Toh, Chia Ming (2021) Study Of Oil Palm Growth Variation And Machine Learning Based Classification With Synthetic Aperture Radar Imagery. PhD thesis, UTAR.
Abstract
In recent years, the increase in demand for food resources led to the rapid expansion of oil palm plantations. To cope with the expansion, a better plantation monitoring system is required which can be done via satellite remote sensing where biophysical properties of palms used by many to determine the overall health of palms. Therefore, it is essential for plantation managers to monitor and taking measures to optimize these parameters of planted palms in the field. Microwave remote sensing can potentially be used to assist agronomists and mangers in monitoring oil palm plantation. This project attempts to explore the relationship between various biophysical parameters of oil palm with microwave backscatter and later uses Deep Learning (DL) to classify the oil palm plots based on these parameters. We first explored this area of study with the use of Radarsat 2’s C band microwave. From this early study, we used simple deep learning network for classification and concluded that C band is not suitable for oil palm remote sensing. We then repeat the study later on using L band satellite image from ALOS-PALSAR 2. Here, we found good correlation between frond cross section area and microwave backscattering coefficient. In addition to parameter study, we also attempted to study how L band microwave backscattering coefficient varies with different stages of Ganoderma infected oil palms. The findings were also reinforced with the use microwave radiative transfer (RT) model. By using collected biophysical parameters as input for simulation, we obtained simulation results which showed good agreement with satellite measured backscattering coefficient. By utilizing the model’s capability to break down backscattering contribution from specific components, we also found that frond is the highest contributor followed by trunk for oil palms shorter than 7 meters. For taller oil palms, the role was reversed as taller trunk contributes more than fronds which are now sparser in nature. We then classify oil palms of different criteria from intensity images of L band satellite. We used several methods such support vector machine (SVM), convolution neural network (CNN) and finally a combination of SVM, CNN and RT model which shows good results with training results of 90% accuracy. The combination method shows good synergy between physical scattering model with classification capability of machine learning methods.
Actions (login required)
![[img]](http://eprints.utar.edu.my/style/images/fileicons/application_pdf.png)
