Remote sensing technique for oil palm age classification using Landsat-5 TM satellite

Shamala, Vadivelu (2014) Remote sensing technique for oil palm age classification using Landsat-5 TM satellite. Project Report. UTeM. (Submitted)

Text REMOTE SENSING TECHNIQUE FOR OIL PALM AGE CLASSIFICATION USING LANDSAT-5 TM SATELLITE.pdf Download (286kB)

Abstract

Age of oil palm is an important variable used in carbon and yield forecasting studies. Conventionally, the age classification of oil palm was made manually by mapping the plantation area. This technique is time consuming and difficult to classify a large area of hectare which causes difficulty for organisation like MPOB to analyse on the yield production. With remote sensing, the nature of acquiring the oil palm age through spectral response is more convenient. Limited studies were concerning the performance of current technique in classification of oil palm age. Mostly were using traditional parametric statistical approaches. Moreover, previous studies for vegetation age prediction were carried using different remote sensing approaches consisting of different resolution and measurements of data. This project demonstrates the procedure/algorithm to classify age of oil palm trees using LANDSAT-5 TM remote sensing data. The study were conducted in two phases; where phase I is the land cover classification whereas phase II is the oil palm age classification. Firstly, region of interest (ROI) was identified and drawn in order to supply training and testing pixels for the supervised classification. Maximum likelihood (ML) classifier was used for land cover classification with overall accuracy of 85.69%. Whereas, three classifiers were studied, such as: ML, Neural Network (NN) and Support Vector Machine (SVM) for oil palm age classification. Two sets of training set, smaller training set and larger training set were compared to obtain a good result. The accuracy of the classifications was assessed by using confusion matrix and decision boundary analysis. In conclusion, SVM could be used to classify oil palm age as it performs the best with highest overall accuracy of 91.89%. It is stable of limited amount and quality of training data. Further study can focus on hybrid techniques for age classification with accuracy assessment using ground truth image instead of ROI.

Actions (login required)

View Item