Abstract: A banana sucker is a vegetative body that grows out from an underground tuber. The growth status of the sucker has a great influence on the mother plant. Traditional method for manually measuring plant morphological parameters is both time-consuming and strongly subjective. This paper proposed a method for extracting the morphological information of banana buds in the field based on a 3D point cloud. A 3D point cloud acquisition system based on an autonomous navigation robot platform was developed to obtain high throughput 3D point cloud of banana suckers in large fields. A measuring algorithm of plant height stem thickness and leaf area based on banana bud sucker point cloud was developed, and a calculating method of stem thickness based on cylindrical surface fitting was proposed to reduce the random error caused by the selection of measuring the position of stem diameters. MLS algorithm was used to smooth the banana stem point cloud and improve the measuring accuracy of stem diameters. Through comparing the effect of Kinect, PMD, ZED and PVP-16 depth sensors on the collection of the banana bud point cloud, the mean absolute percentage error (MAPE) of plant height parameters obtained by Kinect V2 point cloud is 2.96 percentage points smaller than that of PMD camera, 5.63 percentage points smaller than that of ZED camera and 6.92 percentage points smaller than that of PVP-16. The MAPE of the stem diameter parameter obtained by Kinect V2 is 0.21 percentage points smaller than that of the PMD camera and 6.84 percentage points smaller than that of the ZED camera, indicating that the point cloud result obtained by Kinect V2 is superior to other sensors. The mean absolute percentage errors of plant height and stem diameter obtained by Kinect V2 point cloud had the highest accuracy, which was 4.79% and 9.20%, and the root mean square error (RMSE) was 0.055 and 0.044 m, and the determination coefficient R2 was 0.96 and 0.87, respectively. For point clouds collected from different directions using Kinect V2, the point feature histograms algorithm and Iterative Closest Point algorithm (PFH+ICP) were used to register the point clouds. Based on the registered point clouds acquired from the two sides of a banana sucker, a greedy triangle was used to reconstruct the triangular mesh surface of the leaves, and the leaf area of the plants was obtained by calculating the area of all the triangular mesh elements. The mean absolute percentage error of leaf area between automatic registration and manual registration was 16.59%, the RMSE was 197.83 cm2, and the determination coefficient R2 was 0.92, indicating that the registration model could reflect the leaf area of the banana sucker. The results show that the method proposed in this paper based on the point cloud obtained by Kinect V2 is suitable for obtaining the morphological parameters of banana suckers, and can provide a fast and accurate method for measuring the morphological parameters of plant height, stem thickness and leaf area of a banana sucker for orchard management. If the robot can autonomously identify and measure the morphological parameters of banana bud suctioning plants, more accurate data can be obtained and the intelligence of orchard management can be further improved.