This research aims to improve the visual target detection and recognition capabilities of shopping robots in various sales environments by optimizing and improving the YOLO algorithm, in order to improve accuracy and real-time performance. The research method involves embedded spatial hierarchical sampling technology and it adapts to image processing of different sizes, uses a separate convolutional neural network structure to reduce computational complexity, and cultivates a more concise network model by refining the effective data of complex models. Experimental results show that the improved YOLO algorithm performs well in weak Its average accuracy has been significantly improved under light, medium light and strong light environments, especially in the detection of small items. A study shows that improved programming significantly improved the vision of shopping assistance robots. Recognition capabilities enable robots to provide more accurate and faster services in real shopping environments.