Motoki Ichisawa

Research on improving accuracy of paddy rice growth index estimation by UAV-LiDAR measurement

Kazuyoshi Takahashi

In order to maintain the quality of rice, paddy rice growth surveys are currently conducted manually. In the conventional measurement, a worker enters the field and measures the grass height, number of stems, and leaf color of several rice plants. However, the survey results are easily affected by individual differences in the surveyed plants and the measurement methods of the workers. In light of the above, Phan et al. proposed a method for estimating grass height and number of stem in paddy rice using the canopy thickness of rice point cloud (rD) measured by LiDAR. The previous research showed that the method can be applied at a vehicle-mounted LiDAR onboard a UAV System. 
In this paper, we investigated the highly accurate growth index estimation of paddy rice in consideration of the distribution of point clouds for each growing season. We searched for appropriate canopy thickness for each growing season from the optimized stem number estimation result using grid search. For another study, the height distribution of the point cloud is assumed to be a mixed distribution of ground and paddy rice. Using this assumption, we verify whether the point cloud can be divided by clustering with a mixture distribution. According to the results of the stem number estimation, determining the canopy thickness in the early stage of growth in detail get the high accurate estimation result. As a result of the clustering, the point cloud could be divided in the early stage of growth, but it was hardly divided in the late stage of growth. When we examined the canopy thickness of the cluster considered to be a point cloud of paddy rice, we obtained the similar result as the grid search. In the future, it is expected that the division accuracy will be improved by examining clustering in detail.