Palm Oil Plantation Area Clusterization for Monitoring

This paper discusses the use of the clusterization to group palm trees in plantation areas using three categories, i.e. healthy, unhealthy, and non-plantation. Here, overhead images taken by an Unmanned Aerial Vehicle (UAV) were used to view a wider area. Images were divided into several smaller images using sliding windows and extracted using three color feature extraction techniques, i.e. 2D Wavelet Decomposition Color Energy, Principal Component Analysis, and t-Distributed Stochastic Neighbor Embedding (t-SNE). Texture feature extraction techniques used were Local Binary Pattern, Gray Level Co-occurrence Matrix and Segmentation-based Fractal Texture Analysis. Cluster results using the different techniques were compared to determine the optimal feature. Sliding windows were first implemented, and then cropped into small images with the same size as the windows. During clusterization, the K-Means clustering method was used to divide all smaller images into groups with high degrees of similarity. Feature extraction techniques were used individually to divide areas into three categories. The ground truth of the dataset was determined in advance, and results were compared to determine recognition rate. The study shows that dimensionality reduction using t-SNE in RGB color obtained the best clusterization results with 1135 correct patches.