基于卷积神经网络的玉米根茎精确识别与定位研究  

Accurate Identification and Location of Corn Rhizome Based on Faster R-CNN

作　　者：杨洋[1,2] 张亚兰 苗伟 张铁[3] 陈黎卿[1,2] 黄莉莉[1,2] YANG Yang;ZHANG Yalan;MIAO Wei;ZHANG Tie;CHEN Liqing;HUANG Lili(School of Engineering,Anhui Agricultural University,Hefei 230036,China;Anhui Intelligent Agricultural Machinery Equipment Engineering Laboratory,Hefei 230036,China;Chinese Academy of Agricultural Mechanization Sciences,Beijing 100083,China)

机构地区：[1]安徽农业大学工学院,合肥230036 [2]安徽省智能农机装备工程实验室,合肥230036 [3]中国农业机械化科学研究院,北京100083

出　　处：《农业机械学报》2018年第10期46-53,共8页Transactions of the Chinese Society for Agricultural Machinery

基　　金：国家重点研发计划项目(2017YFD0301303;2017YFD0700902-2);安徽省自然科学基金项目(1708085QF148)

摘　　要：为了能精准地识别和定位玉米根茎,本文建立了基于迁移学习方法的玉米根茎检测网络,模拟人眼识别功能从复杂的田间环境中识别和定位玉米根茎,实现履带自走式热雾机玉米行间对行行走。以履带自走式热雾机为图像采集平台获取玉米作物田间图像,采用DOG金字塔算法提取图像中的目标根茎,构成样本训练数据库。通过训练网络,首先实现了单株玉米根茎的精准识别,然后开展玉米作物行间环境下多株玉米根茎精确识别和根茎定位。基于已识别的玉米根茎位置采用最小二乘法拟合行驶路径,试验结果表明,提出的玉米根茎识别方法与传统图像处理的方法相比,具有更好的定位精度,能够实现玉米作物田间路径的准确规划,为履带自走式热雾机玉米行间对行行走提供了技术支撑。In order to identify and locate the maize rhizomes accurately,a maize rhizome detection network based on the migration learning method was established.The function of human eye recognition to identify and locate the rhizomes of the corn from a complex field environment was simulated,which achieved the function of crawler heat fog machine walking along the corn line.Field image of corn was collected by crawler self-propelled hot fogging machine,construction of a precise identification and location model of corn rhizome based on convolutional neural network,and the“DOG Pyramid”algorithm was used to extract maize rhizome as the target from the images,which constituted the training sample database.Through training network,the single maize rhizome was precisely identified firstly,and then were accurately identified and located in the environment of corn crop.The path tracking was obtained by east square fitting algorithm based on the identified maize rhizome location,and the sliding mode track tracking algorithm was used to control the double differential drive motor of the caterpillar chassis to realize the path tracking.The test result showed that the corn root recognition method can identify and locate the maize rhizomes more accurately,the correct rate of identification and location of corn rhizome reached 91.4%,but the traditional image processing method can only reach 67.3%.It can be seen that the method of identifying maize rhizomes proposed had better positioning accuracy,which can better plan the corn field path accurately.The research results provided the key technical support for the crawler self-propelled hot fogging machine self walking along the intercropping of corn.

关 键 词：热雾机 玉米根茎 迁移学习 识别与定位 路径规划 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]