| In order to strengthen substation equipment status monitoring and ensure stable operation of substations,highly reliable identification of substation equipment status information is required.In this paper,a substation small target detection and identification system is designed to intelligently identify substation digital meter readings and small target indicator status information that are subject to decimal point interference,effectively improving the reliability of substation equipment status detection.The main research work and innovations of this paper can be summarized as follows:(1)To address the problem that the extraction of semantic features is easily affected by the small size of small targets in substations,a scale-adaptive detection strategy is adopted and a detection framework based on a localization module and a detection module is designed.Specifically,Faster R-CNN is used as a region localization network to localize and segment the region of interest containing small targets,which then enters the detection module for small target detection.(2)To address the problem of inaccurate recognition of digital meter readings in substations subject to decimal point interference,an algorithm for recognising digital meter readings in substations based on connected domain analysis is designed.In the detection module of the algorithm,the detection accuracy is improved by adding modules for analysing the category and area attributes of the connected domain to reduce the leakage and false detection rates of the detection network YOLO-V4 in detecting high similarity targets.The experimental results show that the average accuracy of the training model is improved from98.29% to 99.92%.(3)Small target indicators in substations are susceptible to optical artefacts and are densely distributed when multiple targets are present,making detection more difficult,so this paper designs an algorithm for identifying the status of small target indicators in substations.The algorithm detection module is based on the high real-time target detection algorithm YOLO-V5,and is improved from both the backbone network and the addition of hybrid attention,which effectively improves the average accuracy of the detection network.Experimental results show that the average accuracy of the improved detection network is improved from 87.4% to 92.7%,and the visualisation test results show that the leakage rate of the model is effectively reduced.(4)On the basis of the substation small target detection and recognition algorithm model,this paper implements the construction of the substation small target detection and recognition system from the front-end user data interaction design and the interface design of the algorithm logic module,and carries out practical tests using the collected substation small target image data.Through theoretical analysis,network model design,experimental verification and system design,this paper realises the establishment of a small target detection and identification system for substations,which provides a reference for the intelligent development of substations. |
