University of Twente Student Theses

Login

Estimation of dry weight and height of rice crops using machine learning algorithms and unmanned aerial vehicle (UAV) data

Ariadji, Farah Nafisa (2020) Estimation of dry weight and height of rice crops using machine learning algorithms and unmanned aerial vehicle (UAV) data.

Link to full-text:https://ezproxy2.utwente.nl/login?url=https://library.itc.utwente.nl/login/2020/msc/nrm/ariadji.pdf
(only accessible for UT students and staff)
Abstract:Ensuring food security remains one of society’s biggest challenges. Rice is a staple food and has an important role in the world’s food system. It is engrained in the tradition and culture of many countries. The Philippines is one of the major rice-growing countries and one third of their food consumption is based on rice and its derived products. The high consumption of rice is not balanced with their rice production. This is because there is limited suitable land for rice cultivation, therefore further production growth depends on increasing yield in existing areas. The prediction and estimation of rice yield is necessary to strengthen food security. One way to predict yield is by monitoring and estimating crop parameters, specifically biomass, as they have direct relationship with yield. In addition, crop height as one of the crop parameters is also considered as suitable indicator for plant dry weight estimation and crop growth. The aim of this study was to accurately estimate the dry weight and height of the rice crop for the early wet season 2016 (2016 EWS) and dry season 2017 (2017 DS) in the Long Term Continuous Cropping Experiment (LTCCE) field in the International Rice Research Institute (IRRI) Experimental Station, Philippines using Unmaned Arial Vehicle (UAV) data, field data and machine learning algorithms. The UAV dataset consisted of a timseries of multispectral images of green, red, red edge, near infrared bands and its derived products including point cloud and Digital Surface Model (DSM) data. Field data consisted of 72 samples or subplots within the 1 hectare LTCCE in both the 2016 EWS and 2017 DS. The UAV data was conducted from 23 May to 3 August 2016 and 13 January to 11 April 2017. The machine learning algorithms used in this study were Artificial Neural Network (ANN), Support Vector Machine (SVM), and Random Forest (RF). In this study, the machine learning algorithms had four different sets of input variables, consisting of the historical total plant dry weight, the average reflectance of four multispectral bands, and vegetation indices. The selection of input variables was based on the correlations between field measured data of plant dry weight and rice height, to average spectral reflectance and vegetation indices. Low correlations between DSM and point cloud height metrics with field measured crop height were observed which prevented further analysis with the machine learning methods. Reasons for this are discussed and ideas for more representative field measurements of rice crop height are suggested. However, for dry weight (our measure of biomass) we demonstrated that for the 2016 EWS dataset, the SVM method performed best in terms of its accuracy with R2 = 0.75 and RMSE of 639 kg/ha. As for 2017 DS dataset, the best model comes from RF method with R2 = 0.88 and RMSE = 671 kg/ha. We conclude that in this study, SVM and RF algorithm have produced better models compare to ANN algorithms in estimating dry weight in rice with high accuracy by using field measured data and UAV data.
Item Type:Essay (Master)
Faculty:ITC: Faculty of Geo-information Science and Earth Observation
Programme:Geoinformation Science and Earth Observation MSc (75014)
Link to this item:https://purl.utwente.nl/essays/85241
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page