Research on water meter data augmentation based on arbitrary style transfer algorithm of large convolutional kernel
扫码查看
点击上方二维码区域，可以放大扫码查看
原文链接
维普
万方数据
针对现有的风格迁移方法在对水表进行数据增强后导致颜色失真,内容保留不完整等问题,提出了一种基于大卷积核的任意风格迁移算法(arbitrary style transfer algorithm of large convolutional kernel,LKAST).首先,针对风格图像使用大卷积核提取风格特征,保留风格特征的高层特征;此外,通过引入新的损失函数,更好的保留迁移结果对内容的保留;最后,通过两组对照实验验证方法的有效性.实验结果表明,该方法能够在模拟水表现场环境的同时保留足够的内容信息,在仅改变数据增强算法的前提下,单次多框目标检测(SSD)算法准确率提升6.84%,YOLOv5准确率提升6.56%.
In view of the limitations of the existing style transfer methods in the data enhancement of water meters.In this paper,we proposed an arbitrary style transfer algorithm based on large convolutional kernel(LKAST).Firstly,the large convolution kernel is used to extract the style features for the style images,and the high-level features of the style features are retained,and a new loss function is introduced to better preserve the content of the migration results.Finally,the effectiveness of the method was verified by two groups of controlled experiments.Experimental results show that the proposed method can simulate the water performance field environment while retaining enough content information,and the accuracy of the SSD object detection algorithm is increased by 6.84%and YOLOv5 by 6.56%under the premise of only changing the data augmentation algorithm.
data augmentationlarge convolutional kernelstyle transferloss function
维普
万方数据
