改进的Faster-RCNN算法在聚乙烯管接头内部缺陷检测中的应用

超声相控阵技术是目前聚乙烯管道热熔接头内部缺陷检测的一种主流方法。提出了基于注意力机制的改进Faster-RCNN目标检测网络用于超声相控阵D扫图聚乙烯管接头内部缺陷检测。针对聚乙烯管道热熔接头内部超声相控阵D扫图小缺陷较多、特征信息容易丢失的问题，将残差网络(ResNet50)与特征金字塔网络(FPN)相结合作为骨干网络，并引入卷积注意力模块(CBAM)自适应细化特征。将SSD网络框架和Faster-RCNN网络框架用于模型训练和测试，使用VGG16、ResNet50、ResNet50+FPN、ACBM+ResNet50+FPN作为骨干网络依次对超声相控阵聚乙烯管道热熔对接接头内部缺陷样本进行训练对比。结果表明，改进的Faster-RCNN网络模型在聚乙烯管接头内部缺陷检测和分类方面有明显改进，对小缺陷的检测性能有了显著的提高。

Application of improved Faster-RCNN algorithm in internal defect detection of polyethylene pipe joint

Ultrasonic phased array technology is a mainstream method to detect the internal defects of polyethylene (PE) pipe hot melt joint. In this paper, an improved Faster-RCNN target detection network based on attention mechanism is proposed for ultrasonic phased array D-sweep PE pipe joint internal defect detection. To solve the problem that there are many small defects in ultrasonic phased array D-sweep pattern inside hot melt joint of PE pipeline and feature information is easy to be lost, residual network (ResNet50) is combined with feature pyramid network (FPN) as the backbone network, and convolution block attention module (CBAM) is introduced to self-refine feature. SSD network framework and Faster-RCNN network framework were used for model training and testing. VGG16, ResNet50, ResNet50+FPN and ACBM+ResNet50+FPN backbone networks were used as backbone networks to train and compare the internal defect samples of ultrasonic phased array PE pipeline hot melt butt joint. The results show that the improved Faster-RCNN network model has obvious improvement in the detection and classification of internal defects of PE pipe joints, and the detection performance of small defects is significantly improved.