基于特征调和筛选的焊缝缺陷语义分割研究

Research on Semantic Segmentation of Weld Defects Based on Feature Harmonic Screening

提出了一种基于特征调和筛选的X射线不锈钢焊缝缺陷语义分割模型。通过池化与特征调和筛选块对输入图像进行下采样,利用特征注意模块为相同层级的特征图进行通道与空间重新赋权,并且使用反卷积逐步恢复特征图大小。实验表明:在X射线不锈钢焊缝缺陷数据集上进行语义分割识别性能时,本方法MIoU指数达到了53.87%,相较于其他方法,本方法具有更高的识别准确度与图像处理速度,具有一定的工程应用价值。

A semantic segmentation model of X-ray stainless steel weld defects based on feature harmonic screening is proposed.The input image is downsampled by pooling and feature harmonic screening blocks,the feature attention module is used to re-weight the channel and space of the feature map at the same level,and deconvolution is used to gradually restore the size of the feature image.The experimental results show that when the semantic segmentation recognition performance is carried out on the X-ray stainless steel weld defect data set,the MIoU index of this method reaches 53.87%.Compared with other methods,this method has higher recognition accuracy and image processing speed,and has a certain engineering value.