Double‐threshold segmentation of panicle and clustering adaptive density estimation for mature rice plants based on 3d point cloud

Abstract

Crop density estimation ahead of the combine harvester provides a valuable reference for operators to keep the feeding amount stable in agriculture production, and, as a consequence, guaranteeing the working stability and improving the operation efficiency. For the current method de-pending on LiDAR, it is difficult to extract individual plants for mature rice plants with luxuriant branches and leaves, as well as bent and intersected panicles. Therefore, this paper proposes a clustering adaptive density estimation method based on the constructed LiDAR measurement system and double‐threshold segmentation. The Otsu algorithm is adopted to construct a double‐threshold according to elevation and inflection intensity in different parts of the rice plant, after reducing noise through the statistical outlier removal (SOR) algorithm. For adaptively parameter adjustment of supervoxel clustering and mean‐shift clustering during density estimation, the calculation relation-ship between influencing factors (including seed‐point size and kernel‐bandwidth size) and number of points are, respectively, deduced by analysis. The experiment result of density estimation proved the two clustering methods effective, with a Root Mean Square Error (RMSE) of 9.968 and 5.877, and a Mean Absolute Percent Error (MAPE) of 5.67% and 3.37%, and the average accuracy was more than 90% and 95%, respectively. This estimation method is of positive significance for crop density measurement and could lay the foundation for intelligent harvest.