Automated plant node detection using terrestrial LiDAR data under field conditions

Plant node is an important phenotypic trait which can be used to characterize plant architecture and evaluate grow conditions. This study presented a method for cotton plant node detection from terrestrial LiDAR data collected under field conditions. The data processing pipeline was proposed according to the morphological features of cotton plants and mainly involved three steps. First, a Laplacian-based contraction was used to extract the 3D curve skeleton; second, a minimum-spanning-tree-based method was developed to segment the main stalk and branches; third, plant nodes were detected through finding the junction point between the main stalk and each branch. The experiment showed that a highly accurate point cloud could be obtained using the LiDAR sensor. The proposed method achieved an accuracy of around 91.75%, and the F1-score was 0.95 between the sensor detected nodes and the ground truth data. Other traits such as internode length and height-to-node ratio could be derived based on this method. This information could be highly useful for monitoring crop growth and yield prediction.