Small Target Detection Algorithm for Industrial Pointer Meters Based on Swin Transformer

To enhance small-object detection accuracy and robustness in pointer instrument detection, this paper proposes an improved STF-YOLOv8 model. It incorporates a Swin Transformer backbone to strengthen feature modeling and small-object recognition. The Focus module and Depthwise Separable Convolution (DWConv) optimize high-frequency feature extraction while reducing computational cost. The SPPCSPC-C3STR structure enhances multi-scale feature interaction, improving detection across different target sizes. The Weighted Intersection over Union (WIoU) loss function refines bounding box regression for higher precision. Experimental results show that STF-YOLOv8 achieves 97.8 % [email protected] on a dataset of 1,000 industrial instrument images, surpassing YOLOv8n by 0.6 %, with an inference speed of 45 FPS. Compared to YOLOv3 and YOLOv5s, the proposed model demonstrates superior accuracy and real-time performance, offering a reliable solution for automated industrial instrument detection.