Research on Grass Recognition of Mowing System Based on Machine Vision

In order to realize the grass recognition in the mower system, plan the moving path of the mower and cut the grass automatically, the target detection algorithm of single shot multibox detector (SSD) and convolutional architecture for fast feature embedding (Caffe) were used to train the grass recognition model on the mower. Pictures of grass cutting field were taken by raspberry pie (RPi) and sent to the working machine. The coordinate values of the grass in the picture were calculated by the working machine and returned to raspberry pie, and the axle rotation angle, the running time and direction of the rear wheel motor according to the coordinate value of the grass were calculated automatically, and then the mechanical parts of the mower were mobilized to mow the grass. The experimental results show that compared with the traditional manual mechanical mower or fence mower, the trained grass recognition model can recognize the grass normally, and the mower can better plan the mowing path automatically, which has a certain weeding effect. The research results realize the combination of machine vision and traditional machinery, and provide some ideas for the future research of intelligent machinery.