基于深度学习网络U-Net的轮胎带束层分割算法研究

Research on Tire Belt Segmentation Algorithm Based on Deep Learning Network U-Net

轮胎成型工艺中的带束层贴合情况影响成品轮胎的质量,而现有的带束层缺陷检测算法存在检测精度低,速度慢的缺点。为了快速精确的分割带束层,提出了一种基于U-Net的带束层分割算法。首先对带束层进行鼓面去除预处理,然后结合带束层区域的特点,对比基于数学形态学的传统图像分割方法与深度学习语义分割网络U-Net,对带束层进行分割。研究结果表明,两种方法都可以很好地标识带束层区域和背景区域,但U-Net稳定性更强,可以满足工程中精度与实时性的要求。

The belt defects in tire forming process affect the quality of the finished tire,and belt segmentation is significant in defect detection.However,the existing belt segmentation methods have low detection accuracy and slow speed.To segment the belt accurately and fast,a belt segmentation technique based on deep learning neural network U-Net is proposed.Firstly,preprocess the belt by removing drum surface.Then,it is segmented according to the characteristics of the belt region.Select an appropriate algorithm by comparing the segmentation effects of traditional image segmentation method based on mathematical morphology and semantic segmentation network U-Net.The result shows that the belt layer area and the background area can be well identified by the two methods.But U-Net is more stable than traditional image segmentation method.So U-Net can meet the requirements of accuracy and real-time in engineering.