响检测速度.
Abstract:The development of social networks and economic globalization gives logo a great commercial value, which makes logo detection have a good application prospect. In fact, logo objects typically occupy a small portion of the image, and the low resolution of the logo makes it difficult to further improve detection performance. Therefore, this study proposes an improved detection method based on Faster R-CNN. This approach combines the generative adversarial networks and a Faster R-CNN framework, uses the network to map lower resolution features to highly expressed high resolution features, and then sends them to fully connected layers for classification and regression. The outcomes of the experiment are evaluated on a publicly available logo dataset. The results show that the method can effectively improve the accuracy of logo object detection without affecting the detection speed of the basic network.
文章编号:     中图分类号:    文献标志码:
基金项目:
引用文本:
黄明珠,黄文清.基于改进Faster R-CNN的Logo目标检测方法.计算机系统应用,2019,28(2):41-48
HUANG Ming-Zhu,HUANG Wen-Qing.Logo Object Detection Method Based on Improved Faster R-CNN.COMPUTER SYSTEMS APPLICATIONS,2019,28(2):41-48

用微信扫一扫

用微信扫一扫