The first problem to be solved in potato cutting fast is the detection of potato seed bud eyes, an improved YOLO v5s-based potato seed bud eye detection method was proposed to improve seed potato eye detection performance. Firstly, by adding the CBAM attention mechanism, the feature learning and feature extraction of the potato bud eye images were strengthened. The influence of the potato surface background similar to the bud eyes on the detection results was weakened. Secondly, the weighted bidirectional feature pyramid BiFPN was introduced to increase the original information of bud eyes extracted by the backbone network and assign weights to feature maps of different scales, making multi-scale feature fusion more reasonable. Finally, it was replaced with an improved and efficient Decoupled Head to distinguish between regression and classification, speed up the convergence speed of the model, and further improve the performance of potato bud eye detection. The test results showed that the precision, recall rate, and average precision of the improved algorithm were 93.3%, 93.4% and 95.2%, respectively, which was 3.2 percentage points higher than that of the original algorithm in the mean average precision, and the precision and recall rate were improved by 0.9 and 1.7 percentage points. The comparative analysis of different algorithms showed that this algorithm had absolute advantages compared with Faster R-CNN, YOLO v3, YOLO v6,YOLOX and YOLO v7 algorithms. The mAP was increased by 8.4 percentage points, 3.1 percentage points, 9.0 percentage points，12.9 percentage points and 4.4 percentage points. In the actual detection application, the average recall rate of the improved algorithm was 91.5%, which was 17.5 percentage points higher than that of the original algorithm, and the missed detection rate was reduced. The method can provide technical support for the next step in the development of a sprout-eye identification device for the intelligent cutting of potato seed potatoes.