Citation

Muliady, Muliady and Lim, Tien Sze and Koo, Voon Chet (2021) Machine-Learning Based Algorithm for Estimating Rice Plant Nitrogen Status. In: 2nd FET PG Engineering Colloquium Proceedings 2021, 1-15 Dec. 2021, Online Conference. (Unpublished)

Text 15 Muliady_abstract.pdf Restricted to Repository staff only Download (66kB)

Abstract

The need for an affordable system to estimate the rice plant N status is one of the important issues in achieving high rice production. Recent studies have shown the promising result of hyperspectral and multispectral imaging technology for plant nutrition estimation, but it costs expensive. On the other hand, a low-cost multispectral camera is inaccurate and has false interpretation issues because it only applies one sensor. This research objective is to develop a machine-learning based algorithm to estimate the rice plant N status based on the multispectral images captured by a low-cost Mapir camera. This research is initiated by creating the dataset from field experiments consisting of vegetation index, light intensity, and SPAD value. The vegetation index is calculated from the multispectral images, and SPAD value is measured by a plant nutrition tester. The dataset will be used to train several Machine Learning algorithms and modify their parameter to obtain the best performance. The most effort taken part is to create the appropriate dataset. The field experiment showed the multispectral images from Mapir have made a significant bias value of the vegetation index. After conducting a series of experiments, the ideal condition to capture the multispectral images of canopy rice leaves is on a sunny day with minimum shadow area. The images were calibrated using 4 point regression method and then segmented using the Near IR channel to separate the leaves from the background. The next step is to clean the data from the outlier and select the data from their fertilizing history. The last step is to apply several machine learning algorithms, and the result showed that Support Vector Regression and Gaussian Process Regression have the best performance. The performance of the algorithm may be different for each rice variety. This research was conducted only for Ciherang rice from West Java-Indonesia in the vegetative phase. The value of this research is improving a low-cost multispectral camera to achieve better accuracy, using a machine learning algorithm.

Item Type:	Conference or Workshop Item (Paper)
Uncontrolled Keywords:	Machine learning, rice plant N status, multispectral images, 4 point calibration
Subjects:	Q Science > Q Science (General) > Q300-390 Cybernetics
Divisions:	Faculty of Engineering and Technology (FET)
Depositing User:	Ms Nurul Iqtiani Ahmad
Date Deposited:	28 Jan 2022 01:54
Last Modified:	28 Jan 2022 01:54
URII:	http://shdl.mmu.edu.my/id/eprint/9886

Downloads

Edit (login required)