Yolo-Rds: An Efficient Algorithm for Monitoring the Quality of Seedling Transplantation

Abstract

During the mechanized transplanting process, the planting angle and standing posture of potting seedlings are important factors in evaluating the survival of the planting, planting uprightness is an important indicator for evaluating planting quality. With the gradual expansion of mechanization, the traditional method of manually checking the uprightness of mechanized planting has gradually become inapplicable. To address this problem, an online monitoring system for the uprightness of transplanting machines based on machine vision is constructed. The system mainly consists of an image acquisition device, a vehicle-mounted computer, and an interactive interface. Furthermore, in order to improve recognition errors caused by the unstructured environment in the field, a planting uprightness monitoring device is designed to integrate photoelectric sensors, a CCD camera, an LED light, and a light hood. An efficient algorithm YOLO-RDS for the detection of the planting posture of potting seedlings is proposed using a single-stage detection paradigm. Firstly, the conventional (HBB) object detection box was improved by introducing an angular prediction head to accommodate different seedling planting inclined posture detection. Secondly, the Rotation Weighted Boxes Fusion (RWBF) detection head was constructed to extract the stem inclination angle using adaptive rotation to obtain the rotation angle of the bounding box in the world space. The algorithm detection AP is 97% and FPS is 22. By encapsulating the algorithm and integrating it, a potting quality monitoring system was developed. A field experiment was conducted on the system, the experiment showed that the average relative error rate of missed planting in system monitoring is 1.13%, the average relative error rate and accuracy rate of uprightness recognition are 3.09% and 92.56% respectively, and the absolute angle error is 0°-3°. This system has good accuracy and stability, meets the requirements of potting seedling uprightness monitoring, and can provide technical support for the evaluation of transplantation machine operation quality.

Keywords: Transplanting, Potting seedlings, Rotation detection, Uprightness recognition, Monitoring system