Research on Improved YOLOv5-Based Small Component Recognition Method of Electrical Drawings

With the development of information technology, there is a demand for digitalization of electrical drawings, in which the identification of electrical components is the basis. We proposed an improved YOLOv5-based method to solve the difficulty in detecting small components due to their obscure characteristics. An improved projection segmentation module was proposed to segment the original drawing into several circuits, which make the feature of elements more obvious. We take the IoU of the prediction box and the ground truth as the distance of K-means++ clustering algorithm to calculate the anchor boxes. Based on the original YOLOv5 structure, we add a new detection channel and two transfer paths of shallow feature map, which can extract more feature information of the small components. In the loss function, we use the SIoU instead of CIoU to calculate the location loss function, which can get more accurate anchor boxes. The experimental results show that the precision of the improved YOLOv5 algorithm can reach 96.4%, which is 35.1%% higher than the original YOLOv5 model, verifying the effectiveness of our method.