Abstract:

For white asparagus selective harvesting is the best harvesting method determined by its growth characteristics. Focusing on the difficulties that the texture and the color of shoot tips are similar with ridge surface under machine vision, the recognition method of asparagus shoots and precise positioning were studied in this research. A changeable scale ROI detection method was proposed, with the fusion of color transformation, histogram averaging, morphology and texture filtering. After that, a harvesting path optimization method of multiple asparaguses was proposed, which solved the problem of harvesting efficiency reduction caused by unreasonable harvesting paths. Firstly, real-time acquisition of the image and individual RGB channel Gaussian filtering were implemented. Based on the HSV color transformation and histogram averaging processing, the asparagus shoot and soil feature clustering analysis were carried out. According to the sprout degrees of asparaguses, the changeable scale ROI detection method was studied. The morphology and the texture of the shoot, and soil were statistically analyzed. According to the texture feature parameters, the position of shoot was determined and its geometric center was calculated. Secondly, in order to improve harvesting efficiency, a path optimization algorithm based on multiple asparaguses was designed according to the locations of the asparaguses and the bins to obtain the optimal harvesting path. Finally, in order to verify the reliability of the proposed methods, asparagus shoot location and harvest verification tests were carried out on the established harvesting test platform. The results showed that the recognition rate of white asparagus in the visual system was more than 98.04%, the maximum positioning error of the center coordinate of the white asparagus shoot was 0.879 mm in X direction and 0.882 mm in Y direction, and the average reduction of end-effector motion distance could be 43.89% after path optimization under different circumstances, the success rate of end-effector localization was 100% and the harvest rate of white asparagus in the laboratory test was 88.13%. The research verified the feasibility of the visual positioning and harvesting path optimization of the white asparagus selective harvesting robot.

Key words: white asparagus, harvesting robot, selective harvesting, visual positioning, harvesting path optimization, asparagus shoot recognition