Scale invariant feature transform technique for weed classification in oil palm plantation

Abstract

This study presents a new and robust technique using Scale Invariant Feature Transform (SIFT) for weed classification in oil palm plantation. The proposed SIFT classification technique was developed to overcome problem in real application of image processing such as varies of lightning densities, resolution and target range which contributed to classification accuracy. In this study, SIFT classification algorithm is used to extract a set of feature vectors to represent the input image. The set of feature vectors then can be used to classify weed. In general, the weeds can be classified as either broad or narrow. Based on this classification, a decision will be made to control the strategy of weed infestation in oil palm plantations. The effectiveness of the robust SIFT technique has been tested offline where the input images were captured under varies conditions usch as different lighting effects, ambiguity resolution values, variable range of object and many sizes of weed which simulate the actual field conditions. The proposed SIFT resulted in over 95.7% accuracy of classification of weed in palm oil plantation. © 2008 Asian Network for Scientific Information.