基于多尺度上下文感知的绝缘子缺陷检测网络

Detection Network for Insulator Defects Based on Multi-scale Context Awareness

针对巡检图像中绝缘子缺陷尺度不一造成检测效果不佳的问题,提出了一种基于多尺度上下文感知的绝缘子缺陷检测网络,称为上下文感知缺陷检测网络(context aware defect detection network,CAD^(2)Net)。该网络采用ResNeSt101架构提高了对图像的特征提取能力。设计了改进特征金字塔结构,构建不同分辨率的丰富语义特征图,以更好地检测不同尺度的目标。同时,在网络的检测单元中增加感受野自适应(adaptive receptive field,Ada-RF)模块聚合多尺度上下文信息,生成更具辨别力的特征,改善网络对不同尺度目标的检测效果。在随机生成缺陷的样本集及公开数据集上的平均检测精度分别达到91.7%及91.0%。结果表明:所提出的缺陷检测网络能够对不同尺度绝缘子的缺陷进行准确识别与定位。

To solve the problem of poor detection results caused by different sizes of insulator defects in inspection images,a multi-scale context aware defect detection network(CAD^(2)Net)is proposed.The network adopts ResNeSt101architecture to improve the feature extraction capability of images.An improved feature pyramid structure is designed to detect targets at different scales by rich semantic feature maps with different resolutions.At the same time,the adaptive receptive field(Ada-RF)module is added to the detection unit of the network to aggregate multi-scale context information,generate more discriminative features,and improve the detection effect of the network on targets of different scales.The average detection accuracy of the randomly generated defects dataset and Chinese power line insulator dataset is 91.7%and 91.0%respectively.The results show that the proposed defect detection network can accurately identify and locate the defects of insulators of different scales.