A navigation method for paddy field management based on seedlings coordinate information

Compared to tillage, transplanting, and harvesting operations in paddy fields, field management operations such as spraying and weeding require completing tasks along the rice seedling rows to minimize seedling damage and reduce operational losses. Visual navigation, as a commonly used navigation method, can identify rice seedling rows, but it is susceptible to complicated paddy field environments. To address these challenges, our study combines UAV (unmanned aerial vehicle) surveying technology with the YOLOv5s object detection algorithm and proposes a navigation method for paddy field management based on seedling coordinate information. Firstly, the YOLOv5s object detection algorithm was used to identify and locate rice seedling points in the UAV image, and a coordinate transformation model was established to extract the geographic coordinates of the seedling points. Secondly, in order to classify the recognized seedling points into different seedling rows, a classification algorithm was proposed, and the improved RANSAC algorithm further promoted the classification effect. Once again, to obtain a smooth navigation path for agricultural machinery navigation, B-spline curves were used to approximate the classified seedling points. Finally, the generated navigation paths were written into the navigation controller, and the navigation performance was tested on flat ground and paddy fields using a vehicle platform. The results of the flat ground test indicated that different bending degrees have a significant impact on the lateral error. When the bending degree was large, the maximum lateral error was less than 7 cm, and when the bending degree was small, the maximum lateral error was less than 4 cm. The navigation accuracy of RMS in the flat ground test was 2.7 cm. In the paddy field experiment, due to the influence of loose soil surfaces and potholes, the maximum lateral error was about 16 cm. The navigation accuracy of RMS in the paddy field experiment was 5.9 cm, which can meet the operational requirements. The experimental results show that the proposed navigation method can achieve automatic navigation of agricultural machinery along curved rice seedling rows under complex paddy field environments, such as light change, seedling shortage, uneven road surfaces, and curved seeding rows.