Abstract:
Efficient, non-destructive cabbage harvesting is crucial for preserving its flavor and quality. Current cabbage harvesting mainly relies on mechanized automatic picking methods. However, a notable deficiency in most existing cabbage harvesting devices is the absence of a root posture recognition system to promptly adjust the root posture, consequently impacting the accuracy of root cutting during harvesting. To address this issue, this study introduces a cabbage root posture recognition method that combines deep learning with traditional image processing algorithms. Preliminary detection of the main root Region of Interest (ROI) areas of the cabbage is carried out through the YOLOv5s deep learning model. Subsequently, traditional image processing methods, the Graham algorithm, and the method of calculating the minimum circumscribed rectangle are employed to specifically detect the inclination angle of cabbage roots. This approach effectively addresses the difficulty in calculating the inclination angle of roots caused by occlusion from outer leaves. The results demonstrate that the precision and recall of this method are 98.7 % and 98.6 %, respectively, with an average absolute error of 0.80° and an average relative error of 1.34 % in posture. Using this method as a reference for mechanical harvesting can effectively mitigate cabbage damage rates.
摘要:
高效,无损收获白菜对于保持其风味和质量至关重要。目前白菜收获主要依靠机械化自动采摘方法。然而,大多数现有的卷心菜收获装置的一个显著缺陷是没有一个根部姿势识别系统来及时调整根部姿势,因此影响采收过程中切根的准确性。为了解决这个问题,本研究介绍了一种将深度学习与传统图像处理算法相结合的白菜根姿态识别方法。通过YOLOv5s深度学习模型对白菜的主根感兴趣区域(ROI)区域进行初步检测。随后,传统的图像处理方法,格雷厄姆算法,并采用计算最小外接矩形的方法来具体检测白菜根的倾角。这种方法有效地解决了计算由外叶遮挡引起的根倾角的困难。结果表明,该方法的准确率和召回率分别为98.7%和98.6%。分别,姿态的平均绝对误差为0.80°,平均相对误差为1.34%。将该方法作为机械收获的参考,可以有效减轻白菜的伤害率。