Automatic oil palm detection and identification from aerial images using multi-scale clustering and normalized cross correlation

Abstract

Oil palm cultivation is one of the most important occupation in South East Asia. Since oil palm plantation covers the wide range of areas, it is very difficult to count the numbers of oil palms manually. This thesis presents the new methods to detect and identify an individual of oil palms in plantation areas from aerial images regardless of their sizes using features including shape and texture. The proposed methods can handle the problem of oil palm identification from aerial image when oil palms are too close to each other such that they are identified as single oil palm. The processes of oil pam detection and identification consist of non-oil palm components and noise filtering, distinguishing oil palms from non-oil palm components (e.g., grass, weed, road, pond or swamp), identifying individual oil palms and counting the numbers of oil palms. In this thesis, oil palms can be detected and distinguished from non-oil palm components by using Butterworth low-pass filter on Fourier spectrum in frequency domain and template matching based on normalized cross correlation. Then the multi-scale clustering method is used to separate the oil palm stand into individual oil palms. The proposed method is used to estimate the population of oil palms on a set of 21 aerial images. All the aerial images were captured by mounting a digital camera on the model airplane that was controlled to fly over the plantation areas. Each image was taken from different view and location. The experimental results show that the proposed methods can detect and identify oil palms correctly with the recognition rate at 96.34%