基于Deeplab v3+的PIV误矢量修复

Deeplab v3+Based PIV Error Vector Repair

在PIV图像后处理过程中,修正其误矢量对获取准确速度矢量场起到关键作用。本研究在模拟和实验矢量图数据库中人为添加误矢量,通过标注添加误矢量图像,用Deeplab v3+算法来识别分割误矢量,并同时生成用于修正的掩膜(mask),其识别分割的损失控制在0.001。对于修复,使用了DeepFill v2算法,用正确矢量图像训练网络,将损失稳定于0.05。将识别生成的mask和原始待修正的PIV图像,放入修正图像,靠输入指令即可实现PIV矢量图像中的误矢量自动识别与修正。结果表明,此方法,可批量处理PIV矢量图像,简单方便且自动化,具有较高的识别分割与修正精度。

During the post-processing of PIV images,correction of their error vectors plays a key role in obtaining accurate velocity vector fields.In this study,the error vectors were artificially added to the simulated and experimental vector map database,and the Deeplab v3+algorithm was used to identify the segmented error vectors by annotating the added error vectors images and generating the mask(mask)for correction at the same time,and the loss of its identification segmentation was controlled at 0.001.For the repair,the DeepFill v2 algorithm was used to train the network with the correct vector images to stabilize the loss at 0.05.The recognition of the generated mask and the original PIV image to be corrected are put into the corrected image,and the automatic recognition and correction of error vectors in the PIV vector image can be achieved by inputting commands.The results show that this method,which can batch process PIV vector images,is simple,convenient and automatic,and has high recognition segmentation and correction accuracy.