PLANT IDENTIFICATION BASED ON A LATE FUSION METHOD WITH PRIORITY WEIGHTS

Abstract

Automatic plant identification based on images is currently very interesting. The challenge with the plant identification problem is the great similarity among species, especially when the number of species is large. In this study, we developed a plant identification method based on the use of late fusion method for identification results on different plant organs. We assigned priority weights by organ/species to the confidence scores of each model. The organ/species with a better identification result was assigned a higher weight. GoogLeNet was used to identify plant based on each organ. Experiments were applied to combine two to six organs according to leaf, flower, fruit, stem, branch, entire. This method is based on combining the product rule using weights assigned to plant organs and species. The experimental results have shown the effectiveness of the proposed method, it outperforms some fusion late methods. The proposed method achieved the highest accuracy when combining 2 organs, 3 organs, 4 organs, 5 organs, and 6 organs with 96.0%, 98.2%, 98.8%, 99%, and 99.2% respectively.