Pollen Recognition Using Multi-Layer Feature Decomposition

We propose a method for recognizing pollen types from images. Unlike other methods that measure visual characteristics directly on the pollen image, our method decomposes the images into layers prior to performing feature extraction. The method measures texture and geometrical characteristics in each layer. We tested our method on 1,060 samples of 30 species of pollen. The same dataset is also used to compare the results of other pollen-classification techniques. The findings show the proposed method’s classification rate is higher than those produced by classical techniques, and the layering technique increases the classification rate over the direct use of the same features.