摘要
利用空洞卷积设置不同空洞率,得到不同感受野的特点,提出一种基于深度卷积Visual Geometry Group19(VGG19)和空洞卷积相融合的结构.所采用的结构不受输入图像尺寸以及分辨率影响,通过设置锯齿状空洞率,扩大网络的感受野,在保持分辨率良好的情况下,可以较为精确地定位目标,提高检测准确性.经验证,该算法在Shanghai-tech标准数据集上具有较高的实验准确率.
A combined structure based on Visual Geometry Group19(VGG19)and dilated convolution with different receptive field was proposed for high density crowd counting in the paper.The structure adopted would not be affected by the size and resolution of the input image.By setting the serration dilation rate,the network receptive field was expanded,and the target could be accurately localized without any loss of resolution,which improved the accuracy of detection.Finally,the experimental results showed that the algorithm had higher accuracy on the standard data set of Shanghai-tech.
作者
盛馨心
苏颖
汪洋
SHENG Xinxin;SU Ying;WANG Yang(College of Information,Mechanical and Electrical Engineering,Shanghai Normal University,Shanghai 200234,China)
出处
《上海师范大学学报（自然科学版）》
2019年第5期479-484,共6页
Journal of Shanghai Normal University(Natural Sciences)
