Rice yield estimation based on K-means clustering with graph-cut segmentation using low-altitude UAV images

Abstract

Abstract Predicting the harvest yield enables farm practices to be modified throughout the growing season, with potential to increase the final yield. Unmanned aerial vehicle (UAV) based remote sensing is a promising way to estimate crop yields. In this study, rice yield was estimated by segmenting grain areas using low altitude RGB images collected using a rotary-wing type UAV. In particular, an image processing method that combines K-means clustering with a graph-cut (KCG) algorithm was proposed to segment the rice grain areas. The graph-cut algorithm was applied to extract the foreground and background of the images. The foreground RGB images were converted to the Lab colour space and then K-means clustering was used to label pixels based on colour information. The area of the rice grains in the images was calculated from the clustered images. Using this grain area information, the rice yield of the field could be estimated. Experiments show that the proposed method can segment the grain areas with a relative error of 6%–33%, and it improved the relative error of the previous method (by 1%–31%). The coefficient of determination between the results of the proposed method and the ground truth was found to be 0.98. Furthermore, the relative error of the yield estimation for four field sections was 21%–31%. The results indicate that the UAV image-based grain segmentation has the potential to estimate rice yield accurately and conveniently.