Rice cultivation under uncertain weather conditions often requires farmers to visit paddy fields frequently to monitor rice growth and identify insect or pest infestations. Integrating UAV imagery with machine learning technologies offers a promising monitoring and visual navigation solution. The accuracy of evaluating the distribution of rice seedlings plays a crucial role in precision agriculture. This study aims to assess the performance of machine learning models, including YOLOv8, YOLOv11, and Detectron2, for detecting rice seedlings from UAV imagery collected 20 days after mechanized direct seeding. Based on the detected rice seedlings, we estimated their distribution using centroid-based methods and identified regions devoid of seedlings. Using a grid-based approach, we estimated the number of rice seedlings to be transplanted in these unplanted regions. Experimental results on 22 RGB images demonstrated that YOLOv11x (extra large) and Detectron2 with a ResNet-101 backbone achieved strong overall performance with high accuracy and efficiency. Furthermore, we estimated the quantity of rice seedlings that will be transplanted in a single UAV image.