Background Subtraction for Accurate Palm Oil Fruitlet Ripeness Detection

Abstract

Image background subtraction is an important and essential process in many computer vision applications as allows for a more effective processing of the foreground objects. Various methods have been proposed for performing background subtraction in the literature. In this study, we investigated various background subtraction to automatically identify the correct class of the foreground objects. There are only a few major producers of palm oil and Malaysia is the world’s second-largest producer and exporter of palm oil in terms of volume. In 2019, the gross domestic product (GDP) contribution from palm oil in Malaysia was estimated to be around 37.6 billion ringgit to Malaysia’s economy or at 2.7 percent of the country’s GDP. Among the many major industries, it is one of Malaysia’s primary industries, and a main agricultural export. There are various studies to automatically identify fruit ripeness, ranging from mangos to strawberries, etc. In addition, there have been some work in recent years to identify the maturity of the palm oil fruit bunches, and the use of Raman spectroscopy on individual fruitlets, etc. This study investigates the effect of background subtraction on the performance of a deep neural network to accurately identify the ripeness of palm oil fruitlets i.e. ripe, unripe and over ripe. This was compared with a feature based probabilistic approach.