关键词: 果树, 机器视觉, 靶标探测, DSP, 横框识别算法, 竖框识别算法

Abstract: In order to realize online identification and detection of fruit trees in actual orchards, a fruit tree target identification device based on machine vision is designed. Firstly, an online fruit tree target recognition algorithm applied to DSP is proposed, and the background of the image is segmented by HSV chroma segmentation method. Then, the crown information after segmentation （the numerical change law of white pixel occupancy） is judged, and two recognition algorithms, horizontal frame and vertical frame, are designed. A mobile car equipped with identification device is designed and tested in the field. It is found in the pre-experiment that the change of illumination has little influence on the recognition effect. Further experiments were carried out under the illumination of 650Lux. The results showed that when the vehicle speed was 0.2 m/s, 0.4 m/s and 0.8 m/s, the recognition accuracy of the horizontal frame recognition algorithm was 88%, 84% and 34%, respectively, and the recognition accuracy was decreasing with the increase of the vehicle speed. From the time when the camera aligns with the center of the target plant to the time when the system determines that there was an effective target plant, the detection delay of 0.4-0.8 s was generated. Under the same conditions, the recognition accuracy of the mullion recognition algorithm was 88%, 86% and 84% respectively, and the recognition delay was short （0.2-0.3s）. The advantages of the vertical frame recognition algorithm are fast recognition speed and rich feedback position information, while the horizontal frame recognition algorithm can synchronously analyze the size of the crown contour of fruit trees. The position information obtained from the two algorithms is conducive to further optimizing and controlling the spraying time of a single plant.

Key words: fruit tree, machine vision, target detection, DSP, horizontal frame recognition algorithm, vertical frame recognition algorithm