摘要
针对现有大田农机高精度卫星导航成本高,且受气象环境影响较大的问题,该文提出了一种基于D-S-CNN的农田多传感器融合的自动导航方法,实现田间低成本、高精度定位导航。首先,将低成本卫星导航、惯性导航、视觉导航传感器的数据进行预处理,分别得到拖拉机的位置、姿态信息,然后通过D-S证据理论进行传感器失效检验,最后将三种传感器的位置、姿态信息以及D-S证据信任函数矩阵输入CNN神经网络进行信息融合实现田间高精度定位。实验表明,采用基于D-S-CNN的农田多信息融合导航方法,在离线测试中,无传感器失效情况下定位精度为4 cm,单个传感器失效情况下定位精度为10 cm,定位时间约17 ms;在实际测试中,无传感器失效情况下定位精度为6 cm,单个传感器失效情况下定位精度为13 cm,且该算法具有较好的鲁棒性和实用性。
The existing agricultural machinery precise navigation method using GPS/BD costhighlyand also strongly influenced by thecomplicated field microclimate.This paper proposed a farmland intelligent navigation method based on multi-sensor and visual fusion using D-S-CNN to realize low cost and high precision positioning navigation infield.Firstly,satellite navigation,inertial navigation and visual navigation information were got after preprocessing.Then the sensors failure test was carried out and the D-S evidence trust matrix was calculated by using D-S evidence theory.Finally,the high precision position navigation was achieved by using sensors data of position,posture and D-S evidence trust matrix as input.The experiments results showed that the position accuracies were 4 cm and 10 cm in the cases of none and one failure sensor respectively.And the positioning time was 17 ms.The experimental position accuracies in actualworkcondition were 6 cm and 13 cm in the case of none and one failure sensor with good robustness.
作者
俞娟
卢伟
曾梦洁
赵思佳
YU Juan;LU Wei;ZENG Mengjie;ZHAO Sijia(Sanjiang University,Nanjing 210012,China;College of Artificial Intelligence,Nanjing Agricultural University,Nanjing 210031,China)
出处
《中国测试》
CAS
北大核心
2021年第12期106-113,119,共9页
China Measurement & Test
关键词
智能拖拉机
导航
传感器信息融合
卷积神经网络
intelligent tractor
navigation
sensor information fusion
convolution neural network
