Automatic and accurate segmenting a single poplar leaf is very necessary for non-contact extraction of plant leaf phenotype. However, a single leaf segmentation is a challenging task, especially for the complexity of field poplar seedling planting environment. An automatic leaf segmentation method combined SegNet with 3D point cloud clustering was proposed. In the proposed approach, to obtain accurate sample images, the Kinect V2 camera was firstly calibrated. Subsequently, the RGB and depth data were aligned, the background was filtered, and the RGB and deep fusion data of poplar seedling were collected. Then, for RGB and deep fusion data, a large number of samples were labelled and SegNet was utilized to segment poplar seedling leaf and trunk candidate regions. Finally, in order to better segment single poplar leaves, 3D point cloud of leaf regions were reconstructed by using the RGB-D fusion data of poplar leaf regions separated by SegNet, and kd-tree based on geometric distance was introduced to classify single leaves. The performance of the proposed method was verified by various comparative experiments for poplar seedlings in different growth environments. SegNet and FCN were used to segment the leaf region and stem region of poplar seedlings respectively. The results showed that the precision of SegNet for leaf and stem detection were 94.4% and 97.5% respectively, and the intersection over union (IoU) were 75.9% and 67.9% respectively, which was better than that of FCN. In order to find the suitable segmentation threshold for a single poplar leaf segmentation, the comparison experiments of different threshold segmentation using kd-tree for single and multiple poplar seedling leaf areas were performed. The experiment results validated that the proposed method can segment poplar leaves not only for a single poplar seedling, but also for multiple poplar seedlings.