The seedling line extraction of automatic weeding machinery in paddy field

Abstract

Mechanical weeding can solve the problems of environmental pollution caused by chemical weeding and high labor intensity of manual weeding. However, the traditional mechanical weeding method will result in the seedling injury due to the bending of the seedling rows. Introducing deep learning into traditional mechanical weeding methods can help the system accurately identify seedling rows and avoid crushing seedlings. In this paper, an identification method for straight and curved seedling rows based on sub-region growth and outlier removal is proposed, which can effectively solve the influence of factors such as the complex paddy field backgrounds, the weed distribution of different densities, and the curvature changes of the seedling rows, and has good robustness. The seedling line extraction mainly includes the image collection of rice seedlings during the two weed germination periods, the detection of rice seedlings based on the improved PFL-Yolov5 model, and the fitting of the center lines of straight-curved seedling rows based on sub-region growth and outlier removal. Experiments indicated that this method had good effects on the identification of seedling rows in different paddy field backgrounds, and with different weed densities, and straight and curved seedling rows, and the average identification angle error was less than 0.5°.