Improved statistical recognition algorithms for oil palm ripeness identification

Abstract

Awareness for high quality crude oil is crucial in oil palm production. Proper grading process is important to ensure only the ripe fruits are taken into consideration for the maximum level of oil content. Currently, researchers focus mainly on providing an automatic grading system using various techniques such as producing digital numbers, oil palm colorimeter, photogrammetric grading, fuzzy or neuro-fuzzy technique and so on. Even though some of them have more than 85% accuracy, it is only valid in controlled environment. However, when they are applied in real situation with uncontrolled environment, the accuracy can drop to less than 50%. So far, there is limited study on suitable colour model conducted on oil palm ripeness identification. Most researchers use RGB colour model to determine an oil palm ripeness. This research looks into the suitability and performance of HSV colour model in classifying an oil palm ripeness. Distance Measurement and Linear Discriminant Analysis are chosen as methods to classify an oil palm ripeness in this study. Histogram is used as a feature vector for feature extraction method while colour as a feature to be analysed. Images of oil palm were captured by an expert in the form of JPEG images. Preprocessing is then performed to remove noise and background from the images. Subsequently, images are transformed into histogram and mean value are extracted. Selected Distance Measurement such as Euclidean Distance, Nearest Neighbour, Furthest Neighbour and Mean Distance are then used for feature matching process. An Oil Palm Ripeness Identification algorithm is proposed, wherein an elimination technique is also introduced in the process. In addition, a Multiple Features Technique is also proposed to find the best feature which brings a very good recognition rate for selected Distance Measurement. The results show that 98% accuracy have been obtained in comparison with other researchers’ work.