YOLOv9-SAM: A Novel Image Segmentation Method for Industrial Pointer Meter

In industrial settings, meters are widely employed for diverse measurement operations. In practice, pointer-style meters are designated for areas with flammable or explosive hazards that require manpower to read the meters. In order to increase the manual efficiency of labor and avoid errors that can result from human interpretation. In this paper, we propose a fuse Segment Anything Model (SAM) with You Only Look Once version 9 (YOLOv9) to resolve industrial pointer meter segmentation. Our method includes two components: (1) we enhance the YOLOv9 with an attention module, which achieves better precision in detecting the point meter. (2) the detection result will be introduced into the prompt module to make the Segment Anything Model focus more on the point meter, meanwhile making the process faster. Our proposed method achieved a detection accuracy of 92.5%, surpassing the original YOLOv9, 87.7%. As for the segmentation part, our method realized a stable, intact dial segment with 90% less operation time. Our enhanced detection and segmentation model has reached a better detection precision and a more accurate segmentation compared to origin methods.