Aiming at the key challenges in potato seedling detection, this paper proposes a target detection model based on YOLO-PS (You Only Look Once with Pyramid Seedling). In this model, MobileNetV4-backbone was introduced into the detection backbone to enhance the feature extraction capability of seedlings in different states. At the same time, the DLKA attention mechanism was introduced into the detection head to enhance the model's ability to extract and focus on the local features of potato seedlings in one step. In order to optimize the precise positioning of the bounding box, the Focal Loss function was used as the loss function of the model, and finally Pyqt5 was used to design a convenient and reliable interactive interface for the potato seedling identification system. YOLO-PS model was experimentally verified to exhibit excellent performance in the potato seedling detection task. On the test set, the precision  of the model reached 94.75%, the recall  was 95.58%, and the mean average precision  was as high as 96.67%. It effectively proved the effectiveness and superiority of the model in potato seedling detection. This study not only provides a reliable technical means for automated monitoring of potato seedlings, but also provides new ideas and methods for seedling detection of similar crops.