Build Auto Annotate Oil Palm Image Datasets Using Template Matching Correlation and Balanced Iterative Reducing and Clustering Using Hierarchies Algorithms

Data is the key component of any machine learning project. Data annotation process in general is one of the most important and time-consuming tasks and it is used manually to label the dataset and it takes almost 70% of the time just to analyze the dataset. This dataset is critical to drive computational development, robustness, and confidence in results. Oil Palm image can be used to predict the production that can be obtained. In one image of oil palm images, there are one or more images to be able to determine which palm tree is or not. Data annotations are generally used manually by doing bounding box images. In this paper, it is proposed by combining the Template Matching Algorithm with fast normalized cross correlation and balanced iterative reducing and clustering using hierarchies (BIRCH) to be able to collect image datasets, so as to facilitate and speed up the process of collecting images for auto annotate image datasets on oil palm. Based on the results of experiments conducted using original images taken by drones from a height of 150 m from the ground, 205 images have been captured. This algorithm detected coordinate points with the time consuming 30 minutes 24 seconds. Based on 10 tested images, an average accuracy of 89.90% was detected as oil palm images for the dataset. This shows a significant important thing that must be done in the auto annotate process can be used to give more accurate and faster on image detection.