COUNTING OIL PALM BUNCH WITH IMAGE ANALYSIS

• Main Author: ALRIANA HARYADI MOEL (NIM : 13513051), IGNATIUS
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/27925
• Institution: Institut Teknologi Bandung

<p align="justify">Now object counting field has already know by many people. But counting oil palm bunch still done by farmer with manual counting and errors often occur. Manual counting by farmer is easily manipulate. Using image analysis method that already use in another counting object problem, solution for counting oil palm bunch is obtained. <br /> <br /> <br /> Heuristic approach used because it is simple. The color feature used to identify oil palm bunch. In this final assignment there are three core stages that is image enhancement, image segmentation, and image counting. In image enhancement there are graylevel transformation, histogram equalization, and median filtering. In image segmentation there are image binarization, dilation process, and erosion proses. For object counting there are two types of method, the first one connected component labelling and the second one is k-means clustering. <br /> <br /> <br /> Testing data for this final assigment is 77 instances. This data consist of oil palm bunches that is still hanging in the tree. There are 4 scenario testing. Scenario 1 and scenario 2 will use graylevel as parameter for image binarization, scenario 3 will use hue value as parameter for image binarization, and scenario 4 will combine binary image that using graylevel and hue value. Scenario 1, 3, and 4 are using connected component labelling for counting object while scenario 2 is using k-means clustering for counting object. The result is scenario 1 has the lowest MSE value with weight of 8,052. MSE valuse means of errors weight based on another object that counted as an oil palm bunch. <br /> <br /> <br /> Base on the result after testing, the conclusion is that image enhancement and image segmentation stage is the most imporntant stage in this reasearch. The heuristic approach for determine the threshold and coeffisient for formula in every stages is the reason for raising the image quality that is increasing the result. Errors in the result caused by another object that count as oil palm bunch, an oil palm bunch that count more than once, and an oil palm bunch that not count.<p align="justify">