摘要
茶叶嫩芽自动识别分类是实现采茶机器人精采名优茶的关键技术。由于茶叶嫩芽与背景中茶叶差别很小,且茶叶嫩芽形状多样,有一芽一叶和一芽二叶等多种形式,给自动识别带来很大难度。基于Faster-RCNN深度学习神经网络模型多维度进行茶叶嫩芽识别。首先对网络性能进行分析,选取较优的网络模型;在此基础上,研究一幅图像中包含嫩芽的不同数量、形态、拍摄角度、光照条件多维度对识别性能的影响。结果发现,光照条件和拍摄角度对嫩芽识别影响较大。所采用的Faster-RCNN深度学习模型对45°角度拍摄、晴天环境下单株集中一芽两叶的茶叶嫩芽识别效果最佳,同时阴天和90°拍摄时识别效果较差。研究为后续实现机器人现代化智能化的名优茶精采提供了技术支持。
The automatic identification and classification of tea-buds was the key technology for picking famous brand tea with robots.There was little difference between tea-buds and background tea-leaves,and the shapes of tea-buds have one-bud-one-leaf and onebud-two-leaves,which brought great difficulty to identify tea-buds.The paper utilizes Faster-RCNN deep learning neural network to identify tea-buds in multiple dimensions.The network performance was analyzed and a better network model was selected firstly.After that,the identification performance of the selected network was studied.The effects of numbers and shapes of tea-buds,different shooting angles and lighting conditions on identification performance were studied.The results show that lighting conditions and shooting angles have a greater impact on the identification performance.Under sunny environment and 45°shooting angle,the one-bud-twoleaves in a single plant have the best identification result.While the identification performance is poor in the environment of cloudy and 90°shooting angle.The paper provides a potential technical support for the development of tea picking robots.
作者
许宝阳
高延峰
Xu Baoyang;Gao Yanfeng(Shanghai University of Engineering Science,Shanghai Collaborative Innovation Center of Intelligent Manufacturing Robot Technology for Large Components,Shanghai 201620,China)
出处
《农业装备与车辆工程》
2023年第2期19-24,共6页
Agricultural Equipment & Vehicle Engineering
基金
上海市自然科学基金(21010501600)。
关键词
茶叶嫩芽
深度学习
神经网络
目标识别
目标分类
tea bud
deep learning
neural networks
target recognition
target classification
