Plant Recommender System Based on Multi-label Classification

In this paper, a plant recommender system using 2D digital images of leaves is proposed. This system made use of feature fusion technique and the multi-label classification method. Feature fusion technique is used to combine the color, shape, and texture features. Invariant moments, color moments, and Scale Invariant Feature Transform (SIFT) are used to extract the shape, color, and texture features, respectively. The multi-label classification method is capable of classifying samples in more than one class. In multi-label classification method, the nearest neighbor classifier with different metrics is used to match the unknown image with the training images and assigns five different class labels (i.e. recommendations) for each unknown image. The proposed approach was tested using Flavia dataset which consists of 1907 colored images of leaves. The experimental results proved that the accuracy of feature fusion method was much better than all other single features. Moreover, the experiments demonstrated their robustness to provide reliable recommendations.