Ridge Detection and Perceptual Grouping Based Automatic Counting of Rice Seedlings Using UAV Images

Automatic counting of rice seedling stand is desirable but challenging. In this paper, we propose a new method to detect and count drill-seeded, delay-flooded rice seedlings using unmanned aerial vehicle (UAV) images. First, a coarse-to-fine dual scale ridge detection method is developed to detect individual rice research plot at a coarse scale and its associated rice seedlings at a fine scale. Then, the skeleton of each detected rice seedling region is extracted and separated into small skeleton segments for rice seedling structure analysis. Finally, the skeleton segments are grouped based on seedling structure by a graph-cut based perceptual grouping. We evaluated the proposed method using more than 20,000 rice seedlings from 20 UAV images and the results demonstrated that our method can detect and count rice seedling with a high average accuracy of 89.37%.