Coal mine tunneling anchor operation is a key factor affecting the efficiency of tunneling, how to accu-rately determine the location of anchor holes under the tunnel conditions with many interfering factors is one of the key links to improve the anchoring speed. At present, most of the studies on anchor hole locali-zation consider the interference factors in a single way, and the influence of complex factors on anchor hole identification under multi-scale changes has not been studied yet. Therefore, an improved anchor hole localization method under multi-scale complex environment of YOLOv8s is proposed. First, in order to simulate the real conditions of coal mine roadways, three levels of influence scales are established for the main factors affecting anchor hole localization, such as light intensity, dust and water mist concentration, shooting distance, etc., and the initial data set is constituted by the combination of each influence factor. Image preprocessing operation is performed before anchor hole recognition to optimize the overall image quality. Then, the YOLOv8s deep learning network model is improved by adding the multi-scale void at-tention mechanism (MSDA), which improves the feature extraction ability of the model for small targets under complex conditions. Finally, the pixel coordinates of the anchor hole center are calculated to find its actual 3D coordinates. The actual detection results show that: The average detection accuracy of the im-proved YOLOv8s model reaches 91%, which is 5% higher compared to the YOLOv8s model; the improved YOLOv8s model has a better detection ability, which can accurately detect the location of the anchor holes in the image under the interference of the influencing factors of different scale combinations; the number of frames per second (FPS) of the processed image of the improved YOLOv8s model is maintained at 171 frames/s, which can meet the real-time requirements of the model detection function.