Objective To improve the quality and yield of liquor by achieving efficient and accurate detection of alcohol content during the liquor gathering process and to develop an detection model of alcohol content in liquor gathering based on improved YOLOv5s.Methods This study replaces the original feature extraction module of the YOLOv5s model with the lightweight ShuffleNetV2 module to reduce the model's depth, making it more compact. The convolutional block attention module (CBAM) dual-channel attention mechanism is added to the feature extraction process to capture features from different dimensions. The SIOU loss function is used to replace the original model's loss function. A novel method for alcohol content detection based on the improved YOLOv5s model is proposed.Results The improved model achieves an accuracy of 91.9%, with a model size of 6.7 MB. The recall rate and mean average precision (mAP) are 89.3% and 96.3%, respectively, showing an increase of 10.3% and 12.3% over the original YOLOv5s model. Compared to current mainstream models like YOLOv3, YOLOv5m, and YOLOv8, the mAP has increased by 44.3%, 9.3%, and 13.1%, respectively.Conclusion The improved YOLOv5s model proposed in this paper provides high accuracy in detecting alcohol content during the liquor gathering.