Improved RM-YOLO Model Based on Transformer for Intelligent Monitoring Meters

Pointer meters are still used in large numbers in power engineering because of their reliability in complex electromagnetic environments, but the complex electromagnetic and machinery environment of substations causes huge problems for intelligent identification to identify the meter. In order to solve the current problem of automatic recognition of pointer meters in substations, this paper proposes an improved RM-YOLO model based on Transformer. By introducing the Transformer module in the backbone, it enhanced the model's ability to integrate shallow and deep feature information. The fusion of multi-scale feature enhances the dynamic adjustment accuracy of meter edge feature at different scales. To solve the increased data volume and reduced computational speed caused by the introduction of the Transformer, the RMNet module is then introduced to replace the RepVGG module, significantly reducing the complexity of the model, effectively improving its recognition speed and further enhancing its ability to mine deep data. The method has been validated to achieve an average target recognition accuracy of 94.7%, enabling effective and accurate identification of pointer meters by inspection robots.