基于DeepLabv3+的电网引流线夹中心螺杆自动化分割  

Automatic Segmentation Method of Center Screw of Power Grid Drainage Clamp Based on DeepLabv3+

作　　者：李佳馨 刘君[1] 燕振华 LI Jia-xin;LIU Jun;YAN Zhen-hua(School of Information Engineering,Nanchang Hangkong University,Nanchang 330063)

机构地区：[1]南昌航空大学信息工程学院,南昌330063

出　　处：《南昌航空大学学报（自然科学版）》2023年第4期73-81,96,共10页Journal of Nanchang Hangkong University(Natural Sciences)

基　　金：国家自然科学基金(61963030)。

摘　　要：基于计算机视觉的线夹孔洞定位是实现电网引流作业自动化的关键环节之一。线夹的中心螺杆作为线夹孔位的重要定位标志,对其进行自动化分割是实现线夹孔位定位的有效手段。为此提出了一种基于DeepLabv3+的线夹中心螺杆自动化分割方法,同时将DeepLabv3+的分割性能与主流卷积神经网络U-Net、SegNet和PspNet进行比较。结果显示,基于DeepLabv3+的分割方法对线夹中心螺杆的分割不但具有96.78%的总体精度,在敏感性和DICE相似性定量分析中也都表现较好。该方法能够实现线夹孔位的自动化定位,为实现电网引流作业自动化提供了一种行之有效的方法。The location of wire clamp hole based on computer vision is one of the key points to realize the automation of power grid drainage operation.The center screw of the wire clamp is an important mark of the position of the wire clamp hole,so the automatic segmentation of which is an effective means to realize the position of the wire clamp hole.Therefore,an automatic segmentation method for wire clamp center screw based on DeepLabv3+is proposed in this work.The performance of the proposed method is also compared with that of the mainstream convolutional neural networks PspNet,SegNet and U-Net.The result shows that the segmentation of DeepLabv3+not only achieves 96.78%overall accuracy,but also performs better in sensitivity and DICE similarity quantitative analysis.The proposed method could detect the automatic position of wire clamp hole,providing a proven and effective method for the automatic drainage operation of power grid.

关 键 词：DeepLabv3+ 螺杆自动分割 深度卷积神经网络 图像语义分割 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]