Automatic Labeling of Rice Seedlings in Unmanned Aerial Vehicles Images

Smart agriculture has been researched in these years. With the development of artificial intelligence (AI) and Unmanned Aerial Vehicles (UAV) technology, AI-based object detection of UAV images helps to develop smart agriculture. Therefore, we propose automatic rice seedling labeling from a UAV image system based on YOLOv4. Many studies have shown great performance in object recognition from images. However, detecting small targets such as rice seedlings in UAV images is more difficult than traditional object recognition. In addition, the small number of data is also a problem to improve performance. Therefore, applying YOLOv4 and using the dataset from the AIdea contest in 2021, the proposed model is trained with the original UAV image data for data augmentation to detect small objects. We also design the user interface to upload the target images and visualization of the result. According to the experiment result, the proposed method showed an F1-score of 0.84 and improved the performance of rice seedling detection.