Nearest neighbor for histogram-based feature extraction

Abstract

Manual grading process of Fresh Fruit Bunch (FFB) leads to misconduct and human error while inspecting the right category of fruits for the purpose of oil palm production at the mill. It is extremely important to identify the degree of ripeness of FFB are at 95% level of confidence as mentioned by Malaysian Palm Oil Board (MPOB). Therefore, wrong evaluation of graded fruits will result wrong report regarding the oil content. However, the most critical part of oil palm grading is the fruit classification. Error in classifying the right category of FFB will cause error in estimating the oil content. Research done by Federal Land Development Authority (FELDA) at mills show the estimated oil content for ripe fruit is 60%, while underripe is 40% and unripe is only 20% minus water and dirt. This indicates the importance of the right classification of FFB during grading process is essential to prevent from mistakenly claim low quality fruits as the good ones. Problem will occur while receiving the grading report claiming the high percentage of Basic Extraction Rate (BER) by the appointed graders while they have been proven to be poor quality fruits during oil production process. Fruit ripeness identification based on color is hard to measure especially when it involves the color intensity. The most suitable color space must be carefully selected to determine the right color especially when the color intensity is involved. HSV color space has proven to be a good choice because it has all the colors in the channel. Besides, it also offers color intensity which can be in variety level of intensity degrees. This paper explores the use of Nearest Neighbor Distance for histogram-based fruit ripeness identification. Promising results are obtained when value elements of HSV gives the highest recognition rate towards both ripe and unripe category.