| 黄禹康,熊凌,刘洋,邓攀,但斌斌.基于改进U-Net网络的吹氩图像分割方法[J].高技术通讯(中文),2022,32(1):50~56 |
| 基于改进U-Net网络的吹氩图像分割方法 |
| Argon blowing image segmentation method based on improved U-Net model |
| |
| DOI：10.3772/j.issn.1002-0470.2022.01.006 |
| 中文关键词: 深度可分离卷积； U-Net； MobileNet； 吹氩图像； 图像分割 |
| 英文关键词: depthwise separable convolution, U-Net, MobileNet, argon blowing image, image segmentation |
| 基金项目: |
|
| 摘要点击次数: 1772 |
| 全文下载次数: 1189 |
| 中文摘要: |
| 针对传统U-Net网络模型参数量大、图片处理时间长、无法满足工业生产实时性要求的问题，提出了一种改进U-Net网络的吹氩图像分割方法。该方法以U-Net框架为主体，使用传统U-Net网络的特征融合模块高效利用图像的特征信息，利用MoblieNet网络中的深度可分离卷积方法替代传统卷积，降低网络的参数量和计算量，缩短了分割所需的时间。实验结果表明，改进的U-Net网络在保持精度的同时，具有良好的实时性。与传统U-Net网络相比，其参数量缩小15倍，在GPU上运行平均耗时降低6倍。改进的U-Net网络处理一张分辨率为224×224像素的图片的平均耗时为30ms，可以满足工业生产对图像处理实时性的要求。 |
| 英文摘要: |
| Aiming at the problem that the traditional U-Net network model has large parameters and long processing time, which cannot ensure the real-time requirements of industrial production, an improved U-Net network method for argon blowing image segmentation is proposed. This method takes the U-Net framework as the main body and uses the feature fusion module of the traditional U-Net network to make efficient use of the image feature information. In the meantime, the depthwise separable convolution method in the MoblieNet network is used to replace the traditional convolution, which reduces the number of network parameters and calculations, and shortens the time required for segmentation. Experimental results show that the improved U-Net network has excellent real-time performance while maintaining accuracy. Compared with the traditional U-Net network, the number of parameters is reduced by 15 times, and the average running time on GPU is reduced by 6 times. The average time for the improved U-Net network to process a picture with a resolution of 224224 pixels is 30ms, which can satisfy the real-time requirements of industrial production. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|
