Vision systems in agriculture – Lagrangian particle tracking and field robotics

Zhang, Zhongzhong

Permalink

https://hdl.handle.net/2142/101671 Copy

Description

Title

Vision systems in agriculture – Lagrangian particle tracking and field robotics

Author(s)

Zhang, Zhongzhong

Issue Date

2018-07-10

Director of Research (if dissertation) or Advisor (if thesis)

Zhang, Yuanhui

Doctoral Committee Chair(s)

Zhang, Yuanhui

Committee Member(s)

• Wang, Xinlei
• Chamorro, Leonardo P.
• Chowdhary, Girish

Department of Study

Engineering Administration

Discipline

Agricultural & Biological Engr

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

Ph.D.

Degree Level

Dissertation

Keyword(s)

computer vision, particle tracking, field robotics

Abstract

This dissertation presents a study about two systems that utilize techniques in computer vision -- a Lagrangian particle tracking (LPT) system and a high-throughput phenotyping system. Both systems have important application in in agricultural engineering. Lagrangian particle tracking (LPT) enables investigation of turbulence in the Lagrangian reference frame and thus plays a key role in understanding problems such as diffusion, mixing, and transport. First of all, the LPT system is used to study the Lagrangian features of circular and semi-circular jets in the intermediate and far fields. Grid-interpolated velocity are used to validate the measurements, and confirmed the negligible effect of the pipe shape on the mean flow in the intermediate field. Several volumetric regions are defined to get Lagrangian statistical description of the flow from categorized particle trajectories. Probability density functions (PDF) of the velocity fluctuations, particle acceleration, and curvature of the trajectories reveal common and distinctive features of the jets. The first one shows departure from the Gaussian distribution away from the core, and the acceleration exhibits heavy tails in the two jets; however the curvature PDF reveals distinctive footprint of the pipe shape. Secondly, we improve the accuracy and robustness of the velocity and acceleration estimation of the LPT system that used to suffer from considerable variance due to the noise in particle position detection. A long short-term memory network is trained by synthetic trajectories to predict particles' state in a Burgers vortex. Compared to the baseline methods reported in literature, our model results in lower root mean errors for particle velocity and acceleration estimates. Additionally, the errors of our model remain consistent when we increase the measurement noise by a factor of two. Finally, we develop a high-throughput phenotyping system based on ground robots. By using the latest image recognition algorithm, the system is able to autonomously count corn stands by driving through the fields. The system replaces the traditionally time consuming and labor intensive process. Through an intensive season-long field study in real corn-fields, we demonstrate that the algorithm is robust against interferences from leaves and weeds. In particular, the system has been verified in corn fields at the growth stage of V4, V6, VT, R2, and R6 from five different locations. The robot predictions agree well with the ground truth with the correlation coefficient $R=0.96$. These results indicate a first and significant step towards autonomous robot-based real-time phenotyping using small ground robots for corn, sorghum, and other crops.

Graduation Semester

2018-08

Type of Resource

text

Permalink

http://hdl.handle.net/2142/101671

Copyright and License Information

Copyright 2018 Zhongzhong Zhang

Owning Collections