FRESH FRUIT BUNCH RIPENESS ESTIMATION BASED ON MULTISPECTRAL IMAGE ANALYSIS ON RGB HSV XYZ LAB YCbCr COLOR SPACE

• Main Author: Endra Prasetya, Oktavian
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/39166
• Institution: Institut Teknologi Bandung

The high demand for palm oil makes palm oil production increased every year. The quality of fresh fruit bunch is a major factor in increasing production. The quality in question is the level of ripeness of the fresh fruit bunch itself. There are 3 levels of fresh fruit bunch ripeness including raw, ripe and overripe. To determine the ripeness of fresh fruit bunch can be done by observing the color, texture and number of fruit in a bunch of oil fresh fruit bunch. Several studies have been done to improve the grading quality of fresh fruit bunch. One research area that can be used to grade the ripeness level of fresh fruit bunch is image processing. By using the image of fresh fruit bunch, a lot of information can be obtained in more detail such as the color, texture and number of fruits found in bunch of fresh fruit bunch. Further research states that fresh fruit bunch image can be observed in several different wavelengths, images from 1 object produced from several wavelengths are referred as multispectral images. Multispectral image contain more information than normal image, it can can be used to show oil and water content in fresh fruit bunch. In this study the authors used multispectral images obtain from LED lighting and camera modules assistance on visible light and Near Infrared (NIR) wavelengths. By using the analysis of each component in the RGB, rgb normalization, HSV, CIE XYZ, CIE LAB, and YcbCr color space which are assisted by using the support vector machine (SVM) classification method, wavelengths and color space components can be used for grading the ripeness level of fresh fruit bunch. The results of the analysis show that the wavelength that is suitable for the grading of fresh fruit bunch ripeness is a wavelength of 940 nm and by using the selected components in each color space as the SVM classification feature, the accuracy of the system is 97,04%