基于YOLOv5深度学习的茶叶嫩芽估产方法

针对丘陵地区小规模茶园估产难度高，估产手段少等问题，采用基于YOLOv5的目标检测算法和田间抽样调查法，对丘陵地区小规模茶园估产问题进行研究。在茶园中随机抽取9个有代表性的茶叶生长点；使用目标检测算法识别抽样点茶叶嫩芽数目；利用最小二乘法拟合茶叶嫩芽产量与数目间的线性关系；结合抽样点识别出的嫩芽数目、抽样点面积、线性拟合关系和茶园整体面积估算出茶园茶叶嫩芽产量。结果表明:1)基于YOLOv5的目标检测算法对茶叶嫩芽识别的精度为99.02%，平均准确率为90.14%；2)茶叶嫩芽数目和产量间有高度线性关系，决定系数R²为0.999 8；3)通过算法估计的茶叶嫩芽产量与实际采收产量相对误差为29.56%。本研究能够较为方便的估算出茶园茶叶嫩芽产量，在茶叶生长时期为农户提供产量相关的数据支持，便于茶叶生产的前期管理。

Yield estimation method for tea buds based on YOLOv5 deep learning

Aiming at the problems of the high difficulty and only few methods of yield estimation in small-scale tea plantations, a YOLOv5-based target detection algorithm and field sampling method were combined to study the yield estimation problem of small-scale tea plantations in hilly areas. Firstly, nine representative tea growing points were randomly chosen in a tea plantation, and YOLOv5-based target detection algorithm was adopted to identify the number of tea buds at the sampling points. Secondly, the linear relationship between the yield and the number of tea buds was fitted through the least square method. Finally, the yield of tea buds in the tea plantation was estimated by combining the number of shoots identified at the sampling points, the area of the sampling points, the linear fitting relationship and the overall area of the tea plantation. The results indicate that: 1)The highest accuracy of the YOLOv5-based target detection algorithm for tea buds identification reaches 99. 02%, with an average accuracy of 90. 14%; 2)There is a highly linear relationship between the number of tea buds and yield, and the determination coefficient R² is 0. 999 8; 3)The relative error between the estimated tea buds yield and the actual harvest yield is 29. 56%. In general, the method constructed in this study can estimate the yield of tea buds in tea plantations. This study provides farmers with yield-related data support more comprehensively and conveniently during the tea growing period, which facilitates the preliminary management of tea production.