Comparison of machine learning algorithms for classification of LiDAR points for characterization of canola canopy structure

Recent changes to plant architectural traits that influence the canopy have produced high yielding cultivars in rice, wheat and maize. In breeding programs, rapid assessments of the crop canopy and other structural traits are needed to facilitate the development of advanced cultivars in other crops such as Canola. LiDAR has the potential to provide insights into plant structural traits such as canopy height, aboveground biomass, and light penetration. These parameters all rely heavily on classifying LiDAR returns as ground or vegetation as they rely on the number of ground returns and the number of vegetation returns. The aim of this study is to propose a point classification method for canola using machine learning approach. The training and testing datasets were clusters sampled from field plots for flower, plant and ground. The supervised learning algorithms chosen are Decision Tree, Random Forest, Support Vector Machine, and Naive Bayes. K-means Clustering was also used as an unsupervised learning algorithm. The results show that Random Forest models (error rate = 0.006%) are the most accurate to use for canola point classification, followed by Support Vector Machine (0.028%) and Decision Tree (0.169%). Naive Bayes (2.079%) and K-means Clustering (48.806%) are not suitable for this purpose. This method provides the true ground and canopy in point clouds rather than determining ground points via a fixed height rely on the accuracy of the point clouds, subsequently gives more representative measurements of the crop canopy.