基于RT-GLV的变电站电力人员绝缘手套穿戴检测方法OA

The insulating gloves worn by power personnel in substations were small target in size and were easily obscured.Aiming at the problem that general feature fusion networks often lost small target information,a multi-scale small target feature fusion network named STPFM was constructed.The RT-DETR-R18 model was improved,and the RT-GLV model was designed for detecting whether power personnel were wearing insulating gloves.Firstly,the STPFM network was used to replace the CCFM network.The SSFF module and TFE module of the network were utilized to fuse multi-scale feature information.In addition,a small target detection layer with the SSFF module as the core was added to enhance the model′s ability to learn small target information.Secondly,to address the issue of excessive model parameters after replacing the STPFM network,a lightweight PB Block module was constructed.Only the modules in the P4 and P5 layers of the Backbone network,which contained less small target information,were replaced.It not only lightened the model but also reduced the loss of small target information.Finally,the PI-oUv2 loss function was adopted to enhance the model′s learning ability for both easy and difficult samples.The ex-perimental results showed that the RT-GLV model performed excellently in the detection of whether power personnel were wearing insulating gloves.Compared with the RT-DETR-R18,the mAP@0.5 was increased by 2.1 percent-age points,the F1 score was increased by 1.6 percentage points,and the number of model parameters was reduced by 21.5%.In terms of small target detection,the AP@0.5 of wearing insulating gloves was increased by 1.4 per-centage points,and the AP@0.5 of not wearing insulating gloves was increased by 6.4 percentage points.Moreo-ver,the model′s detection speed reached 91.3 frame per second,meeting the requirements of accuracy and real-time performance for detecting whether power personnel were wearing insulating gloves.