关键词: 旋转目标检测, 电表铅封螺钉, YOLOv8obb, 多尺度特征融合, 注意力机制

Abstract: To solve the problem of difficult to accurately and quickly locate the protecting cover seal screw in the scene of automatic installation for single-phase smart meters,a rotating detection algorithm GROOVE-YOLO based on improved YOLOv8obb was proposed,which could obtain central position and rotating angle of the target screw through a single detection. Firstly,the P2 detection head was added and the P5 feature layer was deleted to build a tiny object detection network,which boosted small object detection capability while simplifying the network structure; secondly,the bidirectional feature pyramid network was fused and learnable weights were applied to improve the effect of multi-scale feature fusion and enrich the model feature expression; finally,the global attention mechanism was employed to highlight the key information in features and further enhance the detection accuracy. The experimental results showed that precision,recall,average precision of the improved algorithm reached 93.2 %,90.5 %,95.7 %,which were increased by 13.5 %,17.1 %,14.3 % compared with the original model. The average pixel distance error of the center point was 1.4 pixels,the average rotating angle error was 4.8°,and the detection speed reached 119 f/s. The proposed method can effectively improve the position accuracy of the protecting cover seal screw and meet the actual installation requirements.

Key words: rotating object detection, meter seal screw, YOLOv8obb, multi-scale feature fusion, attention mechanism