A combination of feature ranking approaches for rice images classification

Abstract

In smart agriculture, computer vision is applied to identify rice seeds instead of being investigated by experts. In this paper, we considered three types of feature descriptors, such as Local Binary Patterns (LBP), Histogram of Oriented Gradients (HOG) and Gradient Domain Image Stitching (GIST) to characterize rice seed images. However, this method raises the problem of dimensional phenomena and it is necessary to select the relevant features to have a compact and better representation. A new combination of feature selection methods is proposed to represent all the relevant information from different single feature selection methods. The experimental results show that our approach outperforms the results from the state-of-the-art.