摘要
Abstract
Aiming at the problems of dense distribution of elements,similar elements and large number of small-sized elements in power dispatching control system diagrams,which lead to poor recognition effect,an improved you only look once v7(YOLOv7)element recognition method for the power dispatching control system diagrams is proposed.Firstly,the lightweight dilated reparam block net with cross stage partial and efficient layer aggregation network(DRBNCSPELAN)module is used to replace the efficient layer aggregation network(ELAN)module in the backbone network to capture spatial patterns of different scales simultaneously.Secondly,an information-guided fusion module is proposed to replace the Concat in the neck network,and the sequeeze-and-excitation(SE)attention mechanism is integrated to enhance the global information interaction ability.Then,the minimum point distance intersection over union(MPDIoU)loss function is introduced to improve the recognition effect of the element bounding box.Finally,experimental validation is performed via the power dispatch control system diagram dataset.The results show that compared with the baseline model,the precision,recall and mean average precision of the proposed method are improved by 5.1 percentage points,3.1 per-centage points and 3.5 percentage points respectively,which is helpful to achieve accurate recognition of elements in the power dispatching control system diagrams.关键词
深度学习/电力调度控制系统图/图元识别/YOLOv7/SE注意力机制Key words
deep learning/power dispatching control system diagrams/element recognition/you only look once v7(YOLOv7)/sequeeze-and-excitation(SE)attention mechanism
