基于图像分割的改进Mask R-CNN绝缘子检测方法

Improved Mask R-CNN Insulator Detection Method Based on Image Segmentation

目标检测技术在很多领域有着广泛的研究和应用,如电力行业、军事目标检测等,目前学术界已经提出了很多的目标检测方法,取得了很大进展。近年来,电力行业对输电线路风险预警要求急剧上升,因此高效可靠的目标检测算法必不可少。主要针对输电线路绝缘子目标检测精度低、单张图片检测速度慢、未能满足实际应用的问题,提出基于注意力机制的可形变多尺度绝缘子目标分割模型。该模型将Mask R-CNN主干网络从RestNet101-FPN替换为可形变特征网络ResNet18-dcn;其次,在主干网络后增加注意力机制模块,将主干网络可形变特征网络ResNet18-dcn的3、4、5阶段输出进行多尺度信息融合,提升网络的识别能力及效率。通过实验证明,改进模型能够满足当前输电线路绝缘子识别所需。

Target detection technology has been widely studied and applied in many fields,such as power industry,military target detection,etc.At present,many target detection methods have been proposed in academia,and great progress has been made.In recent years,the demand of power industry for transmission line risk early warning has risen sharply,so efficient and reliable target detection algorithm is essential.This paper mainly aims at the low detection accuracy of transmission line insulator target and the slow detection speed of single image,which cannot meet the practical application problems.A deformable multi-scale insulator target segmentation model based on attention mechanism is proposed.This model replaces the Mask R-CNN backbone network from RestNet101-FPN to the deformable feature network ResNet18-dcn.Secondly,the attention mechanism module is added behind the backbone network to conduct multi-scale information fusion by outputting the stage 3,4 and 5 of the deformable feature network ResNet18 dcn of the backbone network,so as to improve the recognition ability and efficiency of the network.The experimental results show that the improved model can meet the needs of insulator recognition of transmission lines.