基于深度学习的机器人疏果前毛桃目标识别方法  

Recognition of Peach Target Before Fruits Thinning by Robot Based on Deep Learning

作　　者：谢圣桥 宋健[2] 汤修映[1] 白阳 Xie Shengqiao;Song Jian;Tang Xiuying;Bai Yang(College of Engineering,China Agricultural University,Beijing 100083,China;Weifang University,Weifang 261061,China)

机构地区：[1]中国农业大学工学院,北京100083 [2]潍坊学院智能农业装备实验室,山东潍坊261061

出　　处：《农机化研究》2023年第6期183-187,共5页Journal of Agricultural Mechanization Research

基　　金：山东省重点研发计划项目(2019GNC106144)。

摘　　要：为实现自然环境下疏果前期毛桃目标的识别,提出了基于Faster R-CNN的毛桃目标识别方法,设计的网络由ResNet-50特征提取网络、区域生成网络和感兴趣区域子网组成,并以ResNet-50为基础网络,用来提取目标特征。工作时,区域生成网络依据ResNet-50提取的特征生成感兴趣区域,感兴趣区域子网依据目标特征和感兴趣区域进行毛桃的识别与定位;对图像进行扩增后,随机选取1920幅作为训练集、240幅作为验证集。用测试集中的240幅图像对模型进行测试,结果表明:目标识别方法能有效识别出自然环境下的毛桃目标,准确度为92.97%,误识率为7.03%,召回率为84.62%,平均检测速度为0.189s/幅,可实现疏果前期毛桃目标的识别,模型具有较好的鲁棒性和泛化能力。To identify previous peach before fruit thinning,the peach target recognition method based on Faster R-CNN is proposed.The network consists of ResNet-50,region proposal network(RPN)and region of intersect(RoI)subnets.ResNet-50 is the base network and is used to extract the target features.RPN generates ROI based on the features extracted by ResNet-50.Then,the ROI subnet identifies and locates the peach based on the target features and ROI.After amplification of the images,1920 images were randomly selected as the training set and 240 images as the validation set.Test the model with 240 images from the test set.The results show that the method can effectively identify peach targets in the natural environment.The accuracy of recognition is 92.97%,the false recognition rate is 7.03%,the recall rate is 84.62%,and the average detection speed is 0.189s/frame.This method enables the identification of peach in the pre-thinning stage.The model has good robustness and generalization ability.

关 键 词：毛桃 目标识别与定位 机器人 深度学习 Faster R-CNN 

分 类 号：S126[农业科学—农业基础科学] TP242[自动化与计算机技术—检测技术与自动化装置]