Flower Classification Based On Single Petal Image And Machine Learning Methods

This research presented a novel automatic flower classification system based on computer vision and machine learning techniques. First, we obtained in total 157 petal images of three alike categories using a digital camera. After pre-processing, we extracted color features and wavelet entropies from the petal images. Then, principle component analysis was utilized for feature reduction. Finally, four different classifiers, Support Vector Machine, Weighted k Nearest Neighbors, Kernel based Extreme Learning Machine, and Decision Tree, were trained to recognize the categories of the petals. 5-fold cross validation was employed to evaluate the out-of-sample performance of the classifiers. The experimental results showed that Weighted kNearest Neighbors performed the best among all four classifiers with an overall accuracy of 99.4%. The proposed approach is efficient in identifying flower categories in comparison with state-of-the-art methods.