Abstract: In order to realize integration and simplification of PET beverage bottles level detection system in beltline of beverage production and replace trigger instruction based on traditional sensors with method of integrated machine vision in liquid level detection program of bottled PET beverage to achieve rapid identification and positioning of PET bottled beverage liquid level in beltline, a real-time dentification and positioning method based on improved YOLOv7 was proposed for PET beverage bottle liquid level in beltline．Based on original YOLOv7, original SPPCSPC pooled pyramid structure was improved to a faster SPPFCSPC structure, and original loss function was improved by using SIoU loss function．Experiments showed that improved YOLOv7 model recognized PET beverage bottles and bottled liquid level whose part of samples contained color distortion and noise with a recognition accuracy of 98.9% and 96.3%, and recognition time of per image was 12.1 ms．The model could still achieve high-precision liquid level recognition and positioning of bottled beverages despite color distortion, multiple noise points, and image rotation in collected image samples．