Abstract: The blockage of the conveying system of sugarcane harvester is one of the main factors affecting the working efficiency of the harvester, and the excessive feeding amount is the main reason for the blockage of the harvester. Under the complex conditions of sugarcane conveying, the traditional contact detection method has problems such as reduced accuracy, so that a method based on visual recognition to detect the feeding amount was proposed. In order to improve the calculation accuracy of the feeding amount, the three‑dimensional data model was established by the improved SGBM algorithm，and then SIFT was used for feature extraction in the optimization process, which improved the accuracy and robustness of the three‑dimensional reconstruction of sugarcane under complex conditions such as occlusion. Finally, the feeding amount was calculated. The experimental results showed that when feeding 1 to 5 sugarcanes, the mean volume error of the improved algorithm was reduced by 3.26%, 5.68%, 6.32%, 8.45%, 7.81%, respectively, and by 6.29%, 4.24%, 6.15%, 12%, and 10.28%, respectively. The identification of sugarcane feeding amount on the conveying prototype was realized.

Key words:  harvesting machinery, sugarcane harvester, seeding amount detection, three?dimensional reconstruction, visual recognition

摘要： 甘蔗收获机输送系统堵塞是影响收获机工作效率的主要因素之一，而喂入量过大是导致收获机堵塞情况发生的主要原因。在甘蔗输送条件复杂情况下传统的接触式检测方法存在精度降低等问题，所以提出一种基于视觉识别检测喂入量的方法。为提高喂入量的计算精度，通过改进的SGBM算法建立三维数据模型，在优化过程中利用SIFT进行特征提取，提高算法在遮挡等复杂情况下对甘蔗的三维重建精度和鲁棒性，最后进行喂入量计算。试验结果表明，在喂入1~5根甘蔗时，改进后的算法平均体积误差分别减少3.26%、5.68%、6.32%、8.45%、7.81%，存在蔗叶遮挡时分别减少6.29%、4.24%、6.15%、12%、10.28%，该方法实现在输送样机上对甘蔗喂入量的准确识别。

关键词: 收获机械, 甘蔗收获机, 喂入量检测, 三维重建, 视觉识别