At present, the cutting of seed potatoes is mainly accomplished manually, which caused a series of problems, such as heavy labor intensity, low efficiency and high cost. Thus, the automated cutting of seed potatoes is urgently needed to be solved, especially with the rising cost and decreasing availability of labor. The first and foremost step for automated cutting is the recognition of potato buds. An improved faster region convolutional neural network (Faster R-CNN) scheme was proposed to achieve better recognition performance for potato buds. Data augmentation technique was leveraged to expand the potato dataset. Faster R-CNN model was trained based on the expanded dataset, and experimental results on the test set indicated that the recognition precision was 91.67%, recall rate was 84.09% and F1 was 87.72%. The average running time was 0.183 s. On this basis, an improved Faster R-CNN approach was proposed. Gaussian weight reduction function was adopted to optimize the nonmaximum suppression (NMS) algorithm in Faster R-CNN. For the detection boxes which had overlaps with M greater than or equal to the threshold Nt, the corresponding scores was decayed in the improved Faster R-CNN, rather than setting them to zero in Faster R-CNN. Besides, a strategy of online hard example mining (OHEM) with the optimized NMS algorithm was adopted in the improved Faster R-CNN. Experimental results on the test set demonstrated that the improved Faster R-CNN scheme achieved a precision of 96.32%, a recall rate of 90.85% and an F1 of 93.51%, which were increased by 4.65 percentage points, 6.76 percentage points and 579 percentage points, respectively, compared with Faster R-CNN. Moreover, the average running time of the improved scheme was 0.183s, which was the same to that of Faster R-CNN. Namely, the improved scheme could achieve better recognition performance without incurring any noticeable additional computational overhead, thus satisfying the requirements for realtime processing. Consequently, the improved Faster R-CNN approach was effective for potato bud recognition and could lay a solid foundation for future automated cutting of seed potatoes. 