Computer aided identification of the Ficus L. species by the lamina shape

Abstract

Alexander Z. Gluhov, Ivan I. Strelnikov

The development of computer aided plant species determination is the urgent task of the botanical science. Identification is often bases on the morphology of the lamina. It is promising to describe the leaf shapes through the harmonic values of elliptic Fourier decomposition, but the effectiveness of this approach requires further verification. Another task is a comparative evaluation of different classification algorithms. The work was conducted on the 2812 leaves images of the 15 Ficus L. species. To solve the described tasks the optimal set of the Fourier decomposition parameters was determined. The best results are achievable by using the classification with 18 Fourier harmonics. Number of reference points on the outline does not affect the result of the models. We compared an identification accuracy of the 30 classification algorithms. Random forest algorithm had the highest classification accuracy – 98%. Combining different prediction algorithms by stacking improves the efficiency of the leaf shapes recognition.