In the automatic intelligent picking of famous tea sprouts, the images obtained by the robot vision system have the following problems: the highlighted surface of a tea sprout leads to identification omissions, and the color distinction rate between the sprout and old leaves is low, resulting in an incomplete tea sprout segmentation and high segmentation error rate for old tea leaves. A tea sprout recognition segmentation method based on an improved watershed algorithm is proposed in this study. First, images of naturally grown tea leaves in a tea garden are collected. After an experimental analysis and comparison, the collected tea samples are smoothed by Gaussian filtering to remove noise, split the channels, obtain R, G, and B components, and analyze their characteristics. Second, the optimal adaptation threshold T′ is determined using the minimum error method. For all pixels of the B component, the pixel values are set to be greater than the threshold of zero. Third, image operations are performed on the G and B′ components to obtain G-B′ components, and the minimum error method is used to obtain the best adaptation thresholds T1 and T2 and enhance them via piecewise linear transformation to improve the distinction between the young leaves and the background in the image. Lastly, binarization is performed, and the Canny operator is utilized for edge detection. The foreground and background areas are determined, the unknown area is calculated and marked, and the watershed function is used to complete the segmentation. A comparative experiment is conducted by comparing the 100 samples collected using the threshold segmentation algorithm, watershed segmentation algorithm, and the proposed segmentation algorithm. One group is randomly selected among tea samples numbered 1–10, and another nine groups of tea samples are selected with a number interval of 10 for a total of 10 groups. Their experimental data are analyzed. Results show that the improved algorithm has an average segmentation accuracy rate of 95.79%, and it improves the accuracy and integrity of the segmentation of tea leaves.